Methods in

Molecular Biology 1256

Avraham Rasooly
Keith E. Herold Editors

Mobile Health
Technologies
Methods and Protocols
METHODS IN M O L E C U L A R B I O LO G Y

Series Editor
John M. Walker
School of Life and Medical Sciences
University of Hertfordshire
Hatfield, Hertfordshire, AL10 9AB, UK

For further volumes:

http://www.springer.com/series/7651
 Mobile Health Technologies

Methods and Protocols

Edited by

Avraham Rasooly
National Cancer Institute, Rockville, Maryland, USA

Keith E. Herold
Fischell Department of Bioengineering, University of Maryland, College Park, Maryland, USA
Editors
Avraham Rasooly Keith E. Herold
National Cancer Institute Fischell Department of Bioengineering
Rockville, MD, USA University of Maryland
College Park, MD, USA

ISSN 1064-3745 ISSN 1940-6029 (electronic)

ISBN 978-1-4939-2171-3 ISBN 978-1-4939-2172-0 (eBook)
DOI 10.1007/978-1-4939-2172-0
Springer New York Heidelberg Dordrecht London

Library of Congress Control Number: 2014957644

© Springer Science+Business Media New York 2015

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction
on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation,
computer software, or by similar or dissimilar methodology now known or hereafter developed. Exempted from this
legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied specifically for
the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer. Permissions
for use may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution
under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication, neither
the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be
made. The publisher makes no warranty, express or implied, with respect to the material contained herein.

Printed on acid-free paper

Humana Press is a brand of Springer

Springer is part of Springer Science+Business Media (www.springer.com)
Preface

Mobile health (mHealth) is a rapidly developing concept that is defined as “mobile

computing, medical sensor, and communication technologies for healthcare.” It is an
emerging field based on the development and proliferation of mobile devices that have vari-
ous sensors and significant computing power. mHealth offers promising approaches for
medical and public health service delivery in low-resource settings, which should help to
improve access to medical services for underserved populations globally. There are now
mHealth applications for medical diagnostics, biodetection, in vitro diagnostics, imaging,
and physiological measurements, including applications as diverse as accelerometers for
quantifying Parkinson’s disease tremor to devices for aiding neurosurgery.
Objectives of the Book: The primary aim of this book is to help present the emerging field
of mHealth by providing examples of work in this field. The book includes research manu-
scripts on several types of mHealth technologies along with their application in clinical
medicine and medical research. The authors were encouraged to discuss their technologies
in the context of the medical or research utility, and to address accuracy and speed of use.
Scope of the Book: The book describes mHealth technologies in three main mHealth
areas: in vitro and environmental testing, physiological and anatomical measurements, and
imaging. As a technology-oriented book, the chapters include technical information about
materials, methods, and protocols including discussion of pitfalls and lessons learned by the
developer in using the methods. The chapters also provide examples of the utility of each
technology and discussion of potential for clinical and research applications.
Target Audience: This book is designed to make mHealth more accessible and under-
standable to engineers, medical professionals, molecular biologists, chemical, and physical
science researchers developing mHealth technologies. We hope it will also be useful as a
teaching tool for bioengineers, biomedical engineers, medical professionals, and
biologists.
Book Organization: This book is divided into three parts: technologies for in vitro and
environmental testing, mHealth technologies for physiological and anatomical measure-
ments, and mHealth technologies for imaging. These are the most common areas targeted
by mHealth applications. In each part, the chapters are arranged based on the specific clini-
cal utility.
The section focused on mHealth technologies for sample analysis includes technologies
for microbial analysis (detection of HIV, Mycobacterium tuberculosis, and Dengue and malaria-
transmitting mosquitoes), cancer-related technologies, hematology, exposure to environmen-
tal compounds, and protocols on general detection methods relevant to mHealth such as
microfluidics and lab-on-a-chip technologies. This section includes a broad range of tech-
nologies such as cytometry, immunological assays, optical and electrochemical detection, gas
chromatography, and a variety of lab-on-a-chip and lab-on-paper assays.
The section on mHeath physiological and anatomical measurements includes chapters
on heart rate and sounds (stethoscopy), measuring body vital signs, monitoring essential

v
vi Preface

and pathological tremors and Parkinson’s disease, EEG and sleep disorders, and a mobile
device application for surgery support.
The section on mHealth imaging technologies includes chapters on cervical cancer and
skin cancer analysis and the imaging technologies of microendoscopy and skin lesion
imaging.
We would like to express our thanks and gratitude to the authors whose hard work and
excellent contributions are helping make mHealth technologies more accessible to clini-
cians, engineers, and researchers. We appreciate the authors’ time and especially their
patience during a long and arduous review process.
Our hope is that mHealth will play an instrumental role in improving access to medical
procedures including early detection, diagnostics, and treatment through the development
of new portable and accessible devices, and that this will lead to improved health
technologies.

Rockville, MD, USA Avraham Rasooly

College Park, MD, USA Keith E. Herold
Contents

Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

PART I MHEALTH TECHNOLOGIES FOR IN VITRO

AND ENVIRONMENTAL TESTING
1 Mobile Device for Disease Diagnosis and Data Tracking
in Resource-Limited Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Tiffany W. Guo, Tassaneewan Laksanasopin, Archana A. Sridhara,
Samiksha Nayak, and Samuel K. Sia
2 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal
NA Amplification, and Real-Time Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Michael G. Mauk, Changchun Liu, Mohamed Sadik,
and Haim H. Bau
3 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need . . . . . . . . . 41
B. Veigas, E. Fortunato, and P.V. Baptista
4 Immunofluorescence Microtip Sensor for Point-of-Care
Tuberculosis (TB) Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Jong-Hoon Kim, Kyong-Hoon Lee, Gerard A. Cangelosi,
and Jae-Hyun Chung
5 Improving Lateral-Flow Immunoassay (LFIA) Diagnostics
via Biomarker Enrichment for mHealth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
James J. Lai and Patrick S. Stayton
6 Microfluidic Toner-Based Analytical Devices: Disposable,
Lightweight, and Portable Platforms for Point-of-Care Diagnostics
with Colorimetric Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Karoliny Almeida Oliveira, Fabrício Ribeiro de Souza,
Cristina Rodrigues de Oliveira, Lucimeire Antonelli da Silveira,
and Wendell Karlos Tomazelli Coltro
7 Detection of Protein Biomarker Using a Blood Glucose Meter . . . . . . . . . . . . 99
Tian Lan, Yu Xiang, and Yi Lu
8 Microchip ELISA Coupled with Cell Phone to Detect Ovarian
Cancer HE4 Biomarker in Urine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
ShuQi Wang, Ragip Akbas, and Utkan Demirci
9 Point-of-Care Rare Cell Cancer Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . 123
David Issadore
10 Mobile Flow Cytometer for mHealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly

vii
viii Contents

11 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer

in the Developing World . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Bing Yu, Vivek Krishna Nagarajan, and Daron G. Ferris
12 Opto-Fluidics Based Microscopy and Flow Cytometry
on a Cell Phone for Blood Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Hongying Zhu and Aydogan Ozcan
13 Optofluidic Device for Label-Free Cell Classification from Whole Blood . . . . . 191
Tsung-Feng Wu and Yu-Hwa Lo
14 A Wearable Sensing System for Assessment of Exposures
to Environmental Volatile Organic Compounds . . . . . . . . . . . . . . . . . . . . . . . 201
Cheng Chen, Francis Tsow, Xiaojun Xian, Erica Forzani,
Nongjian Tao, and Raymond Tsui
15 Quantitative Point-of-Care (POC) Assays Using Measurements
of Time as the Readout: A New Type of Readout for mHealth . . . . . . . . . . . . 213
Gregory G. Lewis and Scott T. Phillips
16 Smartphone-Based Fluorescence Detector for mHealth . . . . . . . . . . . . . . . . . . 231
Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly
17 Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves
for mHealth Medical Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly
18 Spectrometry with Consumer-Quality CMOS Cameras . . . . . . . . . . . . . . . . . . 259
Alexander Scheeline
19 Mobile Phone Based Electrochemiluminescence Detection
in Paper-Based Microfluidic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Jacqui L. Delaney and Conor F. Hogan

PART II MHEATH TECHNOLOGIES FOR PHYSIOLOGICAL

AND ANATOMICAL MEASUREMENTS

20 iStethoscope: A Demonstration of the Use of Mobile Devices

for Auscultation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Peter J. Bentley
21 iPhysioMeter: A Smartphone Photoplethysmograph
for Measuring Various Physiological Indices . . . . . . . . . . . . . . . . . . . . . . . . . . 305
Kenta Matsumura, Peter Rolfe, and Takehiro Yamakoshi
22 Smartphone Attachment for Stethoscope Recording . . . . . . . . . . . . . . . . . . . . 327
Jeff Thompson
23 Use of Smartphones and Portable Media Devices for Quantifying
Human Movement Characteristics of Gait, Tendon Reflex Response,
and Parkinson’s Disease Hand Tremor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Robert LeMoyne and Timothy Mastroianni
24 Measuring Tremor with a Smartphone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Benoit Carignan, Jean-François Daneault, and Christian Duval
25 The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations . . . 375
Dmitri V. Poltavski
 Contents ix

26 Smartphone Based Monitoring System for Long-Term Sleep Assessment. . . . . 391

Alexandre Domingues
27 Intracranial Ventricular Catheter Placement with
a Smartphone Assisted Instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Ulrich-W. Thomale

PART III MHEALTH CANCER IMAGING TECHNOLOGIES

28 High-Resolution Microendoscope for the Detection of Cervical Neoplasia . . . 421
Benjamin D. Grant, Richard A. Schwarz, Timothy Quang,
Kathleen M. Schmeler, and Rebecca Richards-Kortum
29 Skin Lesions Image Analysis Utilizing Smartphones and Cloud Platforms . . . . 435
Charalampos Doukas, Paris Stagkopoulos, and Ilias Maglogiannis
30 Melanoma and Other Skin Lesion Detection
Using Smart Handheld Devices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
George Zouridakis, Tarun Wadhawan, Ning Situ, Rui Hu,
Xiaojing Yuan, Keith Lancaster, and Courtney M. Queen

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
Contributors

RAGIP AKBAS • Civil Engineering Department, Özyeğin University, Istanbul, Turkey

JOSHUA BALSAM • Division of Biology, Office of Science and Engineering, FDA,
Silver Spring, MD, USA; University of Maryland College Park (UMCP), College Park,
MD, USA
P.V. BAPTISTA • CIGMH, Departamento de Ciências da Vida, Faculdade de Ciências e
Tecnologia, Universidade Nova de Lisboa, Caparica, Portugal
HAIM H. BAU • University of Pennsylvania, Philadelphia, PA, USA
PETER J. BENTLEY • Department of Computer Science, University College London,
London, UK
HUGH ALAN BRUCK • Department of Mechanical Engineering, University of Maryland,
College Park, MD, USA
GERARD A. CANGELOSI • Department of Environmental and Occupational Health
Sciences, University of Washington, Seattle, WA, USA
BENOIT CARIGNAN • Département des Sciences Biologiques, Université du Québec à
Montréal, Montréal, QC, Canada; Centre de Recherche de l’Institut Universitaire de
Gériatrie de Montréal, Montréal, QC, Canada
CHENG CHEN • Biodesign Institute, Arizona State University, Tempe, AZ, USA
JAE-HYUN CHUNG • Department of Mechanical Engineering, University of Washington,
Seattle, WA, USA
WENDELL KARLOS TOMAZELLI COLTRO • Instituto de Química, Universidade Federal de
Goiás, Goiânia, GO, Brazil; Instituto Nacional de Ciência e Tecnologia de Bioanalítica
(INCTBio), Campinas, SP, Brazil
JEAN-FRANÇOIS DANEAULT • Centre de Recherche de l’Institut Universitaire de Gériatrie
de Montréal, Montréal, QC, Canada; Department of Neurology and Neurosurgery,
Montreal Neurological Institute & HospitalMcGill University, Montreal, QC, Canada
JACQUI L. DELANEY • Department of Chemistry, La Trobe Institute for Molecular Science,
La Trobe University, Melbourne, VIC, Australia
UTKAN DEMIRCI • Demirci Bio-Acoustic-MEMS in Medicine (BAMM) Laboratory,
Canary Center at Stanford for Early Cancer Detection, Stanford University School
of Medicine, Palo Alto, CA, USA
ALEXANDRE DOMINGUES • Instituto Superior Técnico, Technical University of Lisbon,
Lisbon, Portugal
JOÃO DOMINGUES-SANCHES • Instituto Superior Técnico, Technical University of Lisbon,
Lisbon, Portugal
CHARALAMPOS DOUKAS • Department of Digital Systems, University of Piraeus,
Piraeus, Greece
CHRISTIAN DUVAL • Centre de Recherche de l’Institut Universitaire de Gériatrie de
Montréal, Montréal, QC, Canada; Département de KinanthropologieUniversité du
Québec à Montréal, Montréal, QC, Canada

xi
xii Contributors

DARON G. FERRIS • Department of Obstetrics and Gynecology, Georgia Regents University,

Augusta, GA, USA
E. FORTUNATO • CENIMAT-I3N, Departamento de Ciências dos Materiais, Faculdade de
Ciências e Tecnologia, Universidade Nova de Lisboa, Caparica, Portugal
ERICA FORZANI • Biodesign Institute, Arizona State University, Tempe, AZ, USA
BENJAMIN D. GRANT • Department of Bioengineering, Rice University, Houston, TX, USA
TIFFANY W. GUO • Department of Biomedical Engineering, Columbia University, New
York, NY, USA
CONOR F. HOGAN • Department of Chemistry, La Trobe Institute for Molecular Science,
La Trobe University, Melbourne, VIC, Australia
RUI HU • Department of Electrical and Computer Engineering, University of Houston,
Houston, TX, USA
DAVID ISSADORE • University of Pennsylvania, Philadelphia, PA, USA
JONG-HOON KIM • Department of Mechanical Engineering, University of Washington,
Seattle, WA, USA
JAMES J. LAI • Department of Bioengineering, University of Washington, Seattle, WA, USA
TASSANEEWAN LAKSANASOPIN • Department of Biomedical Engineering, Columbia
University, New York, NY, USA
TIAN LAN • GlucoSentient, Inc., Champaign, IL, USA
KEITH LANCASTER • Department of Electrical and Computer Engineering,
University of Houston, Houston, TX, USA
KYONG-HOON LEE • NanoFacture, Inc., Bellevue, WA, USA
ROBERT LEMOYNE • Department of Biological Sciences, Northern Arizona University,
Flagstaff, AZ, USA
GREGORY G. LEWIS • Department of Chemistry, The Pennsylvania State University,
University Park, PA, USA
CHANGCHUN LIU • University of Pennsylvania, Philadelphia, PA, USA
YU-HWA LO • Department of Electrical and Computer Engineering, University of
California, San Diego, La Jolla, CA, USA
YI LU • Department of Chemistry, University of Illinois at Urbana-Champaign, Urbana,
IL, USA
ILIAS MAGLOGIANNIS • Department of Digital Systems, University of Piraeus, Piraeus,
Greece
TIMOTHY MASTROIANNI • Independent, Pittsburgh, PA, USA
KENTA MATSUMURA • Division of Bioengineering and Bioinformatics, Graduate School of
Information Sciennce and Technologg, Hokkaido University, Sapporo, Hokkiado, Japan;
School of Mechanical Engineering, College of Science and Engineering, Kanazawa
University, Kanazawa, Ishikawa, Japan
MICHAEL G. MAUK • University of Pennsylvania, Philadelphia, PA, USA
VIVEK KRISHNA NAGARAJAN • Department of Biomedical Engineering,
University of Akron, Akron, OH, USA
SAMIKSHA NAYAK • Department of Biomedical Engineering, Columbia University,
New York, NY, USA
CRISTINA RODRIGUES DE OLIVEIRA • Instituto de Patologia Tropical e Saúde Pública,
Universidade Federal de Goiás, Goiânia, GO, Brazil
 Contributors xiii

KAROLINY ALMEIDA OLIVEIRA • Instituto de Química, Universidade Federal de Goiás,

Goiânia, GO, Brazil
AYDOGAN OZCAN • Electrical Engineering Department, University of California,
Los Angeles, CA, USA; Bioengineering DepartmentUniversity of California, Los Angeles,
CA, USA; California NanoSystems Institute (CNSI)University of California,
Los Angeles, CA, USA
SCOTT T. PHILLIPS • Department of Chemistry, The Pennsylvania State University,
University Park, PA, USA
DMITRI V. POLTAVSKI • University of North Dakota, Grand Forks, ND, USA
TIMOTHY QUANG • Department of Bioengineering, Rice University, Houston, TX, USA
COURTNEY M. QUEEN • Department of Tropical Medicine and Global Health, Duke
University, Durham, NC, USA
AVRAHAM RASOOLY • National Cancer Institute, Rockville, MD, USA
REBECCA RICHARDS-KORTUM • Department of Bioengineering, Rice University, Houston,
TX, USA
PETER ROLFE • Department of Automatic Measurement and Control, Harbin Institute
of Technology, Harbin, Heilongjiang, China; Oxford BioHorizons Ltd., Maidstone, UK
MOHAMED SADIK • University of Pennsylvania, Philadelphia, PA, USA
ALEXANDER SCHEELINE • SpectroClick Inc., Champaign, IL, USA
KATHLEEN M. SCHMELER • Department of Gynecologic Oncology and Reproductive
Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA
RICHARD A. SCHWARZ • Department of Bioengineering, Rice University, Houston, TX,
USA
SAMUEL K. SIA • Department of Biomedical Engineering, Columbia University, New York,
NY, USA
LUCIMEIRE ANTONELLI DA SILVEIRA • Instituto de Patologia Tropical e Saúde Pública,
Universidade Federal de Goiás, Goiânia, GO, Brazil
NING SITU • Department of Computer Science, University of Houston, Houston, TX, USA
FABRÍCIO RIBEIRO DE SOUZA • Instituto de Química, Universidade Federal de Goiás,
Goiânia, GO, Brazil
ARCHANA A. SRIDHARA • Department of Biomedical Engineering, Columbia University,
New York, NY, USA
PARIS STAGKOPOULOS • Department of Digital Systems, University of Piraeus, Piraeus,
Greece
PATRICK S. STAYTON • Department of Bioengineering, University of Washington, Seattle,
WA, USA
NONGJIAN TAO • Biodesign Institute, Arizona State University, Tempe, AZ, USA
ULRICH-W. THOMALE • Division of Pediatric Neurosurgery, Charité Universitätsmedizin,
Berlin, Germany
JEFF THOMPSON • Visual Art & Technology, Stevens Institute of Technology,
Hoboken, NJ, USA
FRANCIS TSOW • Biodesign Institute, Arizona State University, Tempe, AZ, USA
RAYMOND TSUI • Raydis LLC, Tempe, AZ, USA
B. VEIGAS • CIGMH, Departamento de Ciências da Vida, Faculdade de Ciências e
Tecnologia, Universidade Nova de Lisboa, Caparica, Portugal; CENIMAT-I3N,
Departamento de Ciências dos Materiais, Faculdade de Ciências e
TecnologiaUniversidade Nova de Lisboa, Caparica, Portugal
xiv Contributors

TARUN WADHAWAN • Department of Computer Science, University of Houston, Houston,

TX, USA
SHUQI WANG • Demirci Bio-Acoustic-MEMS in Medicine (BAMM) Laboratory, Canary
Center at Stanford for Early Cancer Detection, Stanford University School of Medicine,
Palo Alto, CA, USA
TSUNG-FENG WU • Materials Science and Engineering Program, University of California,
San Diego, La Jolla, CA, USA
XIAOJUN XIAN • Biodesign Institute, Arizona State University, Tempe, AZ, USA
YU XIANG • Department of Chemistry, Tsinghua University, Beijing, China
TAKEHIRO YAMAKOSHI • Department of Information and Systems Engineering,
Faculty of Information Engineering, Fukuoka Institute of Technology, Fukuoka, Japan;
School of Mechanical Engineering, College of Science and Engineering, Kanazawa
University, Kanazawa, Ishikawa, Japan
BING YU • Department of Biomedical Engineering, University of Akron, Akron, OH, USA
XIAOJING YUAN • Department of Engineering Technology, University of Houston, Houston,
TX, USA
HONGYING ZHU • Electrical Engineering Department, University of California, Los
Angeles, CA, USA
GEORGE ZOURIDAKIS • Department of Engineering Technology, University of Houston,
Houston, TX, USA; Department of Computer ScienceUniversity of Houston, Houston,
TX, USA; Department of Electrical and Computer EngineeringUniversity of Houston,
Houston, TX, USA
 Part I

mHealth Technologies for In Vitro and Environmental Testing

 Chapter 1

Mobile Device for Disease Diagnosis and Data Tracking

in Resource-Limited Settings
Tiffany W. Guo, Tassaneewan Laksanasopin, Archana A. Sridhara,
Samiksha Nayak, and Samuel K. Sia

Abstract
Here we describe a low-cost mobile device that combines cell-phone and satellite communication
­technologies with fluid miniaturization techniques for performing all essential functions of enzyme-linked
immunosorbent assay (ELISA). Disease-specific antigens are immobilized on the microfluidic surface, and
disease specific antibodies are captured on the surface and visualized with silver–gold amplification. The
diagnostic result is automatically determined by the device by measuring the absorbance through the
silver–gold amplification in the microchannel. The results are displayed for the user and are synchronized
to a remote patient record. The overall system aims to be portable, robust, low-power, and fully utilize the
ability of mobile devices for bringing better health care to resource poor areas.

Key words HIV, Immunoassay, Microfluidics, Diagnostics, Mobile diagnostics, GSM/satellite

­communication, Global health

1  Introduction

In resource-limited settings, the rapid adoption of mobile phones

has enabled remote communication of voice and limited data at a
low cost. Such mobile technologies are beginning to be used to
improve public health and patient care [1, 2]; examples include
text messaging for improving adherence to malaria [3] and HIV
treatment [4, 5], transmission of images for telemicroscopy [6],
and personal digital assistants for collecting laboratory results [7].
However, there are major areas of health care services, including
preventive diagnostics, that remain unavailable to patients in
remote settings. Ideally, these services would also be linked with
mobile technologies that could access health records and hence
provide a full context of patient history. Reflecting the view that
access to patient records has the potential to markedly improve
patient care, over two dozen countries in the developing world are
implementing electronic databases that collectively contain hundreds

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_1, © Springer Science+Business Media New York 2015

3
4 Tiffany W. Guo et al.

of thousands of patient records [8, 9]. Rwanda alone has over

90,000 HIV/AIDS patients for whom care is currently facilitated
by electronic records [10].
A handheld device that could perform laboratory-quality diag-
nostic tests, and access patient health records [11, 12], would be
valuable for many reasons [13], including improved monitoring of
disease outbreaks [14], rapid transmission of field results to health
experts, increased effectiveness in allocating medications to differ-
ent communities [15], immediate treatment or quarantine of
patients in difficult-to-reach settings, and reduction in human-­
caused error in data transcription for health records.
While satellite-based communication has been virtually unex-
plored for POC health care, it offers several potential advantages:
(1) Broad coverage. Satellite phones are the preferred choice for
communication in extremely remote areas because they offer uni-
versal geographical coverage and do not rely on local infrastructure
such as cell-phone towers. (2) Scalability. Because satellite com-
munication operates independently from local infrastructure that
varies from country to country and over time, it can be rapidly and
reliably scaled up to all regions of the world. (3) Automated loca-
tion tracking. GPS coordinates are automatically communicated
along with each test result. (4) Cost and portability. While some
portable transceivers are still relatively expensive, difficult to install,
and consume large amount of power, transceivers for the Iridium
satellite constellation are less expensive, portable, and easy to install
and interface. The messaging cost by satellites can be brought
down to rival cell phone-based SMS by using the short-burst data
(SBD) service, which is commonly used today in maritime naviga-
tion. While the general perception of satellite-based communica-
tions is costly, a subscription fee of $15 per month and a usage fee
of $1.50 per kilobyte (kb) of transmitted data are charged for each
registered satellite modem. This can represent a minimal cost com-
pared to approximately $2 of production cost that will be spent per
test. Although satellite transceivers tend to be more expensive than
cell phones, the cost of satellite transceivers has been decreasing
over the years. Today, a small sized satellite transceiver (Iridium
SBD 9602) can be purchased commercially for less than $400.
Here, we developed a low-cost mobile device with fluid minia-
turization techniques called mobile microfluidic chip for immuno-
assay on protein markers (mChip) for performing all essential
functions of ELISA and integrated with cell-phone and satellite
communication technologies that send information wirelessly to a
central health record.
Our mobile device (Fig. 1a, c) consists of three main mod-
ules for liquid handling, signal detection, and data communica-
tion. All modules are controlled by an 8-bit microcontroller
along with peripheral components packaged onto a printed
c­ircuit board (PCB).
Fig. 1 Overview of mChip device. (a) The image of mChip device shown here includes a cassette inside.
(b) Transmitted light micrograph of channel meanders. Scale bar is 1 mm. (c) (Left ) Inside view of the mobile
device, with three main modules for liquid handling (orange ), signal detection (red ), and data communication
(green ). All modules are controlled by an 8-bit microcontroller along with peripheral components packaged onto
a printed circuit board. (Right ) Illustration of microfluidic cassette, with areas for reagent storage, sample meter-
ing, analyte capture and detection, and waste storage. At time of assay, a fingerprick of blood is collected in a
capillary tube (connecting regions of reagent storage with biochemical analysis) and drawn through the cassette
by vacuum generated by a micropump. Sample and reagents are eventually collected and stored as waste in a
membrane filter. (d) Schematic diagram of reagent storage sequence in PE tubing (reagents separated by air
gaps). Reagent delivery is passive with no requirement for moving parts on the chip. A preloaded sequence of
reagents passes over four detection zones, each zone characterized by dense meanders coated with capture
proteins, before exiting the chip to a disposable syringe used to generate vacuum for fluid actuation. (e) Illustration
of biochemical reactions in detection zones at different immunoassay steps. The reduction of silver ions on gold
nanoparticle conjugated antibodies yields signals that can be read with low cost optics (for quantification) or
examined by eye. Reprinted from Nature Medicine [16] and Clinical Chemistry [18] (Color figure online)
6 Tiffany W. Guo et al.

We used high-throughput manufacturing microfluidic ­cassettes

(Fig. 1b, c (Right)) in transparent polystyrene which is easily scal-
able at low cost and with features ranging from tens of microns to
millimeter scales. A preloaded reagent on cassette delivers a speci-
fied sequence of reagents over a series of detection zones on the
cassette, each with dense meanders coated with capture proteins.
Sample and reagents are eventually collected and stored as waste in
a membrane filter on the cassette, driven by the vacuum pressure
from a diaphragm pump within the device. The signal detection
module, consisting of pairs of low cost light emitting diodes
(LEDs) and photodetectors, and the data communication compo-
nent are integrated into the device and controlled by an 8-bit
microcontroller with peripheral components packaged onto a
printed circuit board. A satellite modem and GSM/GPRS module
enable flexibility and low costs in data communication through the
use of short burst data and short message services, while maintain-
ing secure transmission of results and mechanisms for patient
confidentiality.

2  Materials

2.1  Microfluidic 1. Plastic microfluidic cassette (customized, OPKO Diagnostics).

Immunoassay 2. 2  μg/mL HIV Antigen, gp41-gp36 envelope chimera for
HIV-1 and HIV-2 (Biolink).
3. 10 μg/mL Rabbit anti-goat IgG (Invitrogen).
4. Bovine Serum Albumin (BSA).
5. 10× phosphate buffered saline (PBS): 140 mM NaCl, 15 mM
Phosphate buffer, pH 7.3. For 1× PBS, dilute the 10× stock
with ACS grade water.
6. ACS grade water.
7. Blocking agent/sample diluent: 1 % BSA and 0.05 % Tween
20 in 1× PBS.
8. 50 mM Carbonate–Bicarbonate buffer pH 9.6 (Sigma-Aldrich).
9. 1.6 μg/mL Gold conjugated anti-human IgG (Customized,
OPKO diagnostics).
10. Gold Diluent: 3 % BSA–0.2 % Tween 20 in 1× PBS.
11. Wash buffer: 0.05 % Tween 20 in 1× PBS.
12. Polyethylene (PE) tubing (Zeus Inc.).
13. Polycarbonate (PC) tubing (customized, OPKO Diagnostics).
14. Clear Sealing adhesive tape (Customized, OPKO Diagnostics).
15. Silver Enhancer A and Silver Enhancer B (customized, OPKO
Diagnostics).
16. 60 mL syringe.
 Mobile Device for Resource-Limited Settings 7

2.2  Determining 17. Super-bright red light-emitting diodes (Digikey).

Assay Results 18. Photodiodes (Hamamatsu).
19. 8-bit microcontroller (ATmega32, Atmel).

2.3  Results 20. Compact satellite modem (Iridium 9601 SBD Transceiver).
Communication 21. GSM/GPRS module (LinkSprite SM5100B).

3  Methods

3.1  Microfluidic 1. Plastic cassettes (Fig. 1c (Right)) are sonicated for 10 min in
Immunoassay deionized water mixed with detergent, rinsed off with deion-
ized water and dried using N2 gas to remove any dust or par-
3.1.1  Surface
ticulates, which may block the microfluidic channels.
Modification of COC
Cassette 2. Next, 20 μL of each antigen is spotted on top of their corre-
sponding meander zones for physisorption (Fig. 1e). Working
concentration of antigens and positive control are prepared in
carbonate–bicarbonate buffer.
3. Test format for zones: negative control–HIV 1/2–positive
control (see Note 1).
4. Physisorption of antigens and antibodies is carried out in
humid chamber at room temperature for 2 h.
5. After physisorption, droplets are aspirated and each meander
zone is washed two times with 1× PBS to remove any unbound
antigen and dried with N2 gas.
6. The cassettes are sealed by application of a clear adhesive tape
over the entire cassette surface.
7. Channels are filled by flowing through the blocking agent
using negative pressure. Blocking agent is incubated in chan-
nels for 1 h at room temperature. This step is important to
prevent any nonspecific protein adsorption.
8. The channels are cleared and the cassettes are either run imme-
diately or stored in dry chamber at 4 °C.

3.1.2  Reagent Loading 1. Working solution (75× dilution) of patient sample (Sera/
Plasma) is prepared in sample diluent (see Note 2).
2. Working solution of gold conjugated anti hIgG is prepared in
gold diluent.
3. A series of reagents is loaded manually using 1 mL syringe to
draw reagents into PE tubing. Air spacers of ~1.6 μL (0.3 cm)
are used to separate the reagents.
4. For sera/plasma, the reagent sequence (Fig. 1d) was a plug of
diluted sample ~27 μL, four plugs of wash buffer ~1.6 μL each,
8 Tiffany W. Guo et al.

one plug of diluted gold conjugated anti-human IgG ~13.5 μL,

two plugs of wash buffer ~1.6 μL each, and four plugs of dis-
tilled water ~1.6 μL each.
5. A PE tube with reagents is connected to the reagent storage
port using a short PC tube connector to fit the inlet hole of
microfluidic cassette to load the reagent plugs into the on-­
board reagent storage using negative pressure.
6. Silver Enhancer A and B are loaded into another part of reagent
storage using a micropipettor.

3.1.3  Assay Operation To operate the test, user draws less than one microliter of whole
blood (or 5 cm of diluted sera/plasma) in a capillary tube, con-
nects the tube into a microfluidic cassette, and inserts the cassette
into device; the tube forms a fluidic connection between the pre-
loaded reagents and the detection zones. Sample and reagents are
drawn through the cassette by vacuum generated by a micropump.
The liquid handling module includes a diaphragm micropump
(Fig. 1c) to generate negative pressure, a vacuum regulator to reg-
ulate the vacuum pressure applied to the microfluidic chip, and
two pressure sensors to monitor pressure at the outlet of the pump
and the inlet of the chip in real-time. With the use of pressure sen-
sors, we are able to precisely control the pressure drop and flow
rate in the fluidic chip. The ability to vary flow rate over a wide
range via vacuum pressure affords flexibility in adjusting assay time.
Operating within the range of vacuum pressure in the mChip
device, we found it possible to achieve flow rates of 2.5–20 μL/min
using a hand pump as a substitute. At the optimal pressure, the
running times of the plugs are ~4.5 min for sera/plasma samples,
~2.5 min for the gold anti-human IgG antibody, and ~25 s per
wash. After the final water wash, silver solution A and silver solu-
tion B (composed of silver nitrate and reducing agent, respectively)
is mixed through the mixing point on the cassette before passing
through the meanders. Silver reagents are run continuously for
4 min (see Note 3). The overall assay time should be 20 min or less
(see Note 4).

3.2  Data Acquisition We programmed the device to run silver enhancement for 4 min
and Analysis and measure optical transmittance through each analysis zone at
the beginning and end of the duration of silver enhancement.
At the end of the assay, a silver film develops at the detection zones
according to the amount of analytes captured. The signal detection
module of the mChip device contains four pairs of low-cost, super-­
bright red light-emitting diodes (LEDs) and photodetectors and a
microcontroller. Each of the four detection zones on the fluidic
chip is sandwiched between a pair of LED and photodetector. The
silver development determines the amount of red light (660 nm)
that can be transmitted. The intensity of transmitted light mea-
sured by the photodetector is converted from analog signals to
 Mobile Device for Resource-Limited Settings 9

digital values by the microcontroller. The absorbance value or

­optical density (OD) is calculated using the following equation:

æ I ö
OD = - log10 ç ÷
è Io ø

Io and I = light intensities measured at the detection zones before

and after the assay, respectively. Io should be taken before the assay,
while channels are filled with water or wash buffer. I should be
taken exactly after 4 min of continuous silver solution in the
channels.
After running numerous samples with known disease status,
receiver-operator curves (ROC) can be made to illustrate tradeoffs
between sensitivity and specificity at different cutoff values.
The ROC is used to determine a normalized OD cutoff value for
classifying disease positive and negative results.

A compact satellite modem (Iridium 9601 SBD Transceiver) and a

3.3  Results
GSM/GPRS module (LinkSprite SM5100B) are integrated to the
Communication
POC device to communicate diagnostic results from a remote set-
3.3.1  Hardware ting to a centralized database (Fig. 2). The satellite modem oper-
and Software Set-Up ates with a data service called “short burst data” (SBD) provided
by the Iridium satellite network. SBD is used due to its low
operational cost, with minimal latency in data transmission
­
(approximately 5 s) and global coverage. The basic architecture of
the Iridium system includes the satellite network, the ground net-

Email:
Diagnostic results:
Perform assay: From: [email protected]
Subject: SBD Msg From Unit: 300034013408780 Name Value
Date: May 26, 2010 4:28:44 PM EDT
satellite To:
Reply-to:
Sam Sia
[email protected]
UNIT_ID 1
1 Attachment, 13 bytes Save Quick Look TEST_ID 1
MOMSN: 0 TEST_DATE 11/18/09
MTMSN: 0
PATIENT_ID 1001
Time of Session (UTC): Wed May 26 20:28:43 2010
Session Status: 00 – Transfer OK decode AGE 36
Message Size (bytes): 13

Unit Location: Lat = 40.816348 Long = -74.009028 GENDER Female

CEPracius = 2
PREGNANT TRUE
3000340134...d (13 bytes)
RESULT_0 0.02
RESULT_1 0.485
RESULT_2 0.52
SMS message: RESULT_3 0.459

cellphone 01001da60fd249051e5821cb00

Fig. 2 Representative responses obtained during the data communication step. At the end of each assay, the
optical density of each detection zone was displayed on the LCD screen. Transmission of the diagnostic results
is initiated via the satellite or cell phone network. The former resulted in the receipt of an e-mail to a predes-
ignated e-mail address, with the diagnostic data as attachments. Note that the testing location was also
reported in this e-mail. If the results were transmitted via the cell phone network, 13 characters (as hexadeci-
mals) representing the encoded diagnostic results were received as an SMS message at a predesignated cell
phone number. Finally, the results were decoded into readable values. In this case, observed optical densities
are 0.02, 0.485, 0.52, and 0.459 at the four detection windows respectively [17]
10 Tiffany W. Guo et al.

Fig. 3 Picture of mChip device with an external antenna at Muhima Hospital in

Kigali, Rwanda. Reprinted with copyright permission from Clinical Chemistry [18]

work (Earth gateways), and a satellite modem. When a user initi-

ates a data transmission from the field, the modem connects to an
overhead satellite, and the data is relayed among satellites around
the globe until it reaches a satellite that is above the appropriate
Earth gateway, which downloads the data and sends it as an e-mail
to a predesignated e-mail address via the Internet. An extendable
antenna is also connected to the POC device to enable RF com-
munication with the satellite (Fig. 3). The GSM/GPRS module
can connect to any portable network operating in the EGSM900,
GSM850, DCS1800, or PCS1900 band frequency. The module is
able to communicate with the microcontroller, receive the coded
results and send it as a SMS message to a pre-stored recipient num-
ber. An advantage of attempting to send the message via GSM first
and satellite second is that less power is consumed than in the
reverse sequence.

3.3.2  Messaging Protocol In order to minimize the size of data to be transmitted (and
hence reduce the cost of data transmission), all the data is format-
ted, including date/time of test, patient’s information, and test
results, into a 15-byte binary string. As opposed to sending data
on a character-­by-character basis, the messaging format treats
each type of data (mobile device identifier number, date and time
of test, patient identifier number, age, gender, pregnancy status,
absorbance values to three decimal places for each analysis zone,
 Mobile Device for Resource-Limited Settings 11

Bit position Bit 119-112 Bit 111-95 Bit 94-83 Bit 82-73 Bit 72-66 Bit 65
Name UNIT_ID TEST_DATE TEST_TIME PATIENT_ID AGE GENDER
Bit 64 Bit 63-54 Bit 53-44 Bit 43-34 Bit 33-24 Bit 23-16 Bit 15-0
PREGNANT RESULT_0 RESULT_1 RESULT_2 RESULT_3 RESERVE CHECK

Name Bit length Description

UNIT_ID 8 Instrument identifier, value 0-255
TEST_DATE 17 Time of test, in the format of (MMDDYY)
TEST_TIME 12 Time of test, in the format of (HHMM)
PATIENT_ID 10 Patient identifier, value 0-1023
AGE 7 Patient’s age
GENDER 1 Patient’s gender, Male = 1, Female = 0
PREGNANT 1 Pregnancy? True = 1, False = 0
RESULT_0 10 Raw absorbance value of zone 1, value 0-1023
RESULT_1 10 Raw absorbance value of zone 2, value 0-1023
RESULT_2 10 Raw absorbance value of zone 3, value 0-1023
RESULT_3 10 Raw absorbance value of zone 4, value 0-1023
RESERV 8 Reserved bits for error code / notification
CHECK 16 2-byte checked sum required by service provider
total 120 bit (15 bytes)

Fig. 4 Messaging protocol for formatting diagnostic data into a 15-byte binary string to reduce the size of data
to be transmitted. In this study, private fields related to patient’s information (e.g., age, gender, and pregnancy
status) were not used. Reprinted with copyright permission from Clinical Chemistry [18]

and error notification) as a numerical value and transmits it as a

binary string. This format allows us to significantly reduce data
size while maintaining the privacy of patient data and the amount
of information being transmitted. For instance, transmitting the
test date “123109” (i.e., 12/31/2009) as 8-bit characters will
require 48 bits, whereas transmitting it as a binary string
“11110000011100101” (binary representation of 121309) will
require only 17 bits. This streamlined messaging format is to
maintaining a low operating cost to transmit POC diagnostics
results via satellite communication (using the inexpensive SBD
service). “Short burst data” service (SBD, for messages under
205 bytes) can be used to enable the reliable use of satellite-based
transmission at cost that matches SMS. While transmission of
large files is still relatively costly, we encoded all the diagnostic
information of each test into a tiny 15 byte binary file. Our sam-
ple messaging protocol is shown in Fig. 4.

3.3.3  Data Confidentiality The data transmission must be secure. As one example, because
in Transmission data is first encoded as a 15-byte binary string before it is transmit-
ted through the satellite network, decoding the information on the
receiver end would require a password-protected lookup table.
Within the data messages, even if they are intercepted and decoded,
patients are identified only by numerical ID’s known only to health
care providers. In the future, if even higher level of data security is
required, Iridium SBD transceivers can transmit data using the
Enhanced Mobile Satellite Services (EMSS). EMSS is a Department
12 Tiffany W. Guo et al.

of Defense (DoD) enhancement to allow commercially available

mobile satellite technologies (such as Iridium) to perform data
transmission through a DoD dedicated gateway, with unique DoD
features such as end-to-end encryption and protection of sensitive
user information.

3.3.4  Automation An 8-bit microcontroller (ATmega32, Atmel) is used to control

with a Microcontroller and integrate all the three modules together. All the electronic
components (including the microcontroller, power regulator, and
pressure sensors) are mounted onto a small custom designed
printed circuit board (6.4 × 6.4 cm) that operates on a 9 V battery.
A software routine is programmed in C into the microcontroller of
the POC device such that: (1) it prompts the user to initiate a test;
(2) controls the liquid handling and signal detection modules with
precise timing; (3) at the end of the test, it computes absorbance
values, displays results to the user on a liquid crystal display (LCD),
and prompts the user to perform data transmission; (4) when a
button is pushed subsequently, the microcontroller formats the
test data into a string of binary numbers and initiate a transmission
session with the satellite modem or the GSM/GPRS module using
AT commands (a set of standard instructions for controlling
modem); (5) and reports the result of data transmission to the
user. An Atmel AVR© STK500 development board is used along
with the AVR Studio 4 IDE to program the microcontroller.

4  Notes

1. This platform can be extended for other immunoassays for

both antigens and antibodies detection, in both singleplex and
multiplex formats. The first step to adapt this platform for a
new disease is attachment of the target marker to the plastic
surface. Multiple techniques for surface chemistry modifica-
tion, for example, avidin–biotin, may be used to create chemi-
cal bonds between capture molecules and surface to strengthen
the attachment. Furthermore, blocking agents, amount of cap-
ture molecules, amount of secondary antibodies, incubation
time, and other parameters in the assay may need to be opti-
mized to achieve the best performance for each application.
2. While running the tests with sera or plasma samples, we dilute
the sample to eliminate prozone effect or hook effect. The
dilution at which the antibody titer is just enough to give out
a positive result for a positive sample is chosen. For whole
blood, we use neat sample and lower concentration of gold
conjugated antibodies.
3. The silver enhancement time is variable depending on the
test format and antigens tested on the mChip platform.
 Mobile Device for Resource-Limited Settings 13

Real-time observation and measurement of silver development

can be used to optimize run time. Optimized conditions will
allow for maximum range, or difference in optical density
between positive control and negative control zones.
4. Certain runs can be rejected if there is significant signal in the
negative control or insufficient signal in the positive control.
Errors can be introduced due to poor sample integrity from
degradation, significant lag time between mixing the silver
reagents and introduction to microfluidic channels, presence
of dust in reagent tubes, or debris introduced into the channels
that can obstruct or slow flow. Syringe-driven vacuum pressure
can be disconnected and reapplied in case of clogging or
obstructed flow to help clear the channel.
Text from Subheadings 1, 3.2 and 3.3 were reprinted from
ref. 18, with permission from Clinical Chemistry.

References
1. Morris K (2009) Mobile phones connecting 8. Willyard C (2010) Focus on electronic health
efforts to tackle infectious disease. Lancet records. Electronic records pose dilemma in
Infect Dis 9:274 developing countries. Nat Med 16:249
2. Lester R, Karanja S (2008) Mobile phones: 9. Fraser HS, Biondich P, Moodley D et al (2005)
exceptional tools for HIV/AIDS, health, and cri- Implementing electronic medical record sys-
sis management. Lancet Infect Dis 8:738–739 tems in developing countries. Inform Prim
3. Zurovac D, Sudoi RK, Akhwale WS et al Care 13:83–95
(2011) The effect of mobile phone text-­ 10. Nsanzimana S, Ruton H, Lowrance DW et al
message reminders on Kenyan health workers’ (2012) Cell phone-based and internet-based
adherence to malaria treatment guidelines: a monitoring and evaluation of the national
cluster randomised trial. Lancet 378:795–803 antiretroviral treatment program during
4. Pop-Eleches C, Thirumurthy H, Habyarimana rapid scale-up in Rwanda: Tracnet, 2004–
JP et al (2011) Mobile phone technologies 2010. J Acquir Immune Defic Syndr
improve adherence to antiretroviral treatment 59:e17–e23
in a resource-limited setting: a randomized 11. NIH. A summit on mobile technologies as a
controlled trial of text message reminders. platform for health research and healthcare
AIDS 25:825–834 delivery. http://www.fic.nih.gov/news/
5. Lester RT, Ritvo P, Mills EJ et al (2010) events/mhealthsummit.htm. Accessed Nov
Effects of a mobile phone short message ser- 2012
vice on antiretroviral treatment adherence in 12. Schreiner S (2009) Inaugural mhealth summit
Kenya (WelTel Kenya1): a randomised trial. successfully brings together health, technology
Lancet 376:1838–1845 and policy communities to advance techno-
6. Bellina L, Missoni E (2009) Mobile cell-­ logical innovation in global healthcare.
phones (m-phones) in telemicroscopy: increas- Business Wire. October 30, 2009
ing connectivity of isolated laboratories. Diagn 13. Kost GJ (2002) Part III: Point-of-care testing
Pathol 4:19 in health system, community and field. In:
7. Blaya JA, Cohen T, Rodriguez P et al (2009) Kost GJ (ed) Principles & practice of point-of-
Personal digital assistants to collect tuberculo- care testing. Lippincott Williams & Wilkins,
sis bacteriology data in Peru reduce delays, Baltimore, p 386
errors, and workload, and are acceptable to 14. CDC. Flu activity & surveillance. http://www.
users: cluster randomized controlled trial. Int J cdc.gov/flu/weekly/fluactivitysur v.htm.
Infect Dis 13:410–418 Accessed Nov 2012
14 Tiffany W. Guo et al.

15. Dzenowagis J, Kernen G (2005) Connecting 17. Cheung YK (2010) Fabrication of multi-­
for health: global vision, local insight. World component hydrogel microstructures and micro-
Health Organization, Geneva. Accessed Dec devices. Dissertation, Columbia University.
2012 Publication Number: 3447863
16. Chin CD, Laksanasopin T, Cheung YK et al 18. Chin CD, Cheung YK, Laksanasopin T et al
(2011) Microfluidics-based diagnostics of (2013) Mobile device for disease diagnosis and
infectious diseases in the developing world. data tracking in resource-limited settings. Clin
Nat Med 17:1015–1019 Chem. doi:10.1373/clinchem.2012.199596
 Chapter 2

Microfluidic Devices for Nucleic Acid (NA) Isolation,

Isothermal NA Amplification, and Real-Time Detection
Michael G. Mauk, Changchun Liu, Mohamed Sadik, and Haim H. Bau

Abstract
Molecular (nucleic acid)-based diagnostics tests have many advantages over immunoassays, particularly
with regard to sensitivity and specificity. Most on-site diagnostic tests, however, are immunoassay-based
because conventional nucleic acid-based tests (NATs) require extensive sample processing, trained opera-
tors, and specialized equipment. To make NATs more convenient, especially for point-of-care diagnostics
and on-site testing, a simple plastic microfluidic cassette (“chip”) has been developed for nucleic acid-
based testing of blood, other clinical specimens, food, water, and environmental samples. The chip com-
bines nucleic acid isolation by solid-phase extraction; isothermal enzymatic amplification such as LAMP
(Loop-mediated AMPlification), NASBA (Nucleic Acid Sequence Based Amplification), and RPA
(Recombinase Polymerase Amplification); and real-time optical detection of DNA or RNA analytes. The
microfluidic cassette incorporates an embedded nucleic acid binding membrane in the amplification reac-
tion chamber. Target nucleic acids extracted from a lysate are captured on the membrane and amplified at
a constant incubation temperature. The amplification product, labeled with a fluorophore reporter, is
excited with a LED light source and monitored in situ in real time with a photodiode or a CCD detector
(such as available in a smartphone). For blood analysis, a companion filtration device that separates plasma
from whole blood to provide cell-free samples for virus and bacterial lysis and nucleic acid testing in the
microfluidic chip has also been developed. For HIV virus detection in blood, the microfluidic NAT chip
achieves a sensitivity and specificity that are nearly comparable to conventional benchtop protocols using
spin columns and thermal cyclers.

Key words Point-of-care, On-site, Molecular diagnostics, DNA, RNA, Infectious diseases, Pathogen,
Microfluidics, Enzymatic, Amplification, LAMP, Isothermal

1 Introduction

On-site detection of pathogens and other disease markers can

improve the quality and lower the costs of health care [1–10].
Portable on-site tests can also monitor the safety of food [11, 12]
and water supplies [13]. Clinical specimens can be processed in
credit-card sized, plastic microfluidic cartridges (“chips”) that can
be coupled to cell phones and other portable devices for detection,
data analysis, and communications. In particular, a cell phone

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_2, © Springer Science+Business Media New York 2015

15
16 Michael G. Mauk et al.

camera can be used to detect fluorescent signals such as used to

monitor the production of amplicons during the enzymatic ampli-
fication process. This “lab-on-a-chip” technology provides for
Point-of-Care (POC) diagnostics and serves as crucial enabling
technology for mobile health care (mHealth). For example, the
smartphone will transmit the test results to the patient’s medical
files, to the health provider, and, when appropriate and/or required
by law, to public health officials. Based on the test results, the
smartphone application could download instructions from the
cloud to guide the patient. The health provider can contact a phar-
macy and order prescriptions, provide additional guidance remotely,
and encourage/discourage an office visit as appropriate.
This chapter describes the development of a microfluidic cas-
sette for nucleic acid based pathogen assays (with specific applica-
tion to HIV viral load testing), and which is compatible with cell
phone-based end-point detection of fluorescence signals generated
in real-time enzymatic amplification of a pathogen target. The
approach and methods reported here are sufficiently generic and
can be readily adapted to other POC diagnostics applications.
Broadly, in vitro bioanalytical methods can be classified into
three categories: (1) immunoassays that rely on specific antigen–
antibody binding (or analogous protein interactions) for isolation
and labeling of analytes such as antibodies and/or antigens;
(2) nucleic acid-based (molecular) tests (NATs) that utilize
sequence-specific nucleic acid hybridization, and for which the
sensitivity can be greatly enhanced by enzymatic amplification; and
(3) cell-based methods incorporating cell fractionation and sort-
ing, selective labeling, and/or cytometry which are often amenable
to target amplification via cell culturing. Most existing rapid
(<hour), on-site test devices are based on immunoassays [14, 15]
since NATs are comparatively complicated and culturing and other
cell-based methods are time-consuming.
Immunoassays are operationally simple because they are toler-
ant of sample heterogeneity and require few, if any, sample prepara-
tion and processing steps. Immunoassays can be performed with
raw or minimally processed specimens such as blood, urine, saliva,
cell culture, tissue, water, and environmental samples. For exam-
ple, the lateral flow (LF) strip format offers an elegant, simple, and
inexpensive implementation of immunoassays [16]. LF strip immu-
noassay tests, such as home pregnancy tests and drugs-of-abuse
tests, are widely available.
Nucleic acid tests (NAT), also known as molecular assays, have a
crucial advantage over immunoassays in that nucleic acids can be
amplified in vitro by sequence-specific enzymatic reactions, thus
facilitating highly sensitive detection. A single target DNA molecule
can be replicated a billion times within an hour. The specificity of the
test can be tailored by appropriate primer design. Typically, nucleic
acid-based tests offer much greater (often 1,000-fold or more)
Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 17

sensitivity and specificity than immunoassays. Nucleic acid-based

tests can also provide information that cannot be readily obtained
with immunoassays such as discrimination between drug-susceptible
and drug-resistant pathogens and the identification of genes and
gene transcription profiles. Despite their many advantages, molecu-
lar assays are still not commonly used at the point of care and are
generally restricted to centralized laboratories since nucleic acid-
based tests typically require elaborate sample processing to release,
isolate, and concentrate the nucleic acids and remove substances that
inhibit enzymatic amplification. Conventional nucleic acid testing
requires benchtop equipment such as centrifuges, water baths, ther-
mal cyclers, and gel readers; cold storage for labile reagents; dedi-
cated lab areas and hoods to avoid contamination, and highly trained
personnel. Moreover, for molecular analysis of blood specimens,
cell-free plasma is preferred. The use of plasma instead of whole
blood in NATs avoids problems associated with inhibitors (such as
hemoglobin in red blood cells) [17–19], clogging of filters or porous
membranes with cells and cell debris, and complications in interpre-
tation of results related to nucleic acids associated with white blood
cells [20]. The plasma is typically separated from whole blood by
centrifugation. However, such and similar plasma extraction adds
an extra processing step to NAT, further burdening point of care
(POC) applications.
The objective of microfluidics implementations of nucleic acid
tests is to make NAT almost as easy-to-use as LF strip test devices.
As an illustration, we describe a single-use (disposable), plastic,
microfluidic cassette or cartridge (“chip”) that hosts fluidic net-
works of conduits, reaction chambers, porous membrane filters,
and inlet/outlet ports for sample processing and analysis. The
sequential steps of sample metering, lysis of the pathogen target,
NA isolation, reverse transcription (for RNA targets), enzymatic
amplification primed with target-sequence oligos, amplicon label-
ing, and detection are integrated in the microfluidic chip. Fluid
actuation and flow control, temperature control, and optical detec-
tion are provided by supporting instrumentation. Completely
automated operation (without any human intervention) is feasible.
Many microfluidic NAT devices [21–23], including our earlier
prototypes [24–27], utilize PCR (polymerase chain reaction) for
nucleic acid amplification. For example, Chen et al. [26] describe a
microfluidic cassette for PCR-based nucleic acid detection. The
palm-sized cassette mates with a portable instrument [28] that
provides temperature regulation using thermoelectric elements,
solenoid actuation of pouches and diaphragm valves formed on the
chip for flow control and pumping, and LED/photodiode detec-
tion of amplification products labeled with an intercalating fluores-
cent dye. The time needed from sample loading to obtaining test
results is typically less than 1 h.
18 Michael G. Mauk et al.

Although PCR technology is highly developed and PCR

primers sequences are available for many targets, PCR is not opti-
mal for on-site applications. PCR requires precise (±1 °C or better)
temperature control and rapid (>5 °C/s) temperature ramping,
which complicates implementation and increases the cost of instru-
mentation. The high temperatures (~95 °C) required for PCR
places demands on chip design, necessitating strong bonding of
chip components to withstand the pressure of the heated reaction
mixture due to expanding trapped air and thermal expansion of the
liquid phase and tight sealing of the amplification chamber to avoid
evaporation. As an alternative to PCR, isothermal amplification
methods are much easier to implement in on-site applications.
Constant-temperature operation lowers energy consumption and
even allows the use of small-scale exothermic chemical reactions
for heating without a need for any electric power [29]. Thus, a
considerable simplification in chip design and operation is realized
with isothermal amplification methods [30–34] that require,
depending on the amplification scheme, an incubation temperature
ranging from 40 to 65 °C. Examples of isothermal amplification
include Loop-mediated AMPlification (LAMP) to detect DNA
targets [35], one-step reverse transcription (RT) with cDNA
LAMP (RT-LAMP) to detect RNA targets [36–38], Nucleic Acid
Sequence Based Amplification (NASBA) [39, 40], Recombinase
Polymerase Amplification (RPA) [41, 42], and helicase-dependent
amplification (HDA) [43–45]. Tests using isothermal amplifica-
tion are typically faster than PCR (roughly, 30 min vs. 60 min).
Also, LAMP appears to be more robust than PCR. It is more toler-
ant of temperature variations and less susceptible to the adverse
effects of inhibitors [46]. In this chapter, we focus on LAMP. We
have, however, successfully tested our chips with NASBA and RPA
(results not shown) and obtained comparable results to the ones
obtained with LAMP.
To streamline sample processing and flow control, we use
“multifunctional” amplification reactors that incorporate porous
flow-through membranes for immobilization of nucleic acids [47, 48].
A disc-shaped silica- or cellulose-based membrane is embedded in
the chip at the inlet of each reaction chamber. Nucleic acid cap-
ture, washing, amplification, and detection are all combined in a
single chamber that houses the NA binding membrane. The target
nucleic acid captured on the membrane serves as a template for
amplification, thus obviating the need for a separate membrane
elution step (as done in spin column formats), and thereby simpli-
fying chip design and flow control. Multiple amplification reactors
can be accommodated in the same chip (Fig. 1) for concurrent
detection of several targets, as well as calibration reactions, and
positive and negative controls.
The lab-on-a-chip format can facilitate autonomous point-of-
care molecular diagnostics with various levels of automation,
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 19

Fig. 1 (Left ) Photograph of a chip containing an array of three multifunctional amplification reactors. (Right ) A
cross section of a multifunctional amplification reactor featuring the flow-through membrane for the capture
and immobilization of nucleic acids

instrumentation, and functionality. For more convenient use, the

chips can be preloaded with lyophilized amplification reagents
[49–53]. During chip assembly, the LAMP reagents (enzymes,
primers, dyes, salts) are measured and mixed in correct proportions
and pipetted into the chip amplification chamber, dried by subli-
mation, and coated with a low-melting point (~55 °C) paraffin wax
layer [52]. The chip is then sealed with a capping layer. The wax
encapsulation protects the pre-stored reagents when the sample
and wash solutions flow through the amplification reactor. When
the chamber is filled with water and heated to the LAMP incuba-
tion temperature (~65 °C), the wax coating melts, just in time, and
releases the LAMP reagents to amplify the target NAs captured on
the membrane.
For brevity, we describe in this chapter a manually operated
NAT module. The sample and reagents are loaded into the chip
by pipetting in the appropriate sequence. This manually operated
version realizes only some of the advantages of microfluidic sys-
tems. For a commercially viable system, the reagents will be pre-
stored in the chip and the fluid delivery automated. The module
described here is, however, a substantial step towards the ultimate
goal of fully automated operation. We describe the design, fabri-
cation, and operation of the multifunctional reactor module. As a
representative application, we describe the detection of an RNA
virus (HIV) in blood plasma samples. Adaptations of the chip and
protocol for other targets and applications are comparatively
minor. We have used approaches similar to the one reported here
for the detection of lambda phage in water, Gram-positive bacteria
(B. cereus) [24] and gram-negative bacteria (Salmonella) in buffer,
E. coli in urine and stool samples, and HIV virus in saliva [25],
20 Michael G. Mauk et al.

and for genotyping malaria-transmitting mosquito subspecies

using insect tissue [48]. The chips, with modifications, can also be
used for gene profiling as may be needed for cancer screening
[54]. To accommodate blood-based samples, we describe a simple
uninstrumented filtration device for separating plasma from whole
blood. The extracted plasma is then subjected to a chemical lysis
step to solubilize nucleic acids from viral or bacterial pathogens in
the extracted plasma. The plasma-derived lysate is then loaded
into the microfluidic chip for assay of a sequence-specific nucleic
acid biomarker.

2 Materials

Materials used include:

1. Polymer chip materials stock: acrylic (polymethylmethacrylate
PMMA, CYRO Industries, Rockaway, NJ or various suppliers
such as McMaster-Carr) or polycarbonate (PC, e.g., Lexan™
from ePlastics.com or various suppliers such as McMaster Carr)
or cyclic olefin polymer (COC, Topas, Florence, KY), approx.
0.1 and 1 mm thick sheets.
2. Bonding solvent for PMMA: acetonitrile (Sigma-Aldrich); for
PC: acetone (Sigma-Aldrich).
3. Chip sealing tape: Microseal “B” Adhesive (Bio-Rad, Hercules, CA).
4. Surface passivants (optional).
(a) BSA (bovine serum albumin, 1–2 %, Sigma-Aldrich),
(b) PVP-40 (polyvinylpyrrolidone 40, Sigma-Aldrich) or poly-
ethylene glycol (1–2 % PEG, 6,000 MW, Sigma-Aldrich).
5. Polyimide-based, thin film heater (HK5572R7.5L23A, Minco
Products, Inc., Minneapolis, MN).
6. DC power supply (e.g., Model 1611, B&K Precision
Corporation, CA).
7. Nucleic Acid binding phases:
(a) Silica-based phases: Whatman™ GF/F borosilicate glass
fiber filter paper (0.42 mm thick, 0.7 μm pore size), or
(b) Cellulose-based membranes (Whatman FTA™ paper:
3MM Whatman cellulose filter paper, 0.5 mm thick, with
dried Tris–HCl, EDTA, and SDS, which can be washed off
the FTA paper before mounting in the chip).
8. Harris Unicore™ (2 or 3-mm diameter, Ted Pella, Inc.,
Redding, CA) for cutting discs of binding phase for insertion
in chip.
9. Commercial kits with reagents (binding/lysis buffer, wash
buffers, elution buffers) for isolation of nucleic acids:
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 21

(a) Qiagen Blood and Tissue kit (Valencia, CA).

(b) Roche HighPure™ Viral RNA kit or Qiagen QIAamp® kit
were used for RNA isolation.
(c) Elution buffer: Tris-acetate EDTA (TAE) buffer (10×)
(Sigma-Aldrich).
10. Enzymatic Amplification Kits
(a) Loopamp™ DNA amplification kit (Eiken Chemical Co.
Ltd.,Tochigi, Japan).
(b) RPA kit (TwistDX, Cambridge, UK), and the
(c) NASBA kit (Biomérieux, Durham, NC).
11. RNase inhibitor (Ambion®, Life Technologies) for amplifica-
tion reactions with the reverse transcription steps.
12. Fluorescent dyes for nucleic acid detection:
(a) SYTO-9 Green (Invitrogen Corp. Carlsbad, CA) or
(b) EvaGreen (Biotium, Hayward, CA).
13. Positive HIV controls of known virus concentration
(Acrometrix®, Benecia, CA).
14. Published LAMP primer sets [38], HPLC purified (e.g., custom
synthesized by Sigma-Aldrich).
(Suggested vendors are included for convenience, but compa-
rable substitutions are generally permissible.)

3 Methods

3.1 Chip Design The microfluidic chips (Fig. 1) are fabricated in clear (transparent)
and Fabrication plastic materials as bonded laminates. We have tested three differ-
ent thermoplastic polymers: polycarbonate (PC), acrylic (PMMA),
and cyclic olefin copolymer (COC) (Table 1). The chips have been
rapidly prototyped using computer-aided design (CAD) and
computer-aided manufacturing software (SolidWorks™,
MasterCAM™, or AutoCAD™). The 0.1–1-mm thick plastic
sheets are cut with a computer-controlled (CNC) milling machine
(e.g., Haas Office Mill OM-1, Haas Automation, Oxnard, CA) or
with a CO2 laser (e.g., Universal Laser Systems, Scottsdale, AZ,
30–50 W power) to define the microfluidic circuit(s) and other
features (Fig. 2). Chip fabrication and bonding methods for these
materials are discussed in more details in references [55–62]. The
smallest feature size is the channel width, which is ~800 μm. For
acrylic-based chips, the component layers are bonded with a weak
solvent such as acetonitrile. These polymer materials exhibit low
surface adsorption and usually do not require any special surface
treatments; however, dynamic surface passivation with BSA (bovine
serum albumin, 1–2 %), PVP-40 (polyvinylpyrrolidone 40), or
22 Michael G. Mauk et al.

Table 1
Thermoplastic polymer stock for chipsa

Optical
Trade CO2 Laser transparency Autofluore-
Material names cutting (nm) scence Bonding Cost
“Acrylic” Plexiglass®, Yes 300–1,200 Moderate Thermal/ Low
Polymethyl- Lucite®, to high, pressure, solvent
methacrylate Perspex® depending (acetonitrile),
(PMMA) on particular ultrasonic
PMMA welding
Polycarbonate Lexan® No 250–1,200 Moderate to Thermal/pressure, Low
(PC) relatively high solvent (acetone)
Cyclic olefin TOPAS® Yes 200–1,200 Low Thermal/pressure, Moderate
copolymer COC solvent (hexane)
(COC)
a
These materials are available in sheets of thicknesses ranging from 0.1 to 3 mm. One layer of the chip is made from a
PCR sealing film (PCR Sealers™ Microseal B, Bio-Rad, Hercules, CA). This is a PCR-compatible adhesive tape with
good transparency and low autofluorescence

Fig. 2 Microcluidic NAT chip CAD drawings (SolidWorks™) showing assembled, bonded structure and compo-
nent layers. (a) Assembled chip. (b) Top view showing the inlets and outlets and the PMMA tape covering the
reaction chambers. (c) Underside view showing the PCR tape covering the membrane isolation chamber and
conduits. (d) Exploded view. (e) Top view of the main slab showing the uncovered amplification chambers.
(f) Underside view of the main slab (without the PCR sealing tape) showing the isolation chambers with
embedded membranes and the conduits
Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 23

polyethylene glycol (1–2 % PEG, 6,000 MW), added to the reagent

mix, can enhance the amplification efficiency, presumably by reduc-
ing surface adsorption of template, enzymes, and primers [63–66].
The chip materials are generally optically transparent and allow
observation of fluid flow during operation. Portions of the chip
that require transmission of exciting light and fluorescence emis-
sion are made sufficiently thin to minimize autofluorescence.
Autofluorescence associated with various chip materials has been
characterized [67]. We note that the use of PCR-compatible seal-
ing tapes for sealing the chamber avoid any unwanted adsorption
of bonding solvent on the NA binding membrane.
The chip shown in Fig. 1 measures 46 mm × 36 mm × 3.50 mm
and consists of three layers: a top layer made with 250 μm (0.01 in.)
thick PMMA film, a 3 mm (0.118 in.) thick PMMA chip body, and
a 250 μm (0.01 in.) thick PCR Sealers™ tape bottom. Each reac-
tion chamber is ~5 mm long, ~1 mm wide, and 3 mm deep, to
form a volume of ~15 μl. Smaller volumes are also feasible. Both
the top PMMA film and the PCR Sealers™ tape bottom were cut
with a CO2 laser. The chip body (middle layer) was milled with a
(CNC) milling machine to form three separate reactors, membrane
disc supports, and access conduits [47, 48]. The top PMMA film
was solvent-bonded with acetonitrile at room temperature.
Residual solvent was removed by overnight heating at 50 °C. The
solvent bonding can be effected by stacking and aligning the layers
and running a bead (~5–10 μl) of solvent around the periphery of
the stack using a pipette. The solvent wicks into the interface
between adjacent layers and forms a strong bond after about
10 min (see Note 1).
The experimental setup for on-chip LAMP amplification and
end-point fluorescent detection is shown in Fig. 3. Briefly, the
resistive heating system consists of a chip support equipped with a
flexible, polyimide-based, thin film heater (HK5572R7.5L23A,
Minco Products, Inc., Minneapolis, MN) and a type T thermo-
couple (Omega Engineering, each wire 75 μm in diameter, and a
junction diameter of 170 μm). The thermocouple junction was
placed at the interface between the heater and the chip. The chip,
once filled with the LAMP master mix, was fixed to the chip sup-
port with a double-sided adhesive tape, allowing the reactors to
form a good thermal contact with the thin film heater. The heater
was powered by a DC power supply (Model 1611, B&K Precision
Corporation, CA). The power supply was adjusted to maintain the
reactors at 63 ± 0.5 °C. Although the LAMP process is fairly forgiv-
ing to temperature variations, in field applications, it would be nec-
essary to use a closed-loop thermal controller to accommodate
operation over the broad range of ambient temperatures that may
be encountered in various regions and times.
24 Michael G. Mauk et al.

Fig. 3 Experimental Setup for operating microfluidic cassette showing cassette mounted on base supporting
heater and blue LED. Detector or camera is positioned over cassette. Optical filter blocks blue excitation light
to allow detection of green fluorescence from LAMP reaction(s) on cassette. A power/control box provides
power for heater and LED, and auxilliary functions such as data logging and communication with smartphone.
Inset shows photo of real-time LAMP reaction on cassette

3.2 Use of Nucleic The chip features solid-phase extraction to isolate nucleic acids
Acid Binding Phases from samples. The chip described here is based on the method of
for Isolation of Nucleic Boom et al. [68–70] using chaotropic salts to promote selective
Acids binding of NAs to a (silica-based) solid phase, and as adapted by
others [71–77] for microfluidic implementation. Several types of
porous membranes for the nucleic acid (NA) binding phase have
been employed, including “silica” (Whatman™ GF/F borosilicate
glass fiber filter paper, 0.42 mm thick, 0.7 μm pore size) and
cellulose-based membranes (Whatman FTA™ paper: 3MM
Whatman cellulose filter paper, 0.5 mm thick, with dried Tris–HCl,
EDTA, and SDS, which can be washed off the FTA paper before
mounting in the chip). Both these membranes can be readily cut to
the desired shape and size (e.g., 1.5 mm diameter) using a hole
punch (Harris Unicore™, Ted Pella, Inc., Redding, CA), and then
inserted into the chip during its assembly. Other membranes such
as nanoporous alumina (Whatman Nanopore™) can also function
as nucleic acid binding phases, but are difficult to cut to size and
mount in the chip due to their brittleness [52, 53] (see Note 2).
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 25

Table 2
Reagents and buffers for on-chip isolation of DNA targets (i.e., B. cereus)

Reagents Qiagen DNEasy™ kit Generic reagents

Lysis/binding buffer AL lysis buffer (with 5 M 6 M Guanidine HCl, 1 % Tween®, SDS
guanidine HCl) (sodium dodecyl sulfate), or Triton X®
Proteinase K (optional)
Wash buffer 1 AW1 buffer 50 % ethanol; 50 % water
Wash buffer 2 AW2 buffer 50 % ethanol; 50 % water

Table 3
Reagents and buffers for on-chip isolation of viral RNA targets (i.e., HIV virus)

Qiagen QIAamp™
Reagent step viral RNA kit Roche High Pure™ Viral RNA kit Generic reagents
Lysis/binding AVL buffer (contains Lysis/binding buffer 6 M Guanidine HCl
buffer guanidine thiocyanate) Inhibitor removal buffer (4.5 M or 6 M guanidine
guanidine-HCl, 50 mM Tris– thiocyanate
HCl, 30 % Triton® X-100)
Wash buffer 1 AW1 buffer (contains Inhibitor removal buffer (5 M 50 % ethanol; 50 %
guanidine HCl), guanidine-HCl, 20 mM water
~50 % ethanol Tris–HCl, 40 % (v/v)
ethanol, pH 6.6)
Wash buffer 2 AW2 buffer, ~50 % Wash Buffer (approx. 50 % (v/v) 50 % ethanol; 50 %
ethanol ethanol, 20 mM NaCl, 2 mM water
Tris–HCl, pH 7.5)

3.3 Reagents During development stages, reagents from commercial nucleic

acid isolation kits can be used, or alternatively, NA isolation can be
done with generic components as specified in Tables 2, 3 and 4.
DNeasy™ Blood and Tissue kit, which includes AL (lysis and bind-
ing) buffer and AW1 and AW2 ethanol-based wash buffers, was
purchased from Qiagen Inc. (Valencia, CA). Roche HighPure™
Viral RNA kit or Qiagen QIAamp® kit were used for RNA isola-
tion. The Loopamp™ DNA amplification kit was obtained from
Eiken Chemical Co. Ltd. (Tochigi, Japan). The RPA kit was sup-
plied by TwistDX (Cambridge, UK), and the NASBA kit was from
Biomérieux (Durham, NC). RNase inhibitor (Ambion®, Life
Technologies) was added to the amplification reactions with the
reverse transcription steps. SYTO-9 Green (Invitrogen Corp.
Carlsbad, CA) or EvaGreen (Biotium, Hayward, CA) were used as
26 Michael G. Mauk et al.

Table 4
Isothermal amplification specifications

Incubation
Amplification Reaction mix (25 μl, temperature
method Supplier total volume) (°C) Primers
LAMP (Loop- Eiken Chemical 20 mM Tris–HCl (pH 8.8), 60–65 6 primers:
mediated (Tochigi, Japan) 10 mM KCl, 10 mM (NH2) FIP 1.6 μM
amplification) SO4, 8 mM MgSO4, 0.1 % BIP
Tween 20, 0.8 M betaine, 8 U 1.6 μM
Bst DNA polymerase, 1.4 mM Loop-F
dNTPs, and 4.0 μM SYTO® 9 0.8 μM
Green intercalating dye, 8 U Loop-B
RNase inhibitor 0.8 μM
F3 0.2 μM
B3 0.2 μM
RPA (recombinase TwistDX Rehydration buffer 30 μl 37–40 2 primers:
polymerase (Cambridge, UK) Freeze-dried reaction mix primer A
amplification) Template + H2O to 47.5 μl 10 μM
total volume, 4.0 μM SYTO® primer B
9 Green intercalating dye; 10 μM
Add 2.5 μl of 280 mM MgAc
to initiate reaction, 8 U RNase
inhibitor
NASBA (nucleic Biomérieux 5 μl enzyme mix (RNase H, T7- 55–60 2 primers
acid sequence (Durham, NC) RNA polymerase, reverse primer F
based transcriptase), 10 μl NASBA 0.2 μM
amplification) buffer + 5 μl sample, 4.0 μM primer R
SYTO® 9 Green intercalating 0.2 μM
dye, 8 U RNase inhibitor
Positive Controls and their primers are included in these kits
Reverse transcriptase is added to reaction mix for RNA targets, e.g., 0.63 U AMV reverse transcriptase (Invitrogen,
Carlsbad, CA)

DNA binding dyes and were added directly to the amplification

reaction mix at recommended dilutions. Acetonitrile, acetone,
ethanol, and Tris-acetate EDTA (TAE) buffer (10×) were supplied
by Sigma-Aldrich and were used without further purification. For
HIV LAMP testing, we used dilutions of positive controls of
known virus concentration (Acrometrix®, Benecia, CA) and pub-
lished LAMP primer sets [38].

3.4 Fluorescent Fluorescence detection of the amplification products requires a

Reporters nucleic acid binding dye [78] or a molecular beacon [79] as a
reporter of amplicons. For real-time detection, the dye or beacon
is included in the amplification reaction mix. The intercalating dye
nonspecifically binds the double-stranded amplification product as
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 27

Table 5
A few fluorescent reporters for real-time amplification

Excitation Emission
Dye Mechanism wavelength (nm) wavelength (nm)
Ethidium bromide Intercalation of ds-DNA 300/360 590
Loop-Amp Amplification phosphate by product 240–370 515
Fluorescent binds Mg and induces fluorescence 340 (peak)
(calcein)
SYBR Green I Intercalation of ds-DNA 497 520
SYTO-9 Intercalation of ds-CNA 485 498
EvaGreen™ Hybridization/intercalation 490 530

well as sample “background” DNA co-isolated with the target, and

any artifacts such as primer-dimers or misprimed amplicons. The
fluorescence emission signal monotonically increases as more
amplicons are produced. Molecular beacons work by hybridization
of a doubly functionalized (fluorophore and quencher) oligo probe
to the amplicon, such that quenching of the fluorophore by the
otherwise proximal quencher is relieved, and a fluorescence signal
proportional to amplicons concentration is produced. Molecular
beacons generally produce weaker fluorescence than intercalating
dyes but are less prone to nonspecific targeting. The selection of a
particular reporter determines the excitation and emission wave-
lengths of the fluorescence, which informs the choice of light
source, detector spectral response, spectral optical filters
(if needed), and spectral transparency range of the chip material.
A few common fluorescent reporters that can be used in these
devices are listed in Table 5. These dyes are selected for their known
compatibility with enzymatic amplification, i.e., they are not strong
inhibitors of PCR or isothermal amplification at the recommended
dye concentrations. Too high a dye concentration can, however,
inhibit amplification. The autofluorescence of the plastic chip
material creates a background signal (see below). The NA-binding
membrane material also fluoresces and must be excluded from the
detector’s field of view. The active area of the chip for fluorescence
detection can be framed with opaque (non-fluorescing) tape. We
have also successfully used molecular beacon probes based on fluo-
rescence resonance energy transfer (FRET) to measure amplicon
production.

3.5 Real-Time On-chip real-time detection of amplicons requires: (1) an excitation

Detection light source with spectral emission that matches the strong absorp-
tion band of the fluorophore; (2) the transmission of the excitation
28 Michael G. Mauk et al.

light into the amplification chamber(s); (3) filtering out the excita-
tion spectra to prevent its detection; and (4) the detection of the
(longer wavelength) light emitted from the reaction chamber with
a detector having appropriate spectral responsivity. Various
approaches to fluorescence detection on chips are feasible [80–85].
To detect the fluorescence signal on the chip, we have used either
(1) a commercial compact fluorometer such as the Qiagen ESE
Fluo-Sens reader (Model ESE Model ESML 10-MB-3007 with
520 nm excitation and silicon photodiode detector); (2) a portable
(palm-sized) fluorescence microscope (DinoLite™ Model
AM4113T-GFBW) with seven built-in blue LEDs for illumination,
an emission filter with a 510-nm wavelength cutoff, a CCD camera
detection, and a USB connection; and (3) a cell phone camera in
conjunction with LED excitation and an appropriate filter. Both the
Qiagen ESE reader and the portable fluorescence microscope have
their own excitation sources and optical filters and are suitable for
detection with SYTO Green, SYBR Green, and EVA Green dyes.
The Qiagen ESE reader can view only one reactor at a time and
must be mounted on a scanner when the chip houses an array of
reactors. In contrast, the USB-based microscope and the CCD
camera can monitor multiple reactors simultaneously and do not
require scanning. Both the USB-based microscope and the smart-
phone camera require imaging software (e.g., ImageJ® or
MATLAB®) to quantify and process the detected signal. Given the
ubiquity of cell phones even in Third World countries, the use of a
cell phone is an attractive option. By taking advantage of mobile
devices, we can reduce the cost of the instrumentation needed for
the diagnostic devices while at the same time providing the ability
of transferring test results wirelessly to a centralized depository for
archiving, analysis, and public health monitoring.
For smart cell phone-based detection, one can illuminate the
edge of the chip with blue light LED or laser. The light is guided
into the reaction chambers through the transparent chip material.
Lensing and waveguide structures can also be formed in the chip
during the microfluidic circuit machining process to attain more
uniform excitation of the reactor array [86]. LEDs with emission
wavelengths of 450–500 nm are needed to excite many of the
common dyes (Table 5). Unfortunately, many nominally “blue”
LEDs have a significant green component that overlaps with the
emission of the fluorophore. To minimize interference from the
excitation source, suitable bandpass and longpass colored glass or
interference coating optical filters (Thor Labs, Newton, NJ or
Edmund Scientific, Barrington, NJ) must be used to block excita-
tion light from the detector. For example, when using blue emis-
sion (450–500 nm), we employed a Thor Labs LED465E filter
(2.5-mm diameter, nominal emission peak at 465 nm with spectral
FWHM, full-width half maximum of 50 nm). Note that the plastic
chip materials are not very transmissive for UV (for dyes such as
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 29

Fig. 4 Plasma separation device: (a) whole blood inserted into the separator
chamber, (b) sedimentation of blood components, (c) extraction of plasma.
(d) A photograph of a standalone version of the plasma separation device

EtBr or calcein), so waveguiding light into the edge of the chip

may not be always feasible. As an alternative to fluorescence-based
detection, several colorimetric and turbidity assays have been
reported for isothermal amplification [87, 88].

3.6 Plasma For the analysis of blood samples, plasma is first separated from
Extraction from Whole whole blood. Traditionally, plasma is separated from blood by
Blood Using centrifugation. Centrifugation is, however, inappropriate for on-site
a Filtration Module applications. As an alternative means, we developed a passive, low-
cost, pump-free separator (Fig. 4) that utilizes both sedimentation
and filtration to extract plasma from whole blood [89]. The sedi-
mentation process reduces the concentration of the cells in the
blood that is then filtered through an asymmetric plasma separation
membrane (polysulfone, Vivid™ GR, Pall Corp., Port Washington,
NY). Red cells, white cells, and platelets are trapped without lysis in the
large pores on the upstream side of the membrane. The plasma out-
flows through the smaller pores on the membrane’s downstream side.
The separated plasma is collected with a pipette. The plasma separation
module was tested by extracting about 275 μl of plasma from 1.8 ml of
whole blood with known titer of HIV, in less than 7 min, using only
30 Michael G. Mauk et al.

a pipette to load and retrieve the sample. The plasma was then
subjected to lysis to solubilize target nucleic acids from the virus (or
bacteria) of interest, and processed in our microfluidic chip as
described in the next section.

3.7 Chip Operation We focus on the operation of a microfluidic module that integrates
nucleic acid isolation, amplification, and detection using the multi-
functional amplification reactor described above. For clarity and
brevity, only manual operation of the chip is described. The sample
is lysed off-chip by mixing it with a chaotropic salt (guanidinium
chloride) and other optional lysing agents (e.g., detergents, pro-
teinase K, lysozyme). The lysis deactivates any infectious agents in
the sample as well as nucleases that might degrade target NAs. The
lysate is then pipetted through the membrane and the reaction
chamber (see Note 3). The chaotropic salt promotes binding of the
nucleic acids to the membrane. The membrane-bound nucleic
acids are then washed with buffer to remove debris and inhibitors.
In the normal mode of solid-phase extraction (i.e., spin column),
the captured nucleic acids are eluted from the membrane into the
amplification chamber with a low-salt, pH-neutral buffer. In con-
trast, in our device, we simplify the process by forgoing a separate
elution step. The membrane is sited at the inlet of the amplification
chamber and is in contact with the amplification reaction mixture.
The membrane-bound nucleic acids serve directly as templates for
amplification. An important advantage of this configuration is that
the sample volume (the amount of lysate filtered through the
membrane) is largely decoupled from the volume of the amplifica-
tion reaction. Nevertheless, there is an optimum membrane size
for a given sample volume and amplification chamber size. The
membrane must be large enough to capture a high fraction of the
NA from the perfusing lysate, but not too large as to impede
desorption of captured NA during amplification. Also, the silica-
based membrane material partially inhibits enzymatic amplifica-
tion. The cellulose-based FTA membrane shows a better
compatibility with enzymatic amplification, i.e., less inhibition of
amplification. We speculate that inhibition of amplification by the
membrane material results from irreversible adsorption of tem-
plate, primers, and/or polymerase to the membrane since experi-
ence teaches that inhibition effects are reduced when the
concentrations of these three components are increased in the
reaction mix.
As an example of the chip operation, we describe the detection
of HIV virus in blood. The various processing steps are depicted
schematically in Fig. 5. Plasma separated from whole blood was
subjected to a chemical lysis step using a combination of chaotropic
salts, detergents, and enzymes. Nucleic acids were isolated from the
lysate by solid-phase extraction, and the purified, concentrated,
immobilized nucleic acids were subjected to sequence-specific
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 31

Fig. 5 Processing steps for plasma extraction and microfluidic NAT

enzymatic amplification using targeted oligo primer sequences.

The production of amplicon was then monitored by a DNA-binding
fluorescent dye (Fig. 6). Troubleshooting for Amplification/
Detection Step: See Notes 4–6.

3.8 Steps for a Blood 1. A blood sample laden with the suspected pathogen (e.g., HIV
Sample Molecular virus), ranging in volume from 100 μl to 1 ml (depending on
Analysis on a Chip the required limit of detection), is inserted into the plasma
separator. The plasma separator utilizes a combination of filtra-
tion and sedimentation to separate the plasma from the blood.
2. Add 4 volumes of lysis/binding buffer to 1 volume of plasma
(e.g., add 400 μl lysing buffer to 100 μl plasma) and incubate
for 10 min at room temperature.
3. Optionally, add carrier RNA suspension (20 ng per sample) to
the lysis buffer.
4. Pipette a volume of lysate (50–500 μl) into the chip, perfusing
the membrane.
5. Pipette 100 μl ethanol-based wash solution through the mem-
brane, washing the membrane.
32 Michael G. Mauk et al.

Fig. 6 An example of real time detection. Fluorescence emission intensity as a

function of time when the sample contains 0 (negative control), 102, 103, and 104
HIV copies/ml. (Inset ) The threshold time as a function of HIV concentration

6. Repeat the wash step above.

7. Dry the membrane to remove residual ethanol by aspirating or
blowing dry air through inlet port with a 10-ml syringe.
8. Inject amplification mix (with correct proportion of enzymes,
primers, nucleotides, salts, and dye) into the reactor as propor-
tioned for volume of amplification chamber. Flow the reaction
mix through the membrane and fill the amplification chamber.
9. Seal the chip inlet and outlet with PCR sealing tape.
10. For isothermal LAMP amplification, incubate chip at 65 °C on
a hotplate. (Use a “dummy” calibration chip with embedded
thermocouple to adjust hotplate settings.) Foam rubber can be
placed around the chip (with portal opening for camera) to
reduce temperature fluctuations from ambient room drafts.
11. Excite the fluorescent dye and monitor the emission with
either a fluorometer or a camera.
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 33

12. (optional) If desired, a melting curve can be constructed to

verify that the melting temperature is consistent with the
expected amplicons. Alternatively, the amplification products
can be removed from the reaction chamber and subjected to
gel electrophoresis to verify that the size of the amplicons is
consistent with expectations. As yet another option, instead of
real time detection, the amplification products (when the prim-
ers are appropriately functionalized) can be run on a lateral
flow strip for end point detection.

3.9 Analysis Real-time fluorescence emission intensity as a function of amplifi-

cation time for a LAMP-based chip is shown in Fig. 6. The curves
correspond to samples spiked with different HIV virus concentra-
tions and have a characteristic sigmoidal shape (similar to real-time
PCR). The rapid increase in the fluorescence intensity occurs
sooner as the number of target copies in the sample increases [47].
In other words, the threshold time needed to observe emission
intensity above background level can be correlated with the con-
centration of target copies in the sample (Fig. 6, inset). The thresh-
old time is a linear function of the logarithm of the target
concentration. By developing appropriate calibration curves, one
can infer the unknown number of virus copies in a sample. The
results indicate that our cassette’s detection limit is smaller than
1,000 HIV copies/ml.

3.10 Conclusion The microfluidic module and protocol for nucleic acid isolation
and Discussion and amplification described here are a convenient alternative to
benchtop molecular assays. The microfluidic device combines
nucleic acid isolation typically done using commercial spin column
kits with enzymatic amplification and detection as conventionally
done in Eppendorf tubes in a benchtop thermal cycler instrument.
A sample can be processed in the microfluidic chip in less than an
hour. In several applications, such as analysis of an RNA virus in
blood plasma, limits of detection comparable to benchtop assays
(<1,000 virus particles per ml sample) have been realized. This
“lab-on-a-chip” approach is attractive for processing of a small
number of samples. More importantly, the microfluidic module
described here can serve as a component of an autonomous, por-
table, on-site molecular diagnostic system. Such next-generation
devices may incorporate programmed fluidic actuation and flow
control, as well as smartphone detection of amplicons’ fluorescence
emission. Chips, preloaded with lyophilized reagents, primers for a
specified target, and buffers, could be hermetically packaged,
bar-coded, and stored (without refrigeration) with a shelf-life of a
year or more. Chips, customized for various targets, could operate
with the same portable processor. A simple filtration separation
device that extracts plasma samples from whole blood for microflu-
idic nucleic acid testing could be combined with the NAT chip into
34 Michael G. Mauk et al.

a single integrated microfluidic cartridge. The chip designs are

compatible with high-volume, low-cost injection molding, and a
microfluidic test cartridge is projected to cost several dollars per
test. As such, this approach may make molecular diagnostics feasi-
ble in doctors’ and dentists’ offices, home settings, clinics, ambu-
lances, hospital bedsides, laboratory animal facilities, food
processing and retail outlets.

4 Notes

1. Chip delamination or leaks may be the result of poor bonding.

Weak bonds may result from inadequate cleaning of the chip
materials prior to bonding. Fortunately, the use of relatively
low-temperature (≤65 °C) isothermal amplification as distin-
guished from PCR (rapid thermal cycling up to 95 °C) imposes
less stress on the chips, and delamination and leak problems
are rare in these isothermal amplification chips.
2. Addition of “carrier” RNA to the sample often proves helpful
and sometimes crucial, especially for samples with a low number
of target molecules. Lyophilized poly-A RNA (approx. 20 ng
per sample) is added to the sample to improve NA isolation
yield [90–92]. This may be beneficial due to one or a combina-
tion of several effects: (1) the carrier RNA acts as a sacrificial
substrate for RNAases that would otherwise degrade the tem-
plate; (2) the carrier RNA enhances precipitation and binding
of NA to the solid-phase membrane, and (3) small amounts of
NA are strongly and irreversibly adsorbed on the NA binding
phase and cannot be desorbed to serve as templates for ampli-
fication. The added carrier RNA occupies the strong binding
sites on the NA membrane, so that the weakly bound target
NA can be desorbed. The effect of carrier RNA appears to help
DNA analysis as well.
3. Air bubbles in the fluids may interfere with NA isolation and
amplification, so introduction of air bubbles should be avoided
when pipetting liquids into the chip.
4. No amplification observed from positive controls. Possible
causes include:
(a) The template has degraded. If sufficient amount of material
is available, obtain gel electropherograms of the template.
RNA samples are most susceptible to degradation.
To avoid contamination, extract amplicons for gel analysis
as remotely as possible from where the LAMP reagents are
prepared and loaded into the chips (see below).
(b) Reagents are spoiled. Many kits combine reagent compo-
nents so only mixtures of components can be tested.
Reverse transcriptase is most susceptible to degradation
and is the component that should be examined first.
Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 35

(c) Inhibitors persist. The solid-phase extraction process should

generally remove inhibitors. Inhibitors may be intrinsic to
the sample and/or to the chip due to residual bonding
solvent left over in the chip or membrane or due to resid-
ual ethanol. Sample dilution will reduce the concentration
of inhibitors and help mitigate the effects of sample inhibi-
tors. Increasing membrane drying duration will help to
remove any residual ethanol left over from the wash steps.
In the case of blood samples, anticoagulants (e.g., EDTA
chelates Mg needed for polymerases) may interfere with
the amplification process. Proteinase K and other lysing
agents may degrade polymerases, and unlike in PCR, there
is no high temperature heating (>90 °C) in the isothermal
amplification process to inactivate nucleases and protein-
ases. If certain lysing agents are suspected of inhibiting
amplification, their necessity in the protocol should be
reconsidered, or else more rigorous washes (repetitions
and greater total wash buffer volume) should be used to
enhance their removal.
(d) Incorrect or suboptimal incubation temperature. Temperature
overshoots of 10 °C over the recommended incubation
temperature can degrade enzymes. Verify there is no over-
shoot in temperature during heat-up which may abolish
enzyme activity. Also, too low a temperature will reduce
amplification efficiency.
5. Limit of detection in the chip is significantly lower than parallel
benchtop assay. Possible causes:
(a) Non-optimal temperature control. Check amplification
reactor temperature with a calibration chip equipped with
temperature sensor.
(b) Fluorescent dyes degraded. Dyes lose their fluorescence over
time due to bleaching and excessive freeze-thaw. Minimize
exposure of dyes to light as much as possible. Dyes can be
mixed with calibration DNA (such as gel marker ladders)
and tested in chip with fluorometer.
(c) Low NA yield. Add more carrier RNA to the sample to
improve yield of on-chip NA isolation. Check NA yield:
Assay nucleic acid captured on the NA binding membrane
by eluting with 10–50 μl of warm (~50 °C) water or TE
buffer, and measure (total) NA concentration and purity
(260/280 ratio) with a spectrophotometer (e.g.,
Nanodrop™, Thermo Scientific, Wilmington, Delaware)
or fluorometer (TBS-380 with PicoGreen™ dye for DNA
and RiboGreen™ dye for RNA, Turner Bio Systems,
Sunnyvale, California). Since depending on sample matrix,
36 Michael G. Mauk et al.

(total) NA can range from hundreds of picograms to

hundreds of nanograms, it is probably more meaningful
to compare chip NA yields to spin-column yields of com-
parable samples.
6. Amplification is detected for nominally negative samples/
controls. Possible causes include:
(a) Primer-dimer or nonspecific amplification is creating a false
signal. Run amplification product on gel or produce a
melting curve in a benchtop PCR to verify that the ampli-
fication product has the expected size.
(b) Excessive incubation time may lead to the production of
artifacts such as primer dimers. When using LAMP, limit
the duration of the amplification process to less than
45 min.
(c) Excess Mg results in nonspecific amplification. Adjust Mg
concentration according to manufacturer’s protocol for
amplification. Note that EDTA and some other anticoagu-
lants used in blood samples are chelators for Mg.
(d) Contamination from previous runs. Due to their very high
efficiency, the isothermal amplification techniques are par-
ticularly sensitive to contamination. This problem affects
the assay developers, not on-site users. As a precaution, we
prepare reagents and load the chips in separate labs and use
dedicated pipettes. Chips should be left sealed after ampli-
fication to avoid the spreading of any amplicons. Gel elec-
trophoresis of amplification product creates an aerosol of
amplicons, and so should be done as far away from the
sample prep and test area as possible. Experience shows
that molecular biology grade water is often the source of
contamination. When you purchase primers, aliquot the
primers (25–50 runs worth) and cold store separately
(preferably in a distant lab) to avoid contaminating the
entire primer batch. Unfortunately, once the lab area is
contaminated with amplicons, it may prove difficult to
alleviate, so constant diligence is strongly advised.

Acknowledgments

The work reported here was supported, in part, by NIH Grants

U01DE017855 (Bau, Mauk) and K25AI099160 (Liu), and a
grant from the Commonwealth of Pennsylvania’s Ben Franklin
Technology Development Authority through the Ben Franklin
Technology Partners of Southeastern Pennsylvania (Bau, Sadik).
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 37

References
1. Olasagasasti F, Ruiz de Gordoa J (2012) HIV-1 and coinfections. J Clin Microbiol
Miniaturized technology for protein and 49(10):3584–3590
nucleic acid point-of-care testing. Transl Res 15. Lin C-C, Wang J-H, Wu H-W, Lee G-B
160(5):332–345 (2010) Microfluidic immunoassays. J Lab
2. Weigl B, Domingo G, LaBarre P, Gerlach J Autom 15:253–274
(2008) Towards non- and minimally instru- 16. Ngom B, Guo Y, Wang X, Bi D (2010)
mented, microfluidics-based diagnostics Development of lateral flow strip technology
devices. Lab Chip 81:1999–2014 for detection of infectious agents and chemical
3. Wang S, Xu F, Demirci U (2010) Advances in contaminants. Anal Bioanal Chem 397(3):
developing HIV-1 viral load assays for 1113–1135
resource-limited settings. Biotechnol Adv 17. Al-Soud WA, Rådström P (2001) Purification
28:770–781 and characterization of PCR-inhibitory com-
4. Chin CD, Linder V, Sia SK (2012) ponents in blood cells. J Clin Microbiol
Commercialization of microfluidic point-of- 39(2):485–493
care diagnostic devices. Lab Chip 12: 18. Opel KL, Chang D, McCord BR (2010) A
2118–2134 study of PCR inhibition mechanisms using real
5. Myers FB, Henrikson RH, Bone J, Lee LP time PCR. J Forensic Sci 55(1):25–33
(2013) A handheld point-of-care genomic 19. Wiedbrauk DL, Werner JC, Drevon AM
diagnostic system. PLoS One 8(8):e70266 (1995) Inhibition of PCR by aqueous and vit-
6. Giljohann DA, Mirkin CA (2009) Drivers of reous fluids. J Clin Microbiol 33(10):
biodiagnostic development. Nature 462(26): 2643–3646
461–464 20. Burns DE, Ashwood ER, Burtis CA (2007)
7. Gubala V, Harris LF, Ricco AJ, Tan MX, Fundamentals of molecular diagnostics.
Williams DE (2012) Point of care diagnostics: Saunders, St. Louis
status and future. Anal Chem 84(2):487–515 21. Zhang C, Xing D (2007) Minaturized chips
8. Fu E, Yager P, Fioriano PN, Christodoulides for nucleic acid amplification and analysis: lat-
N, McDevitt J (2011) Perspectives on diag- est advances and future trends. Nucleic Acids
nostics for global health. IEEE Pulse 2(6): Res 35(13):4223–4237
40–50 22. Park S, Zhang Y, Lin S, Wang T-H, Yang S
9. Mairfair J, Roppert K, Ertl P (2009) (2011) Advances in microfluidic PCR for
Microfluidic systems for pathogen sensing: a point-of-care infectious disease diagnostics.
review. Sensors 9:4804–4823 Biotechnol Adv 29:830–839
10. Mondal S, Venkataraman V (2007) 23. Lee J-G, Cheong KH, Huh N, Kim S, Choi
Miniaturized devices for DNA amplification J-W, Ki C (2006) Microchip-based one step
and fluorescence based detection. J Indian Inst DNA extraction and real-time PCR in one
Sci 87(3):309–332 chamber for rapid pathogen identification. Lab
11. Yoon J-Y, Kim B (2012) Lab-on-a-chip patho- Chip 6:886–895
gen sensors for food safety. Sensors 12: 24. Wang J, Chen Z, Corstjens PL, Mauk MG,
10713–10741 Bau HH (2006) A disposable microfluidic cas-
12. Tourlousse DM, Ahmad F, Stedtfeld RD, sette for DNA amplification and detection.
Seyrig G, Tiedje JM, Hashham SA (2012) A Lab Chip 6(1):46–53
polymer microfluidic chip for quantitative 25. Chen Z, Mauk MG, Wang J, Abrams WR,
detection of multiple water and foodborne Corstjens PL, Niedbala RS, Malamud D, Bau
pathogens using real-time fluorogenic loop- HH (2007) A microfluidic system for saliva-
mediated isothermal amplification. Biomed based detection of infectious diseases. Ann N Y
Microdevices 14:769–778 Acad Sci 1098:429–436
13. Sakamoto C, Yamaguchi N, Yamada M, 26. Chen D, Mauk M, Qiu X, Liu C, Kim J,
Nagase H, Seki M, Nasu M (2007) Rapid Ramprasad S, Ongagna S, Abrams WR,
quantification of bacterial cells in potable water Malamud D, Corstjens PLAM, Bau HH
using a simplified microfluidic device. J (2010) An integrated, self-contained microflu-
Microbiol Methods 68:643–647 idic cassette for isolation, amplification, and
14. Lochhead MJ, Todorof K, Delaney M, Ives JT, detection of nucleic acids. Biomed
Greef C, Moll K, Rowley K, Vogel K, Myatt C, Microdevices 12(4):705–719
Zhang X-Q, Logan C, Benson C, Reed S, 27. Qiu X, Mauk MG, Chen D, Liu C, Bau HH
Schooley RT (2011) Rapid multiplexed immu- (2010) A large volume, portable, real-time
noassay for simultaneous serodiagnosis of PCR reactor. Lab Chip 10(22):3170–3177
38 Michael G. Mauk et al.

28. Qiu X, Chen D, Liu C, Mauk MG, Kientz T, 41. Piepenburg O, Williams CH, Stemple DL,
Bau HH (2011) A portable, integrated ana- Ames NA (2006) DNA detection using recom-
lyzer for microfluidic-based molecular analysis. bination proteins. PLoS Biol 4(7):e204
Biomed Microdevices 13:809–817 42. Lutz S, Weber P, Focke M, Faltin B, Hoffman
29. Liu C, Mauk MG, Hart R, Qiu X, Bau HH J, Müller C, Mark D, Roth G, Munday P,
(2011) A self-heating cartridge for molecular Armes N, Piepenberg O, Zengerle R, von
diagnostics. Lab Chip 11:2686–2692 Stetten F (2010) Microfluidic lab-on-foil for
30. Chang C-C, Chen C-C, Wei S-C, Lu H-H, nucleic acid analysis based on isothermal
Liang Y-H, Lin C-W (2012) Diagnostic recombinase polymerase amplification (RPA).
devices for isothermal nucleic acid amplifica- Lab Chip 10:887–893
tion. Sensors 12:8319–8337 43. Vincent M, Xu Y, Kong H (2004) Helicase-
31. Duarte C, Salm E, Dorvel B, Reddy B Jr, dependent isothermal DNA amplification.
Bashir R (2013) On-chip parallel detection of EMBO Rep 5(8):795–800
foodborne pathogens using loop-mediated 44. Chow WH, McCloskey C, Tong Y, Hu L, You
isothermal amplification. Biomed Microdevices Q, Kelly CP, Kong H, Tang YW, Tang W
15(5):821–830 (2008) Application of isothermal helicase-
32. Fang X, Liu Y, Kong J, Jiang X (2010) Loop- dependent amplification with a disposable
mediated isothermal amplification integrated detection device in a simple sensitive stool test
on microfluidic chips for point-of-care quanti- for toxigenic Clostridium difficile. J Mol Diagn
tative detection of pathogens. Anal Chem 10(5):452–458
82:3002–3006 45. Mahalanabis M, Do J, Al-Muayad H, Zhang
33. Wu Q, Jin W, Zhou C, Han S, Yang W, Zhu Q, JY, Klapperich CM (2010) An integrated dis-
Jin Q, Mu Y (2011) Integrated glass microde- posable device for DNA extraction and heli-
vice for nucleic acid purification, loop- case dependent amplification. Biomed
mediated isothermal amplification, and online Microdevices 12:353–359
detection. Anal Chem 83:3336–3342 46. Francois P, Tangomo M, Hibbs J, Bonetti E-J,
34. Xu L-Y, Jin W, Zhu Q-Y, Yang W-X, Wu Q-Q, Boehme CC, Notomi T, Perkins MD, Shrenzel
Jin Q-H, Mu Y (2011) Real-time detection of J (2011) Robustness of loop-mediated isother-
loop-mediated isothermal amplification reac- mal amplification reaction for diagnostics
tion on microfluidic chip. 2011 5th applications. FEMS Immunol Med Microbiol
International conference on bioinformatics 62:41–48
and biomedical engineering, Wuhan, China, 47. Liu C, Geva E, Mauk MG, Qiu X, Abrams
IEEE, Piscataway, New Jersey, 2011 WR, Malamud D, Curtis K, Owen SM, Bau
35. Notomi T, Okayama H, Masubuchi H, HH (2011) An isothermal amplification reac-
Yonekawa T, Watanabe K, Amino N, Hase T tor with an integrated isolation membrane for
(2000) Loop-mediated isothermal amplifica- point-of-care detection of infectious diseases.
tion of DNA. Nucleic Acids Res 28(12):E63 Analyst 136:2069–2076
36. Mori Y, Kanda H, Notomi T (2013) Loop- 48. Liu C, Mauk MG, Hart R, Bonizzoni M, Yan
mediated isothermal amplification (LAMP): G, Bau HH (2012) A low-cost microfluidic
recent progress in research and development. J chip for rapid genotyping of malaria-transmitted
Infect Chemother 19(3):404–411 mosquitoes. PLoS One 7(8):e42222
37. Parida M, Posadas G, Inoue S, Hasebe F, 49. Stevens DY, Petri CR, Osborn JL, Spicar-
Morita K (2004) Real-time reverse transcrip- Mihalic P, McKenzie KG, Yager P (2008)
tion loop-mediated isothermal amplification Enabling a microfluidic immunoassay for the
for rapid detection of West Nile virus. J Clin developing world by integration of on-card dry
Microbiol 42(1):257–263 reagent storage. Lab Chip 8(12):2038–2045
38. Curtis KA, Rudoph DL, Owen SM (2008) 50. Fridley GE, Le HQ, Fu E, Yager P (2012)
Rapid detection HIV-1 by reverse- Controlled release of dry reagents in porous
transcription, loop-mediated isothermal ampli- media for tunable temporal and spatial distri-
fication (RT-LAMP). J Virol Methods 151: bution upon rehydration. Lab Chip 12(21):
264–270 4321–4327
39. Compton J (1991) Nucleic acid sequence- 51. Hitzbleck M, Gervais L, Delamarche E (2011)
based amplification. Nature 350:91–92 Controlled release of reagents in capillary-
40. Gulliksen A, Solli LA, Drese KS, Sörensen O, driven microfluidics using reagent integrators.
Karlsen F, Rogne H, Hovig E, Sirevåg R Lab Chip 11:2680–2685
(2005) Parallel nanoliter detection of cancer 52. Kim J, Byun D, Mauk MG, Bau HH (2009) A
markers using polymer chips. Lab Chip 5(4): disposable, self-contained PCR chip. Lab Chip
416–420 9(4):606–612
 Microfluidic Devices for Nucleic Acid (NA) Isolation, Isothermal NA… 39

53. Kim J, Mauk MG, Chen D, Qiu X, Kim J, Gale 68. Boom R, Sol CJ, Salimans MM, Jansen CL,
B, Bau HH (2010) A PCR reactor with an Wertheim-van Dillen PM, van der Noordaa J
integrated alumina membrane for nucleic acid (1990) Rapid and simple method for purifica-
isolation. Analyst 135:2408–2414 tion of nucleic acids. J Clin Microbiol
54. Ziober B, Mauk M, Chen Z, Bau HH (2008) 28:495–503
Lab-on-a-chip for oral cancer screening and 69. Greenfield L, White TJ (1993) Sample prepa-
diagnosis. Head Neck 30(1):111–121 ration methods. In: Persing DH, Smith TF,
55. Becker H, Gärtner C (2008) Polymer micro- Tenover FC, White TJ (eds) Diagnostic molec-
fabrication technologies for microfluidic systems. ular biology: principles and applications.
Anal Bioanal Chem 390:89–111 American Soc. Microbiology, Washington,
56. Black I (1998) Laser cutting of perspex. DC, pp 122–137
J Mater Sci Lett 17:1531–1533 70. Kemp BM, Monroe C, Smith DG (2006)
57. Pfleging W, Baldus O (2006) Laser patterning Repeat silica extraction: a simple technique for
and welding of transparent polymers for the removal of PCR inhibitors from DNA
microfluidic device fabrication. Laser-based extracts. J Archaeol Sci 33:1680–1689
micropackaging. Proc SPIE 6107:e610705 71. Breadmore MC, Wolfe KA, Arcibal IG, Leung
58. Steigert J, Haeberle S, Brenner T, Müller C, WK, Dickson D, Giordano BC, Power ME,
Steinert CP, Koltay P, Gottschlich N, Reinecke Ferrance JP, Feldman SH, Norris PM, Landers
H, Rühe J, Zengerle R, Ducrée J (2007) Rapid JP (2003) Microchip-based purification of
prototyping of microfluidic chips in COC. DNA from biological samples. Anal Chem
J Micromech Microeng 17:333–341 75:1880–1886
59. Snakenborg D, Klank H, Kutter JP (2004) 72. Chen X, Cui DF, Liu C (2006) High purity
Microstructure fabrication with a CO2 laser DNA extraction with a SPE microfluidic chip
system. J Micromech Microeng 14:182–189 using KI as the binding salt. Chin Chem Lett
60. Tsao C-W, DeVoe DL (2009) Bonding of 17(8):1101–1104
thermoplastic polymer microfluidics. 73. Chen X, Cui D, Sun J, Zhang L, Li H (2013)
Microfluid Nanofluid 6:1–16 Microdevice-based DNA extraction method
61. Hsu Y-C, Chen T-Y (2007) Applying Taguchi using green reagent. Key Eng Mater
methods for solvent-assistant PMMA bonding 562–565:1111–1115
technique for static and dynamic μ-TAS 74. Cady NC, Stelick S, Batt CA (2003) Nucleic
devices. Biomed Microdevices 9:513–522 acid purification using microfabricated struc-
62. Umbrecht F, Müller D, Gattiker F, Boutry tures. Biosens Bioelectron 30(19):59–66
CM, Neuenschwander J, Sennhauser U, 75. Wen J, Legendre LA, Bienvenue JM, Landers
Hierold C (2009) Solvent assisted bonding of JP (2008) Purification of nucleic acids in
polymethylmethacrylate: characterization using microfluidic devices. Anal Chem 80:
the response surface methodology. Sens 6472–6479
Actuators A Phys 156:121–128 76. Wolfe KA, Breadmore MC, Ferrance JP, Power
63. Kodzius R, Xiang K, Wu J, Yi X, Gong X, ME, Conroy JF, Norris PM, Landers JP
Foulds IG, Wen W (2012) Inhibitory effect of (2002) Toward a microchip-based solid-phase
common microfluidic materials on PCR. Sens extraction method for isolation of nucleic
Actuators B 161:349–358 acids. Electrophoresis 23:727–733
64. Giordano BC, Copeland ER, Landers JP 77. Chung YC, Jan M-S, Lin Y-C, Lin J-H, Cheng
(2001) Towards dynamic coating of glass W-C, Fan C-Y (2004) Microfluidic chip for
microchip chambers for amplifying DNA via high efficiency DNA extraction. Lab Chip
polymerase chain reaction. Electrophoresis 4:141–147
22:334–340 78. Gudnason H, Dufva M, Bang DD, Wolff A
65. Gonzalez A, Grimes R, Walsh EJ, Dalton T, (2007) Comparison of multiple DNA dyes for
Davies M (2007) Interaction of quantitative real-time PCR: effects of dye concentration
PCR with polymeric surfaces. Biomed and sequence composition on DNA amplifica-
Microdevices 9:261–266 tion and melting temperature. Nucleic Acids
66. Lou XJ, Panaro NJ, Wilding P, Fortina P, Kricka Res 35(19):e127
LJ (2004) Increased amplification efficiency of 79. Leone G, van Schijndel H, van Gemen B,
microchip-based PCR by dynamic surface pas- Kramer FR, Schoen CD (1998) Molecular
sivation. Biotechniques 36:248–251 beacon probes combined with amplification by
67. Piruska A, Nikcevic I, Lee SH, Ahn C, NASBA enable homogeneous, real-time detec-
Heineman WR, Limbach PA, Seliskar CJ tion of RNA. Nucleic Acids Res 26(9):
(2005) The autofluorescence of plastic materi- 2150–2155
als and chips measured under laser fluores- 80. Lee D-S, Chang B-H, Chen P-H (2005)
cence. Lab Chip 5:1348–1354 Development of a CCD-based fluorimeter for
40 Michael G. Mauk et al.

real-time PCR machine. Sens Actuators B 87. Goto M, Honda E, Ogura A, Nomoto A,
107:872–881 Hanaki K (2009) Colorimetric detection of
81. Walczak R, Bembnowicz P, Szczepanska P, loop-mediated isothermal amplification reac-
Dziuban JA, Golonka L, Koszur J, Bang DD tion using hydroxy naphthol blue.
(2008), Miniaturized system for real-time Biotechniques 46(3):167–172
PCR in low-cost disposable LTCC chip with 88. Tomita N, Mori Y, Kanda H, Notomi T
integrated optical waveguide. Twelfth Int’l (2008) Loop-mediated isothermal amplifica-
conference on miniaturized systems for chem- tion (LAMP) of gene sequences and simple
istry and the life sciences, San Diego. pp visual detection of products. Nat Protoc
1078–1080 3(5):877–882
82. Zhu H, Yaglidere O, Su T-W, Tseng D, Ozcan 89. Liu C, Mauk M, Gross R, Bushman FD,
A (2011) Cost-effective and compact wide- Edelstein PH, Colman RG, Bau HH (2013)
field fluorescent imaging on a cell-phone. Lab Membrane-based, sedimentation-assisted
Chip 11:315–322 plasma separator for point-of-care applications.
83. Lee D-S, Chou WP, Yeh SH, Chen PJ, Chen PH Anal Chem 85:10463–10470
(2011) DNA detection using commercial mobile 90. Gallagher ML, Burke WF Jr, Orzech K (1987)
phones. Biosens Bioelectron 26:4349–4354 Carrier RNA enhancement of recovery from
84. Ahmad F, Seyrig G, Tourlousse DM, Stedtfeld dilute solutions. Biochem Biophys Res
RD, Tiedje JM, Hashsham SA (2011) A Commun 144(1):271–276
CCDbased fluorescence imaging for real-time 91. Shaw KJ, Thain L, Docker PT, Dyer CE,
loop-mediated isothermal amplification-based Greenman J, Greenway GM, Haswell SJ
rapid and sensitive detection of waterborne (2009) The use of carrier RNA to enhance
pathogens on microchips. Biomed DNA extraction from microfluidic-based silica
Microdevices 13:929–937 monoliths. Anal Chim Acta 652:231–233
85. Myers FB, Lee LP (2008) Innovations in opti- 92. Shaw KJ, Oakley J, Docker PT, Dyer CE,
cal microfluidic technologies for point-of-care Greenman J, Greenway GM, Haswell SJ
diagnostics. Lab Chip 8:2015–2031 (2008) DNA extraction, using carrier RNA,
86. Ogilvie IRG, Sieben VJ, Floquet CFA, Zmijan integrated with agarose gel-based polymerase
R, Mowlem MC, Morgan H (2010) Reduction reaction in a microfluidic device. Twelfth
of surface roughness for optical quality international conference on miniaturized sys-
microfluidic devices in PMMA and COC. tems in chemistry and the life sciences, San
J Micromech Microeng 20:065016 Diego. pp 1069–1071
 Chapter 3

Mobile Based Gold Nanoprobe TB Diagnostics

for Point-of-Need
B. Veigas, E. Fortunato, and P.V. Baptista

Abstract
Nanotechnology based diagnostics has provided improved tools for pathogen detection and sensitive and
specific characterization of antibiotic resistance signatures. Tuberculosis (TB) is caused by members of
the Mycobacterium tuberculosis Complex (MTBC) and, according to the World Health Organization, is
one of the most serious infectious diseases in the world. Recent advances in molecular diagnostics of TB
have improved both the detection time and sensitivity but they still require specialized technical person-
nel and cumbersome laboratory equipment. Diagnostics at point-of-need is crucial to TB control as it
may provide rapid identification of pathogen together with the resistance profile of TB strains, originated
from single nucleotide polymorphisms (SNPs) in different loci, allowing for a more accurate indication of
the adequate therapy.
Gold nanoparticles have been widely used in molecular diagnostics platforms. Here, we describe the
use of gold nanoprobes (oligonucleotide functionalized gold nanoparticles) to be used in a non-cross-
linking colorimetric method for the direct detection of specific DNA targets. Due to the remarkable optical
properties of gold nanoparticles, this detection system provides colorimetric detection of the pathogen
together with the potential of identification of several single nucleotide polymorphisms (SNPs) involved in
TB resistance to antibiotics. For point-of-need use, we adapted this strategy to a low-cost mobile scheme
using a paper based revelation platform and where the spectral signature is transposed to RGB data via a
smartphone device. This way, identification of pathogen and characterization of resistance signatures is
achieved at point-of-need.

Key words Gold nanoparticles, Nanoprobes, Colorimetric method, Mycobacterium tuberculosis,

Antibiotic resistance, Paper-based diagnostics, Point-of-need detection

1 Introduction

Nanotechnology has provided new and improved approaches for

clinical diagnostics with increased sensitivity at lower costs, suit-
able to be used at point-of-need. Nanodiagnostics may circumvent
current limitations of conventional molecular diagnostic methods
by lowering the complexity of the detection step, thus making pro-
cedures suitable for use in remote locations without robust
laboratory setups without compromising sensitivity and specificity.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_3, © Springer Science+Business Media New York 2015

41
42 B. Veigas et al.

Despite the wide range of nanoscale systems being used for

biomolecular assays, in general, nanoparticle-based systems, such
as gold, have been the most widely used for TB diagnostics [1, 2].
Due to their remarkable optical properties, gold nanoparticles
(AuNPs) have been widely used for DNA/RNA screening
approaches [3]. Amongst these amazing properties, the intense
bright colors and ease of functionalization with relevant biomo-
lecular moieties for bio-recognition (e.g., antibodies and ssDNA
oligonucleotides), have paved the way for several molecular diag-
nostics strategies. Solutions containing AuNPs usually exhibit an
intense red color derived from the localized surface plasmon reso-
nance (LSPR) band, centered around 520 nm for 14 nm diameter
NPs; changes to the medium dielectric may induce AuNP aggre-
gation that results in a red-shift of the LSPR and the solution
changes to blue [4].
Tuberculosis (TB) is caused by members of the Mycobacterium
tuberculosis Complex (MTBC) and is today one of the most serious
infectious diseases in the world, responsible for 1.1 million deaths
and 8.8 million new cases every year [5]. Furthermore, the emer-
gence of multidrug-resistant TB also represents a serious threat to
the TB control and an increasing public health problem [6], lead-
ing to a global need for rapid drug susceptibility testing. Single
nucleotide sequence variations (mutations and/or polymorphisms)
within M. tuberculosis (MTb) genome have been associated with
antibiotic resistance, making these sequences ideal targets for the
development of molecular drug susceptibility testing [7–11].
Several approaches have been developed to improve TB diagnos-
tics, by reducing the detection time, costs and by enhancing easi-
ness, thus allowing their application in resource-poor countries
where the main TB epidemic is observed.
Baptista and coworkers have successfully developed the first
colorimetric method for tuberculosis diagnostics based on
Au-nanoprobes (AuNPs functionalized with sequence specific
ssDNA oligonucleotides via a thiol bond) [12–16]. This low-
complexity assay was capable of detecting a specific DNA sequence
from rpoB gene (RNA polymerase beta-subunit) common to all
MTBC members [12]. The method was also adapted towards
detection of MTBC specimens and characterization of mutations
associated with antibiotic resistance in just a few hours [12, 16, 17].
This assay relies on the colorimetric changes of a solution contain-
ing Au-nanoprobes upon an increase to ionic strength of the
medium (dielectric change): presence of the complementary target
prevents Au-nanoprobe aggregation and the solution remains red
(LSPR absorbance peak at ±520 nm); non-complementary/mis-
matched targets do not prevent Au-nanoprobe aggregation, result-
ing in a visible change of color from red to blue characterized by a
concomitant red-shift in the LSPR to 600–650 nm. Towards appli-
cation at point-of-need, where access to fully equipped laboratory is
not feasible, we integrated the colorimetric method onto a paper-
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 43

based mobile platform [18]. This approach combines the use of

inexpensive paper-based platforms and digital image capture and
analysis with a mobile device to screen for the color development
of the AuNPs system. This concept makes use of quantitative colo-
rimetric correlations using mobile cameras to digitalize results
allowing the measurement of color intensity. Using this concept
with a multizone standard 96 or 384 well paper microplates it is
possible to detect multiple samples and targets in a conventional
easy to use plate format [19, 20]. The wax printed well plate is
impregnated with a predetermined concentration of salt capable of
inducing Au-nanoprobe aggregation, thus yielding the required
colorimetric discrimination. Because of the white background of
the paper, the color contrast is greatly improved without the need
for expensive constituents.
Here, we report on the integration of this colorimetric
Au-nanoprobe assay with a paper-platform for characterization of
MTBC members using 30 ng/μl of amplified DNA that is suit-
able to use at point-of-need. By means of a smartphone and a
simple data analysis tool, the developed mobile diagnostics plat-
form allows quantification of the colorimetric changes and trans-
fer of metadata to a centralized off-site laboratory (Fig. 1), taking

Fig. 1 Au-nanoprobe strategy for detection of MTBC members. Schematic representation of the detection with
gold nanoprobes. The colorimetric assay consists of visual comparison of test solutions after salt induced
Au-nanoprobe aggregation on a [MgCl2] impregnated paper plate. After color development a photo of the paper
plate is captured and a RGB and GPS analysis is performed
44 B. Veigas et al.

less than 50 min to complete after DNA amplification with

comparable specificity and sensitivity to the more commonly used
molecular methods.

2 Materials

Prepare all solutions using ultrapure grade water, e.g., Milli-Q

water (Millipore, USA) purified with a resistivity of 18.2 MΩ cm at
25 °C, and use analytical grade reagents.

2.1 Gold Nanoprobe 1. 1 M DL-Dithiothreitol (DTT) solution, molecular biology

Synthesis Components grade. Store at 4 °C until use.
2. Thiol-modified oligonucleotides (5′-thiol-(CH2)6-ssDNA
oligo) harboring a complementary sequence to the target(s) of
interest (STAB Vida, Lda., Portugal)—(see Note 1). For the
specific detection of the rpoB M. tuberculosis gene (GenBank
accession no. L27989) a probe sequence 5′-thiol-GATC-
GCCTCCACGTCC-3′ was used [MTBC probe]. To detect the
resistant strains the probe sequence 5′-thiol-GCCGACAGTC-
GGCGCTTGTG-3′ was used [rpoS531L]. The rpoB531WT
probe sequence 5′-Thiol-GCCGACAGTCGGCGCTTGTC-3′
derived from the normal rpoB gene was used to detect the
normal gene. Resuspend the lyophilized thiol-modified oli-
gonucleotides in 100 μl of 1 M DTT and incubate at room
temperature for 1 h. Add 900 μl of ultrapure sterile water and
mix gently. Store at −20 °C until use.
3. 10 mM phosphate buffer (pH 8): 9.32 mM Na2HPO4, 0.68 mM
NaH2PO4. Sterilize by autoclaving and store at 4 °C until use.
4. AGEI solution: 2 % (w/v) SDS, 10 mM phosphate buffer
(pH 8). Sterilize by filtration (0.22 μm) and store at 4 °C until
use. Warm up to 25 °C before use.
5. AGEII solution: 1.5 M NaCl, 0.01 % (w/v) SDS, 10 mM
phosphate buffer (pH 8). Sterilize by filtration (0.22 μm) and
store at 4 °C until use. Warm up to 25 °C before use.
6. PBS solution: 0.1 M NaCl, 10 mM phosphate buffer (pH 8).
7. Ethyl acetate (CH3COOC2H5, ≥99.5 %).
8. NAP-5 columns (GE Healthcare, Sweden).
9. Ultrasound bath S10H (Elma, Germany).

2.2 Sample Sample decontamination by the N-Acetyl-L-Cysteine–Sodium

Preparation Hydroxide method [21], and DNA extraction with QIAamp DNA
mini kit.
2.2.1 Total DNA
Isolation Components 1. NALC–NaOH solution.
2. Vortex.
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 45

3. Sterile distillated water.

4. Centrifuge.
5. 50 ml Falcon tubes.
6. Safe-lock 1.5 ml microcentrifuge tubes.
7. QIAamp Mini Spin columns.
8. Collection tubes (2 ml).
9. Proteinase K.
10. Buffer AL Buffer AW1 concentrate (dilute with 125 ml
ethanol).
11. Buffer AW2 concentrate (dilute with 160 ml de ethanol).
12. Buffer AE Buffer TE—10 mM Tris–HCl pH 8.0; 1 mM EDTA
pH 8.0.
13. Ethanol (96–100 %) (see Note 2).

2.2.2 DNA Amplification 1. Thermal Cycler DNA Engine (Bio-Rad, USA).

Components 2. 0.1 U/μl DreamTaq DNA polymerase (Fermentas, Canada).
3. 1× DreamTaq Buffer (Fermentas, Canada).
4. 0.2 mM dNTPs mixture (Bioline, UK).
5. 0.2 μM each oligonucleotide primer (StabVida, Portugal)—see
Table 1.
6. UV–Vis Spectrophotometer NanoDrop ND-1000 (NanoDrop
Technologies, USA).
7. GelRed® (Fermentas, Canada).
8. Agarose (Invitrogen).
9. GeneRuler™ DNA Ladder Mix (Fermentas, Canada).
10. TAE Buffer solution: 40 mM Tris-Acetate, 1 mM EDTA
(pH 8).

2.3 Non-cross- 1. 0.3 M MgCl2.

linking Assay 2. 10 mM phosphate buffer (pH 8).
Components
3. 384 well small volume, LoBase Polystyrene microplates, black
(Greiner Bio-One, Germany).
4. Microplate reader Infinite M200 with Absorbance module
(Tecan, Switzerland).

2.4 DNA and Control Thiol-modified ssDNAs, complementary to a specific 17 bp region

Samples of the M. tuberculosis rpoB gene are used to functionalize the gold
nanoparticles, yielding specific Au-nanoprobes. These nanoprobes
are evaluated in terms of specificity by using synthetic DNA oligo-
nucleotides as targets: one fully complementary (wild-type) and
46 B. Veigas et al.

Table 1
Oligonucleotide sequences

Target Sequence (5′–3′)

rpoB Fwd GAG AAT TCG GTC GGC GAG CTG ATC C
rpoB Rev CGA AGC TTG ACC CGC GCG TAC ACC
MTBC probe Thiol-GAT CGC CTC CAC GTC C
rpoB 531WT probe Thiol-GCC GAC AGT CGG CGC TTG TG
Rpo S531L probe Thiol-GCC GAC AGT CGG CGC TTG TC
MTBC target
rpoB 531 target
rpoB S531L target

another harboring the rpoBS531L mutation. These oligonucle-

otides are used to evaluate each probe’s capability to discriminate
the fully complementary sequence (see Note 3). Clinical isolates
susceptible and resistant to at least rifampicin (RIF) were used as
test samples. Additionally, DNA from M. tuberculosis H37Rv
(ATCC27294T) strain and one specimen previously determined as
non-MTCB (M. kansasii) were also included as control samples.
Direct detection of MTBC and mutations in the rpoB gene leading
to RIF resistance was performed by standard detection assays. After
isolation, confirmation of presence of MTBC members was per-
formed using the AccuProbe® (GenProbe Inc., SanDiego, CA)
method according to the manufacturer’s instructions. DNA was
purified via the QIAamp DNA Mini kit (QIAGEN, Hilden,
Germany) according to the manufacturer’s instructions.

2.5 Paper Platform 1. Sheets of a cellulose substrate Whatman No. 1 Chromatography

Components paper (Whatman International Ltd., Floram Park, NJ, USA),
A5 standard format (210 × 148 mm).
2. Solid ink printer (Xerox ColorQube 8570, Xerox Corporation,
CT, USA).
3. Paper microplates design with a standard 96 or 384 plate
format, taken from the “Microplate Dimensions Guide,
Compendium of Greiner Bio-One Microplates” (Greiner Bio-
One GmbH, Frickenhausen, Germany).
4. Hot plate (Heidolph MR Hei-Tec, Schwabach, Germany).
5. 0.3 M MgCl2.
6. 10 mM phosphate buffer (pH 8).
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 47

3 Methods

3.1 Gold Nanoprobe Gold nanoprobe (see Note 4–6) synthesis was performed by deriva-
Synthesis tizing 13 nm gold nanoparticles with thiol-ssDNA oligonucle-
otides. Detailed materials and methods for the preparation and
handling of such particles can be found in Conde et al. [22].
1. Extract one volume (100–500 μl) of thiol-modified oligonu-
cleotide resuspended in 0.1 M DTT with two volumes of ethyl
acetate.
2. Discard the organic phase (i.e., upper phase) after centrifuging
for 5 min at 21,460 × g.
3. Repeat steps 1 and 2 two more times.
4. Purify the remaining aqueous phase through a desalting NAP-5
column, following the manufacturer’s instructions and using
10 mM phosphate buffer (pH 8) as eluent. Typically, after col-
umn equilibration with the eluent, 500 μl of the aqueous phase
is added and let to enter the column. Afterwards, 1000 μl of
purified thiol-modified oligonucleotide can be collected by
adding 1,000 μl of the eluent to the column.
5. Quantify the purified thiol-modified oligonucleotide by UV/
Vis spectroscopy using the extinction coefficient at 260 nm
provided by the oligonucleotide manufacturer.
6. Mix the purified thiol-modified oligonucleotide with the col-
loidal solution in a 1:160 and 1:200 AuNPs/oligo for MTC
and rpo531 probes, respectively.
7. Add AGE I solution to achieve a final concentration of 10 mM
phosphate buffer (pH 8), 0.01 % (w/v) SDS. Typically, 15.1 μl
of AGE I solution is added to a volume of 3 ml of the solution
prepared in step 6.
8. Sonicate the solution for 10 s using an ultrasound bath and
incubate at room temperature for 20 min.
9. Afterwards, sequentially increase the ionic strength of the solu-
tion in 50 mM NaCl increments by adding the respective vol-
ume of AGE II solution up to a final concentration of 10 mM
phosphate buffer (pH 8), 0.3 M NaCl, 0.01 % (w/v)
SDS. Typically, 104.1, 111.6, 119.8, 129.1, 139.4, and 151 μl
of AGE II solution are added sequentially to the 3,015.1 μl of
the solution prepared in step 7. After each increment, sonicate
the solution for 10 s and incubate at room temperature for
20 min before the next increment—see Note 7.
10. Incubate the solution overnight at room temperature.
11. Distribute the functionalized nanoparticles in 1.5 ml micro-
centrifuge tube and centrifuge for 20 min at 21,460 × g.
48 B. Veigas et al.

12. Discard the supernatant. Wash the resulting oily pellet with
1 ml/microcentrifuge tube of 10 mM phosphate buffer (pH 8)
and centrifuge for 20 min at 21,460 × g.
13. Repeat step 12.
14. Discard the supernatant, wash with 1 ml/microcentrifuge tube
of PBS solution and centrifuge for 20 min at 21,460 × g.
15. Discard the supernatant and finally resuspend the pellet in
500 μl/microcentrifuge tube of PBS solution. Gather the
resulting solutions of each microcentrifuge tube in a polypro-
pylene or glass vial with a conical skirted base.
16. Prepare aliquots of 15 or 0.3 nM of Au-nanoprobes, using
PBS as eluent. The initial nanoprobe concentration can be
determined by the Lambert–Beer law using the molar absorp-
tivity of the respective nanoparticles.
17. Store the nanoprobe stock solutions in the dark at 4 °C until
further use.

3.2 Sample Sample decontamination was performed via the N-Acetyl-L-

Preparation Cysteine–Sodium Hydroxide method [21].
3.2.1 Total DNA Isolation 1. Add an equal volume of NALC–NaOH solution to a 50 ml
tube containing the sample.
Sample Decontamination
2. Vortex for 30 s to liquefy.
3. Incubate 15 min at room temperature.
4. Bring the volume to 50 ml with sterile distillated water and
mix by inversion.
5. Centrifuge at 825 × g for 15 min.
6. Discard supernatant and resuspend sediment in ~3 ml of sterile
distillated water.

DNA Extraction 1. Heat water bath or heating block to 95 °C to use in step 6 and
with QIAamp DNA Mini Kit to 56 °C to use in step 9.
2. Centrifuge 1,000 μl of the sample at 15,493 × g for 10 min.
3. Discard supernatant and add 1,000 μl Buffer TE to the pellet.
Mix by pulse-vortexing.
4. Centrifuge at 15,493 × g for 10 min.
5. Discard supernatant and add 200 μl Buffer TE to the pellet.
Mix by pulse-vortexing.
6. Incubate at 95 °C for 20 min.
7. Pipet 20 μl of Proteinase K QIAGEN into the tube.
8. Add 200 μl Buffer AL to the sample. Mix by pulse-vortexing
for 15 s.
9. Incubate at 56 °C for 10 min.
10. Briefly centrifuge the 1.5 ml microcentrifuge tube to remove
drops from the inside of the lid.
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 49

11. Add 200 μl ethanol (96–100 %) and mix by pulse-vortexing

for 15 s.
12. Briefly centrifuge the 1.5 ml microcentrifuge tube to remove
drops from the inside the lid.
13. Carefully apply the mixture to the QIAamp Mini spin column
(in a 2 ml collection tube).

3.2.2 DNA Amplification A specific Polymerase Chain Reaction (PCR)-amplified 395 bp

fragment of the M. tuberculosis RNA polymerase b-subunit gene
(rpoB, GenBank accession no. L27989) suitable for detection of
MTBC members was used as target for the Au-nanoprobe detec-
tion assay [12].
1. In 200 μl polypropylene thermal-resistant reaction tubes pre-
pare at least three amplification reactions (Negative Control,
Negative Sample, and Positive Sample) by mixing PCR ampli-
fication reagents.
2. PCR amplification is performed in a final volume of 50 ml con-
taining 50 mM KCl, 10 mM Tris–HCl (pH 8.3), 2.2 mM
MgCl2, 200 mM of each dNTP, and 1 U of Taq DNA poly-
merase (Amersham Biosciences, GE Healthcare, Europe,
GmbH), 10 pmol of each primer (P1 5′-GAG AAT TCG GTC
GGC GAG CTG ATC C-3′; P2 5′-CGA AGC TTG ACC
CGC GCG TAC ACC-3′). DNA samples isolated from
M. tuberculosis and non-MTBC Mycobacteria cultures are used
as positive MTBC and non-MTBC (non-complementary)
samples, respectively. DNA from an unrelated organism
(Plasmodium berghei 16s-rRNA was used as non-related target
sample, with the final PCR product with a similar length).
Additionally, prepare the Negative Control reaction by replac-
ing the total DNA with an equivalent volume of water.
3. Temperature cycles consist in 35 cycles of 45 s denaturation
at 94 °C, 45 s annealing at 58 °C followed by 45 s extension
at 72 °C.
4. Amplification is observed by 1 % Agarose gel electrophoresis in
a TAE buffer and quantified by UV–Vis spectroscopy (UV–Vis
Spectrophotometer NanoDrop ND-1000—see Note 8).
Additionally, amplification products can be further confirmed
by direct sequencing.

3.3 Colorimetric 1. Prepare six solutions in 200 μl polypropylene thermal-resistant

Non-cross-linking reaction tubes by only mixing the nanoprobe stock solution
Assay with the 10 mM phosphate buffer, according to Table 2. Do
not add MgCl2 at this point!
3.3.1 Gold Nanoprobe
Characterization 2. Incubate the solutions for 10 min at 95 °C.
3. Allow the solutions to cool down at room temperature for
30 min.
50 B. Veigas et al.

Table 2
Setup for characterization of nanoprobe aggregation

Nanoprobe stock 10 mM phosphate

Tube solutiona (μl) buffer (pH 8) (μl) 0.3 M MgCl2b [MgCl2]final (mM)
1 5 25 – 0
2 5 24 1 μl 10
3 5 23 2 μl 20
4 5 22 3 μl 30
5 5 21 4 μl 40
6 5 20 5 μl 50
a
Nanoprobe stock solutions: 15 nM Au-nanoprobe
b
For the paper platform detection add the MgCl2 solution to each paper microplate well in advance, and let the
solution dry

4. Add the 0.3 M MgCl2 according to Table 2, mix well and spin
down the solutions – see Note 9.
5. Incubate the solutions for 15 min at room temperature and
register their absorption spectra (350–800 nm) using a
UV–visible spectrophotometer or microplate reader.
6. Plot Abspeak/Abs600nm vs. [MgCl2]final to determine the mini-
mum salt concentration needed for a complete nanoprobe
aggregation, where Abspeak is the absorbance peak of the
initial dispersed nanoprobe (e.g., typically, 520 nm for
Au-nanoprobes). For Abspeak/Abs600nm ratios below 1, the
nanoprobe is considered to be fully aggregated. If the nano-
probes do not completely aggregate and change color within
the suggested concentrations of MgCl2, use higher concen-
trations of MgCl2. If the nanoprobe changes color and fully
aggregates at a final MgCl2 concentration of 10 mM, then
derivatization of the nanoparticles with the thiol-modified
oligonucleotides may have not been effective. If so, repeat
nanoprobe synthesis – see Note 10.

3.3.2 MTBC DNA 1. In 200 μl polypropylene thermal-resistant reaction tubes pre-

Detection pare at least four assay solutions (Blank, Positive and Negative
Control, and Sample) by mixing the DNA sample (final con-
centration 10–60 ng/μl) with the Au-nanoprobe solution
(final concentration of 2.5 nM), and an appropriate volume of
10 mM phosphate buffer (pH 8) to make up the total volume
(total volume of 30 μl, considering the volume of salt that will
be added in step 5). If more than one sample is to be analyzed,
more assay solutions can be prepared accordingly.
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 51

2. As a Positive Control use the complementary oligonucleotide

MTBComp and as a Negative Control use an oligo with a non-
related sequence (final concentration of 100 fmol/μl)—see
Table 1 for sequences. Additionally, prepare the blank solution
by replacing the total DNA with an equivalent volume of
10 mM phosphate buffer (pH 8).
3. Incubate the solutions for 10 min at 95 °C.
4. Allow the solutions to cool down for 30 min at room
temperature.
5. Add a predetermined volume of the concentrated MgCl2 solu-
tion to attain the final concentration needed for aggregation of
the nanoprobe as previously determined.
6. Mix well, spin down and incubate the solutions for 15 min at
room temperature.
7. Register the absorption spectra (400–800 nm) using a UV–
visible spectrophotometer or a microplate reader.
8. Determine the Abs525nm/Abs600nm ratio. Compare the ratios of
the Blank, Positive and Negative Control with the ratios of the
sample solutions. Typically, the Blank, Negative Control, and
Negative Samples present a ratio <1, while Positive Control
and Positive Samples present a ratio >1 (see Fig. 2). For the
MTBC Au-nanoprobe, this approach provides indication of
presence or absence of MTBC DNA in the sample. Additionally,
for mutation analysis, a set of Au-nanoprobes targeting the
WT and Mut sequences are used. To simplify the mathematical

Fig. 2 Illustration of the characteristic detection assay results and colorimetric call. (a) MTBC detection: UV/
visible analysis of Au-nanoprobe alone—Blank; in presence of a complementary target—MTBC; and in pres-
ence of a non-complementary target—non-MTBC and non-related, registered 30 min after salt addition.
Results presented in Abs526nm/Abs600nm ratios of the UV/visible spectra. (b) Mutation analysis: ratio of WT probe
divided by Mut probe, calculated for rpoB 531 were values >1 indicate a WT genotype (White) and <1 a Mut
genotype (Black). The bars represent the average of three independent measurements and the error bars
indicate standard deviation
52 B. Veigas et al.

analysis, the ratio between the Abs525nm/Abs600nm value for WT

and Mut Au-nanoprobe can be used. Values >1 identify assays
in which the WT probe was more stable than Mut probe, and
thus the sample presented the WT sequence; whereas values <1
indicate the presence of mutation.

3.4 Paper Platform Starting from sheets of a cellulose substrate Whatman No. 1
Chromatography paper (Whatman International Ltd., Floram
3.4.1 Paper Platform
Park, NJ, USA), A5 standard format (210 × 148 mm) sheets were
Preparation
cut. This paper size fits directly into the manual feed tray of a com-
mercial solid ink printer (Xerox ColorQube 8570, Xerox
Corporation, Norwalk, CT, USA) designed to print a wax based
ink, originating hydrophobic barriers. Paper microplates were
designed with a standard 384 plate format in Microsoft Office
Visio1 (Microsoft Corporation, US). All the measurements were
taken from the “Microplate Dimensions Guide, Compendium of
Greiner Bio-One Microplates” (Greiner Bio-One GmbH,
Frickenhausen, Germany). The printed pattern of a 384 microplate
was placed on a hot plate (Heidolph MR Hei-Tec, Schwabach,
Germany) at 140 μC for 2 min, allowing the wax to melt and
spread vertically through the whole thickness of the paper, creating
the desired hydrophobic pattern. Each well was impregnated with
1 ml of a 0.12 M MgCl2 solution (revelation agent) and allowed to
dry at 25 °C for 10 min. The final microplates were stored at
25 μC, and wrapped in aluminum foil until use.

3.4.2 Paper Platform For assaying with the Gold on Paper platform, a total reaction mix-
Detection ture of 5 ml was used with 2.5 nM Au-nanoprobes in 10 mM
phosphate buffer (pH 8), 0.1 M NaCl and target DNA at a final
concentration of 30 mg/ml. After 10 min of denaturation at
95 °C, the mixtures were allowed to stand for 10 min at room
temperature and spotted onto the well on the paper plate. After
45 min at room temperature for color development, the paper
plate was photographed with a mobile device and RGB analysis
was performed.

3.5 Data Analysis The color pattern on the Gold on Paper was captured with an
HTC Desire android smartphone with a 5 megapixel camera
(2,592 × 1,944 pixels) with autofocus. Photos were taken with
artificial white light without flash. Blank test spots with 10 mM
phosphate buffer (pH 8), were used to normalize the data for
light conditions. The digitalized data were then analyzed without
further manipulation with a free RGB analysis application
(ColorPikr, WiseClue) and transmitted via a 3G network to a per-
sonal computer for image processing with ImageJ™. Each assay
was repeated at least three times and on four different paper
micro-well plates (Fig. 3).
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 53

Fig. 3 Representative Gold on paper assay results. Digital photography and RGB
data analysis of Au-nanoprobe mix in the presence of Au-nanoprobe alone—
Blank; in presence of a complementary target—MTBC; and in presence of a
non-complementary target—non-MTBC and non-related. Ratio of aggregation
calculated in the Smartphone (ratio of average intensity of the Red and Blue
channels) for the assay mixtures. Paper microplate assay performed with—
2.5 nM Au-nanoprobe, 10 mM phosphate buffer (pH 8), 0.1 M NaCl, and sample
DNA at a final concentration of 30 μg/ml in a final volume of 5 μl per test. Image
captured after 45 min. The bars represent the average of three independent
measurements and the error bars indicate standard deviation. The horizontal line
represents the threshold of 1 considered for discrimination between positive and
negative. Statistical analysis was performed using Prism 5 graph pad, using one-
way ANOVA with Tukey’s Multiple Comparison test; ***p < 0.001, n = 3

4 Notes

1. Typical thiol-modified oligonucleotide probe sequences range

between 15 and 25 nucleotides, according to specificity
requirements. Check probe sequence specificity using NCBI
nucleotide BLAST tool (http://blast.ncbi.nlm.nih.gov/Blast.
cgi). Increased nanoprobe specificity is achieved at the oligo-
nucleotide’s 3′ end, so consider any single nucleotide poly-
morphism at this position when designing the probe [23].
2. Always measure the volume of ethanol and water separately
and mix them subsequently. Do not prepare ethanol dilutions
based on volumetric flasks, as the total volume of ethanol/
water will drop after mixing, leading to an erroneous dilution
if a volumetric flask mark is used as reference.
3. Thiol-modified ssDNA, complementary to the M. tuberculosis
rpoB gene is used to functionalize the gold nanoparticles and
produce specific Au-nanoprobes. These nanoprobes are
assessed in terms of specificity by means of synthetic DNA
oligonucleotides.
54 B. Veigas et al.

4. Nanoparticles are stable for months when stored in a clean

container (glass or plastic) at room temperature. Do not freeze
the nanoparticles as this may cause aggregation.
5. The Lambert–Beer law states that the absorbance of a homo-
geneous substance becomes linear with its concentration
according to A = ε.l.C, where A is the substance absorbance
typically at its wavelength peak, ε is the molar absorptivity for
the wavelength of A, l is the optical path length and C is the
substance concentration. Care should be taken not to exceed
an absorbance of 1 so as to avoid deviations to the Lambert–
Beer law. In case the measured absorbance exceeds this value,
dilute the sample and consider the dilution factor when calcu-
lating the original stock concentration.
6. Optionally, morphological characterization of the AuNPs can
be performed by Transmission Electron Microscopy (TEM)
and Dynamic Light Scattering (DLS).
7. Different degrees of nanoprobe functionalization (i.e., oligo-
nucleotide density) are attained by varying the final NaCl con-
centration during the aging process (e.g., 0.1, 0.5, 0.7 M).
Be aware that nanoprobe hybridization efficiency may decrease
with increasing density due to steric hindrance. For further
information, consult Doria et al. [23].
8. For PCR product quantification use a UV–Vis Spectro-
photometer NanoDrop ND-1000. Although sample size is not
critical, it is essential that the liquid column be formed so that
the gap between the upper and lower measurement pedestals is
bridged with sample. Use 1 μl of PCR product solution to
ensure reproducibility. It is best to use a precision pipettor
(0–2 μl) with precision tips to assure that sufficient sample
(1–2 μl) is used. Lower precision pipettors (0–10 μl and larger)
are not as good at delivering 1 μl volumes to the measurement
pedestal. Use the application Module (Nucleic Acid—
concentration and purity of nucleic acid) in the default settings
(DNA-50—for dsDNA).
9. If the nanoprobes do not completely aggregate and change
color within the suggested concentrations of MgCl2, use higher
concentrations of MgCl2. If the nanoprobe changes color and
fully aggregates at a final MgCl2 concentration of 10 mM, then
derivatization of the nanoparticles with the thiol-modified
oligonucleotides may have not been effective. If so, repeat
nanoprobe synthesis.
10. To discriminate between two significantly different aggregation
levels, as for example in YES/NO identification of a given tar-
get, the ratio between the peaks at 520 and 600 nm is usually
used for Au-nanoprobes. However, for identifying small dif-
ferences in aggregation levels between two quantities for the
 Mobile Based Gold Nanoprobe TB Diagnostics for Point-of-Need 55

same target, there is a need to decrease the background noise

in the spectra. This background noise may be strongly reduced
and even solved if an integral of the signal is used, i.e., the area
under the curve (AUC).
11. For mutation analysis, a two-probe approach was developed:
(1) one probe complementary to the wild type (WT) sequence;
and (2) one probe complementary to the mutated (Mut)
sequence. Considering that biological targets can only be WT
or Mut, the redundancy of using the two probes simultane-
ously gives robustness to the result.
12. The printing technology used is based on “Solid Ink” from
Xerox (Xerox ColorQube 8670). Solid Ink is formulated from
a nontoxic mixture of hydrophobic carbamates, hydrocarbons
and dyes that melts around 80 °C, being suitable for piezoelec-
tric printing. The printed-paper is melted in a hot plate
(140 °C, 2 min), allowing the wax to move vertically through
the porous paper, creating hydrophobic barriers that define
hydrophilic reaction zones.
13. RGB analysis preformed using the ratio of the red and blue
channel values. The ratio is calculated using the difference of
each channel to the background—taking a white background as
example (RGB—255; 255; 255). Ratio (red/blue) = (255 − red
channel value)/(255 − blue channel value).

References

1. Azzazy HME, Mansour MMH (2009) In vitro resource limited settings. J Microbiol Methods
diagnostic prospects of nanoparticles. Clin 84:155–160
Chim Acta 403:1–8 8. Miller LP, Crawford JT, Shinnick TM (1994)
2. Veigas B, Doria G, Baptista PV (2012) The rpoB gene of Mycobacterium tuberculosis.
Nanodiagnostics for tuberculosis. In: Cardona Antimicrob Agents Chemother 38:805–811
PJ (ed) Understanding tuberculosis—global 9. Musser JM (1995) Antimicrobial agent resis-
experiences and innovative approaches to the tance in Mycobacteria: molecular genetic
diagnosis. InTech. pp 257–276 insights. Clin Microbiol Rev 8:496–514
3. Baptista P, Pereira E, Eaton P et al (2008) 10. Soini H, Musser JM (2001) Molecular diagno-
Gold nanoparticles for the development of sis of Mycobacteria. Clin Chem 47:809–814
clinical diagnosis methods. Anal Bioanal Chem 11. Telenti A, Imboden P, Marchesi F et al (1993)
391:943–950 Detection of rifampicin-resistance mutations
4. Doria G, Conde J, Veigas B et al (2012) Noble in Mycobacterium tuberculosis. Lancet
metal nanoparticles for biosensing applica- 341:647–651
tions. Sensors 12:1657–1687 12. Baptista PV, Koziol-Montewka M, Paluch-
5. WHO (2010) Global tuberculosis control: Oles J et al (2006) Gold-nanoparticle-probe-
WHO Report 2010. WHO Press, Geneva. based assay for rapid and direct detection of
ISBN 978-92-4-156406-9 Mycobacterium tuberculosis DNA in clinical
6. Deun AV, Martin A, Palomino JC (2010) samples. Clin Chem 52:1433–1434
Diagnosis of drug-resistant tuberculosis: reli- 13. Conde J, de la Fuente JM, Baptista PV (2010)
ability and rapidity of detection. Int J Tuberc RNA quantification using gold nanoprobes—
Lung Dis 14:131–140 application to cancer diagnostics. J
7. Abebe G, Paasch F, Apers L et al (2011) Nanobiotechnology 8:5
Tuberculosis drug resistance testing by molec- 14. Costa P, Amaro A, Botelho A et al (2010)
ular methods: opportunities and challenges in Gold nanoprobes assay for identification of
56 B. Veigas et al.

mycobacteria from the Mycobacterium tuber- 19. Carrilho E, Phillips ST, Vella SJ et al (2009) Paper
culosis complex. Clin Microbiol Infect microzone plates. Anal Chem 81:5990–5998
16:1464–1469 20. Carrilho E, Martinez AW, Whitesides GM
15. Doria G, Franco R, Baptista P (2007) (2009) Understanding wax printing: a simple
Nanodiagnostics: fast colorimetric method for micropatterning process for paper-based
single nucleotide polymorphism/mutation microfluidics. Anal Chem 81:7091–7095
detection. IET Nanobiotechnol 1:53–57 21. Kent PT, Kubica GP (1985) Mycobacteriology: a
16. Veigas B, Machado D, Perdigão J et al guide for the level III laboratory. US Dept of
(2010) Au-nanoprobes for detection of Health and Human Services, Public Health
SNPs associated with antibiotic resistance in Service, Centers for Disease Control, Atlanta, GA
Mycobacterium tuberculosis. Nanotechnology 22. Conde J, Doria G, de la Fuente JM, Baptista
21:415101 PV (2012) RNA quantification using noble
17. Silva LB, Veigas B, Doria G et al (2011) metal nanoprobes: simultaneous identification
Portable optoelectronic biosensing platform of several different mRNA targets using color
for identification of mycobacteria from the multiplexing and application to cancer diag-
Mycobacterium tuberculosis complex. Biosens nostics. Methods Mol Biol 906:71–87
Bioelectron 26:2012–2017 23. Doria G, Baumgartner BG, Franco R et al
18. Veigas B, Jacob JM, Costa MN (2012) Gold (2010) Optimizing Au-nanoprobes for specific
on paper-paper platform for Au-nanoprobe TB sequence discrimination. Colloids Surf B
detection. Lab Chip 12:4802–4808 Biointerfaces 77:122–124
 Chapter 4

Immunofluorescence Microtip Sensor for Point-of-Care

Tuberculosis (TB) Diagnosis
Jong-Hoon Kim, Kyong-Hoon Lee, Gerard A. Cangelosi,
and Jae-Hyun Chung

Abstract
A immunofluorescence microtip sensor was developed for specific detection of Mycobacterium cells in
sputum samples by the combination of electric field, streaming flow, and immuno-affinity binding. The
detection limit was 200 CFU/mL in human sputum, which was comparable to PCR but without requiring
bacteriological culture, centrifugation, or nucleic acid amplification. In spite of the complex nature of
physical, chemical, and biological mechanisms, the simple operation of “dipping and withdrawal” of tips
will allow for screening by minimally trained personnel within 30 min. In addition, the minimal power
requirement (5 W) combined with low assay cost is ideal for point-of-care (POC) screening in resource-
limited settings.

Key words Point-of care diagnosis, Tuberculosis, Sputum, Immunosensor, Electric field, Microtip

1 Introduction

Tuberculosis (TB) remains one of the world’s most pressing public

health problems. WHO estimates for 2009 suggest a global preva-
lence of 14 million cases with 1.68 million deaths, including
380,000 people who were HIV positive [1]. TB cases frequently
occur in low-income countries that have poorly resourced public
healthcare sectors [2]. Delayed diagnosis results in serious conse-
quences for both the prognosis of the patient and onward trans-
mission to fuel the epidemic [3, 4]. Thus, the rapid diagnosis and
follow-up treatment are crucial for the effective control of TB.
A number of methods are available for TB diagnosis but
the performance is not satisfactory. Smear microscopy remains the
cornerstone of TB diagnosis in developing countries. The method
depends upon the quality and bacterial load of the sputum speci-
men and the training and motivation of laboratory technicians
[5, 6]. It is suitable for application in developing areas, relatively
inexpensive, and can be accomplished under field conditions.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_4, © Springer Science+Business Media New York 2015

57
58 Jong-Hoon Kim et al.

Even in the absence of electricity, slides can be examined with a

light microscope using reflected sunlight. However, the sensitivity
of smear microscopy is poor, which can detect only 50 % of all the
active cases of TB. The sensitivity can be as low as 20 % in children
and HIV infected people [7, 8]. Smear microscopy also places a
burden on the patient. A person, who is suspicious of TB, has to
visit a clinic at least twice before a diagnosis can be completed, and
then should return again for the results. In terms of transportation
costs and wages, this process can be expensive. Some people never
complete the testing process, or do not return for their results. It has
been estimated that a single-visit diagnostic test offering 100 %
accuracy could save 625,000 lives per year if widely implemented.
A test with only 85 % sensitivity and 97 % specificity could save
392,000 lives, or 22.4 % of the current annual worldwide deaths
attributable to TB [9].
The gold-standard diagnostic method is bacteriological culti-
vation of Mycobacterium tuberculosis (MTB) from pulmonary
specimens, i.e., sputum. Traditional microbial culture-based tests
are the most common methodologies currently used [10, 11].
Culture of MTB on liquid or solid media is a more sensitive method
for TB diagnosis than smear microscopy. Drug resistance can also
be detected. However, the culture requires extensive laboratory
infrastructure and up to 1 month to yield results, which makes it
impractical for use in most cases. Chest radiography is also used in
some countries, but lacks sensitivity and specificity [12]. In one
study, experts missed approximately 25 % of TB cases in a series of
films [13].
To date, many diagnostic methods have been developed for
rapid detection of MTB; polymerase chain reaction (PCR) [14–17],
latex agglutination [18], enzyme-linked immunosorbent assay
(ELISA) [19–21], radiometric detection [22], genprobe amplified
MTB direct test [23], rapid cultivation detection technique [24],
and flow cytometry [25]. These methods are more sensitive and
rapid than the traditional smear microscopy methods; however, the
methods are not capable of offering POC diagnosis of TB. Thus a
rapid and low-cost method for MTB detection is strongly
demanded to effectively control TB infection [26, 27].
A microtip immunofluorescence sensor (Fig. 1) employs the
concentration methods using flow circulation and an electric field
are combined at different scales to concentrate target bacteria in
1 mL sputum samples onto the surface of microscale tips.
Electrokinetic methods have been used to concentrate bioparticles
including DNA and bacteria. Electrophoresis can concentrate tar-
gets by electrostatic attraction. Electroosmosis generates bulk fluid
motion on the electrode surface, which can be used to concentrate
biomolecules [28, 29]. Dielectrophoresis (DEP) describes the
motion of polarizable particles under a nonuniform electric field.
It has been used to selectively concentrate bioparticles, such as
 Microtip Immunosensor 59

Fig. 1 Microtip immunofluorescence sensor. (a) Sensor composed of a linear

motor, vibration motor, sample well, and microchip. (b) Design of microtips
(reprinted with permission [41])

Streaming flow
Electrohydrodynamic flow
microtips
Dielectrophoresis

solution

Fig. 2 Concentration mechanism of target bacteria in 1 mL sputum samples using

streaming flow, EHD flow, and dielectrophoresis (reprinted with permission [41])

bacteria, DNA, and viral particles [30–32]. The novelty of the

microtip sensor is the integration of the electrokinetic methods in
conjunction with streaming flow for efficient concentration of target
bacteria in 1 mL-liquid. Genus-specific antibodies are immobilized
on the microtip surface for specific capturing. Then, the concentrated
bacteria are detected by using an immunofluorescence.
For rapid concentration of target bacteria, streaming flow, elec-
trohydrodynamic (EHD) flow, and dielectrophoresis were com-
bined as illustrated in Fig. 2. The bacteria in the 1 mL sample well
were transported within 1 mm distance from the microtips by
streaming flow. The steady streaming flow, which was the flow of
bulk solution, was induced by the longitudinal vibration of the
well with amplitude of 45 μm at 31.3 Hz. The fluid motion was
analyzed by tracing 19.0 μm-diameter polystyrene spheres in the
60 Jong-Hoon Kim et al.

a
Top view

Microtip

b
0.004
0.002
Microtip
0
−0.01
−0.005
0 0.008
0.006
0.005 0.004
0.01 0.002
Front view 0
−2

0 (Unit : meter)
2
Y axis

8
Side view
10
0 1 2 3 4 5 6 7 8 9 10
X axis

Fig. 3 (a) PIV analysis of two-dimensional flow for each plane, which is generated by longitudinal vibration of
an aluminum well. (b) Three-dimensional flow that is generated by combining two-dimensional flows (reprinted
with permission [41])

steady state of the flow. Particle image velocimetry (PIV) analysis

showed that three-dimensional circulation flow delivered particles
to the center of the well. The flows were measured for two horizon-
tal-, three vertical-, and three side planes, which were integrated
into three-dimensional flow as shown in Fig. 3. Within the 1 mm
distance, the average flow speed was reduced to
100–200 μm/s according to our observation. The transported bac-
terial cells were delivered in the vicinity of microtip surface by EHD
flow. The EHD flow could be generated by both electroosmotic-
and electrothermal flows. The EHD flow speed was 900 μm/s at
maximum at the microtip end and decreased toward the base- and
chip regions of the microtip. The delivered cells were further
attracted to the microtip surface by dielectrophoresis. In our obser-
vation, the dielectrophoretic force attracted the bacteria onto the
microtip surface out of the streamlines of the electrohydrodynamic
flow. The frequency of 5 MHz was chosen to maximize the concen-
tration efficiency based on our previous research [33–35]. The high
frequency was used to selectively capture the target bacterial cells.
Fewer ions and charged particles, potential contamination sources
of microchip surface, could be attracted at 5 MHz. In our experi-
ment, the antibody performance might not be affected by an elec-
tric field due to multiple layers for functionalization of the microchip
surface. The novel concentration mechanism using a microtip could
detect MTB complex in sputum within 25 min. The detection limit
in sputum was 200 CFU/mL with a success rate of 96 %, which was
comparable to PCR. The success rate was the percentage of suc-
cess among the number of the attempts. To compute the success
rate, 30 sputum samples (6 negative controls and 24 distinct samples
 Microtip Immunosensor 61

spiked with MTB at 200 CFU/mL) were tested. On the basis of a

total of 6 negative controls, 23 tests out of 24 positive samples
showed positive results.
The low-cost microtip sensor can be used for a mobile health
(mHealth) system [36, 37]. Although the sensor currently requires
a fluorescent microscope, the recent development of light emitting
diodes (LED) with fluorescent capabilities using low power con-
sumption has led to the development of inexpensive and robust
LED fluorescent microscopes. WHO has recommended using
these as alternative in resource-limited settings after studies have
shown comparable results to standard light and fluorescent micro-
scopes [38, 39]. The battery-powered system can be beneficial for
remote or resource-limited settings.

2 Materials

2.1 Microtip 1. Microtip has a wavy-shape cantilever structure coated with a

Immunosensor System thin gold layer (30 nm). The microchip is composed of five
microtips (Fig. 1b).
2. Linear motor (Compact linear motor, Model #: DRL28PA1G-
03NG, Oriental Motor Co.), Motor driver (Model #:
CRD5107P, Oriental Motor Co.), Controller (Model #: EMP
401, Oriental Motor Co.)
3. Vibration Motor (Model #: 310-003, Precision Microdrives
Ltd.). It is installed under the aluminum well to generate the
streaming flow.
4. The rectangular shape well is made of aluminum for an electric
connection and it had a height of 4.01 mm, a width of
9.525 mm, and a length of 26.19 mm to contain 1 mL sample.
It is vibrated at a frequency of 31.3 Hz with an amplitude of
90 μm.
5. A function generator (Agilent 33220A, Santa Clara, CA) was
used to apply an AC electric field (20 Vpp at 5 MHz).

2.2 Reagents 1. Phosphate buffered saline (1× PBS, pH 7.4, KH2PO4 1.05 mM,
NaCl 155.17 mM, Na2HPO4-7H2O 2.96 mM).
2. Polyclonal IgY antibodies raised against Mycobacterium tuber-
culosis complex.
3. Polyethyleneimine solution (PEI, 1 % (w/v) in DI water).
4. Biotin-conjugated bovine serum albumin (BSA–biotin,
10 mg/mL in 1× PBS).
5. Streptavidin from Streptomyces avidin (streptavidin, 1 mg/mL
in 1× PBS, greater than 13 units/mg protein, streptavidin has
four binding sites for biotins).
62 Jong-Hoon Kim et al.

6. N-acetyl-l-cysteine (NALC, 4 mg/mL in DI water).

7. Sodium dodecyl sulfate (SDS, 4 % (v/v) in DI water).
8. Fluorescein isothiocyanate isomer I (FITC, powder form,
Sigma #F7250).
9. Pierce EZ-Link Sulfo-NHS-LC-LC-Biotin (10 mM (6.7 mg/
mL), Sulfo-NHS ester of biotin with a double 6-aminocaproic
acid spacer, Thermo Scientific #21338).
10. Borate buffer (50 mM borate, pH 8.5).
11. Sodium azide (0.05 % in 1× PBS).

3 Methods

3.1 Microtip Microtips were fabricated in a microfabrication laboratory (Washington

Fabrication Technology Center, University of Washington) as illustrated in
Fig. 4a. The microtips were fabricated from Si wafers having 100 mm
in diameter.
1. Using low pressure chemical vapor deposition (LPCVD), a
1 μm thick silicon nitride layer was grown on Si wafers (100 mm
in diameter).
2. Rectangular areas were patterned on the back side of a Si3N4
layer using conventional UV lithography and etched using
reactive ion etching (RIE) followed by KOH etching.

a
Si3N4 film
Si wafer
Photoresist
UV lithography Remove the Resist
RIE KOH etching

Coat photoresist UV lithography Au deposit

RIE

Fig. 4 Microfabrication process of microtips. (a) Microfabrication process. (b)

Fabricated microtips (reprinted with permission [41])
 Microtip Immunosensor 63

Linear motor

Microtip
holder

Microtip

Aluminum
well

Vibration motor

Fig. 5 Microtip immunofluorescence sensor composed of a linear motor, vibration

motor, sample well, and microchip

3. The wavy-shape of the tip was then patterned on the Si3N4

layer and etched by the RIE process to create free-standing
microtips.
4. Finally, using electron beam evaporator, the tip surface was
coated with gold for electrical conductivity. Figure 4b shows
microtip images by scanning electron microcopy (SEM,
NanoTech User Facility at University of Washington).

3.2 Microtip Microtip concentrator was composed of linear motor, aluminum

Concentrator well block, vibration motor, and microchip holder as illustrated
Assembly in Fig. 5.
1. The microchip holder was made of plastic with a metal probe
which can hold the microchip. It was positioned to immerse the
microchip at the center of the aluminum well. The metal probe
was connected with a function generator to apply electric field.
2. The supporting structure for a linear motor was machined and
assembled with the linear motor.
3. The size of the aluminum well was 4 mm in height, 10 mm in
width, and 26 mm in length, which could hold 1 mL sputum
sample. The vibration motor was installed under the well to
generate the streaming flow.

3.3 BCG and MTB 5 mL cultures of Difco Middlebrook 7H9 broth with ADC enrich-
(H37Ra) Growth ment, 0.2 % glycerol and 0.05 % Tween 20 were inoculated with
and Preparation the Bacillus Calmette-Guérin (BCG) strain of Mycobacterium bovis
(or MTB; H37Ra) and incubated in a 37 °C incubator for between
64 Jong-Hoon Kim et al.

14 and 21 days. These cells were then pelleted with a microcentrifuge

at 2,600 × g for 5 min in 1 mL aliquots followed by removal of
the supernatant and resuspension of the cell pellet in an equal vol-
ume of 1× PBS. Cell concentration of the BCG suspension was
determined by measuring the OD at 580 nm, with an OD of 0.1
corresponding to a concentration of 6.3 × 107 CFU/mL [40].
Note that the BCG is a surrogate marker for TB.

3.4 Biotin Labeling 1. 10 mM (6.7 mg/mL) Pierce EZ-Link Sulfo-NHS-LC-LC-

of Anti-BCG Biotin (#21338) reagent was prepared by adding 2 mg biotin
reagent to 300 μL glass distilled H2O.
2. 153 μL of the biotin solution of this was then added to 1 mL
of a 14 mg/mL antibody solution and allowed to react for 1 h
at room temperature.
3. Unreacted biotin ester was removed with a PD-10 (GE
Healthcare #17-0851-01) size exclusion gel filtration column,
and the final concentration of IgY was measured for protein
concentration at OD 280 nm.
4. The working concentration of IgY-biotin was prepared by
diluting the antibody to 3 mg/mL in PBS.

3.5 Fluorescein Fluorescein isothiocyanate (FITC, Sigma #F7520) was used to

Labeling of Anti-BCG label the IgY antibody with fluorescein.
1. To label the antibody, 100 μL of a 1 mg/mL FITC solution
(dissolved in 50 mM borate buffer, pH 8.5) was added to a
1 mL solution of 10 mg/mL antibody in 50 mM borate buf-
fer, pH 8.5 resulting in a 4.2-fold molar excess of the FITC
reagent.
2. The FITC-IgY solution was then incubated overnight at room
temperature in the dark.
3. Excess and hydrolyzed FITC was then removed by gel filtration
with a Sephadex™ G-25 column (PD-10, GE Healthcare #17-
0851-01, Piscataway, NJ) and returned to PBS + 0.05 % sodium
azide buffer.
4. Antibody concentration was then determined by OD280 using
a UV spectrophotometer (NanoDrop ND-1000, Thermo
Scientific).
5. The working concentration of IgY-Fluorescein was prepared by
diluting the antibody to 2 mg/mL in PBS with 10 mg/mL BSA.

3.6 Microtip The dipping method was utilized for the functionalization.
Functionalization
1. First, the microtip was immersed into 1 % PEI for 1 min and
(Surface Chemistry)
withdrawn at a rate of 100 μm/s. It was coated as an adhesive
layer.
2. The PEI treated tip was then immersed in biotinylated bovine
serum albumin (10 mg/mL in PBS) for 5 min.
 Microtip Immunosensor 65

15 minutes 3 minutes 7 minutes

Sample Bacteria Immunofluorescence
preparation concentration detection

Aluminum well
Microchip
Sputum NALC
Negative
Glass beads Shaker

Shaker 3mm
SDS Positive

Fig. 6 25 min test procedure of a microtip immunofluorescence sensor; (1) sputum was processed to reduce
the viscosity in a simple 15-min procedure; (2) target bacteria in the 1-mL processed sample were specifi-
cally concentrated onto a microtip by a combination of convective flows, an AC field, and affinity binding for
a total of 3 min; and (3) after binding fluorescein-labeled antibodies, the fluorescence signal was processed
for 7 min. In total, 25 min were required for a single assay, from sample cup to a numerical result (reprinted
with permission [41])

3. The biotin-BSA coated tip was cured for 1 min at 175 °C on a

hot plate.
4. The cured tip was dipped in streptavidin (1 mg/mL in PBS)
for 1 min.
5. For the specific binding, the streptavidin-coated tip was coated
with the biotinylated IgY antibodies (3 mg/mL in PBS) for
5 min.

3.7 Sputum Sample Sputum samples from people known not to have TB were pur-
Preparation chased from BioReclamation, Inc. To dissipate the sputum, we used
NALC and SDS with bead-beating as illustrated in Fig. 6.
1. Sputum spiked with MTB (total 600 μL) was mixed with
300 μL NALC (4 mg/mL) and 3 mm glass beads.
2. The sample was vortexed for 5 min at 200 × g .
3. 300 μL SDS (4 %) was added to the mixture. Additional
vortexing for 5 min was conducted for complete liquefaction.
In the process, the final concentration of BCG in the processed
sputum was the half of each initial concentration.
4. Out of total volume of 1.2 mL, 1 mL was transferred to an
aluminum well without glass beads.

3.8 Concentration The test was conducted by three simple steps (Fig. 6):
and Detection
1. Microtips were dipped into the sample with a vibration and an
applied AC field (20 Vpp at 5 MHz) for a 2-min concentration
step. Subsequently, the microtips were withdrawn at a rate of
100 μm/s (see Note 1).
2. The microtips were immersed in a fluorescein conjugated
antibody solution (10 μL, 2 mg/mL) for 5 min.
66 Jong-Hoon Kim et al.

3. It was then rinsed with deionized water (200 μL) for 5 s.

4. The fluorescence intensity was measured by an epi-fluorescence
microscope.
Biotin and streptavidin are well-known protein molecules
which have strong biological interaction each other. Two biotinyl-
ated layers were used to control the orientation of antibodies. To
study the contribution of streaming flow and an electric field to the
concentration of target bacteria, the functionalized microtips were
used in combination with M. bovis BCG cells at a concentration of
107 CFU/mL in 1× PBS. For counting individual cells, BCG cells
were stained with an intercalating dye (SYTO 9® green fluorescent
nucleic acid stain; Molecular Probes L7007, Invitrogen), visual-
ized by an epi-fluorescence microscope (Olympus BX-41, Olympus
America Inc., Melville, NY), and quantified. A 1 mL sample solu-
tion containing stained BCG cells was loaded on the aluminum
well. For concentration, the microtips were immersed in 1 mL
sample for 2 min (see Note 2). For comparison, four combina-
tional cases with and without vibration and electric field (20 Vpp at
5 MHz) were tested. After 2-min concentration, the number of
captured BCG cells was counted on the microtips surface for each
case. When both vibration and electric field were activated, the
number of captured cells was greatest (Fig. 7). When either vibra-
tion or electric field was applied alone, fewer cells were captured.
For preliminary test for sputum samples spiked with BCG, the
concentrations of sputum samples spiked with BCG were
2 × 102 ~ 2 × 106 CFU/mL by tenfold increments. The images of
individual microtips were digitized and compared at each concen-
tration (Fig. 8a). For dose response test using MTB (H37Ra) in
sputum samples, the concentrations of spiked MTB were 2 × 101,
2 × 102, 2 × 103, and 2 × 104 CFU/mL. The experiment was
repeated four times for each concentration (Fig. 8b). In order to

8000
(N=3) SF : Streaming flow
7000
Number of cells (CFU/mm2)

6000

5000

4000

3000

2000

1000

0
E-field, SF SF E-field None

Fig. 7 Effect of each concentration mechanism. The values are number of captured
cells per microtip (reprinted with permission [41])
 Microtip Immunosensor 67

a b
10
87 683 (N=4)
8

Normalized value
6

0
3107 5503 Control 2x102 2x103 2x104
MTB H37Ra in sputum [CFU/mL]

c
10
(N=24)
Normalized value 8

6
11793 15356
4

0
Control 2x102
MTB H37Ra in sputum [CFU/mL]

Fig. 8 Test results of a microtip immunosensor. (a) Digitized images from sputum samples spiked with BCG.
The concentrations of BCG and the average numbers of white pixels are shown at the bottom and top in the
images, respectively. The scale bar is 50 μm. (b) Dose response for MTB spiked in sputum samples.
(c) Reproducibility test at an MTB concentration of 200 CFU/mL in sputum. Compared with 6 negative controls,
23 out of 24 samples show positive signals. The total success rate is 96 % (reprinted with permission [41])

assess the reproducibility of the microtip sensor when challenged

with diverse spiked sputum samples, an additional 30 sputum sam-
ples (6 negative controls and 24 distinct samples spiked with MTB
at 200 CFU/mL) were tested (Fig. 8c).

4 Notes

1. The speed for the withdrawal of microchip should be precisely

controlled. Usually, the higher withdrawal speed could increase
the nonspecific binding.
2. The longer concentration time could damage the function of
antibodies on the surface. The maximum concentration time
could vary based on the types of antibody.
68 Jong-Hoon Kim et al.

Acknowledgements

We acknowledge the support of CDC SBIR II Contract (200-

2009-31946) and the grant from the Catalysis Foundation for
Health. In addition, J.C., J.K., and W.Y. acknowledge the support of
NSF Career (ECCS-0846454). We appreciate valuable discussion
with Hyun-Boo Lee about the concentration mechanism.

References

1. WHO (2010) Global tuberculosis control: 12. Elliott AM, Luo N, Tembo G, Halwiindi B,
WHO Report 2010. WHO, Geneva Steenbergen G, Machiels L et al (1990) Impact
2. WHO (2009) Global tuberculosis control: sur- of HIV on tuberculosis in Zambia—a cross-
veillance, planning, financing. WHO, Geneva sectional study. Br Med J 301:412–415
3. Lienhardt C, Rowley J, Manneh K, Lahai G, 13. Garland LH (1959) Studies on the accuracy of
Needham D, Milligan P et al (2001) Factors diagnostic procedures. Am J Roentgenol
affecting time delay to treatment in a tubercu- Radium Ther Nucl Med 82:25–38
losis control programme in a sub-Saharan 14. Rodriguez-Lazaro D, D’Agostino M,
African country: the experience of The Gambia. Herrewegh A, Pla M, Cook N, Ikonomopoulos
Int J Tuberc Lung Dis 5:233–239 J (2005) Real-time PCR-based methods for
4. Golub JE, Bur S, Cronin WA, Gange S, Baruch detection of Mycobacterium avium Subsp
N, Comstock GW et al (2006) Delayed tuber- paratuberculosis in water and milk. Int J Food
culosis diagnosis and tuberculosis transmission. Microbiol 101:93–104
Int J Tuberc Lung Dis 10:24–30 15. Thomson LM, Traore H, Yesilkaya H, Doig C,
5. Aber VR, Allen BW, Mitchison DA, Ayuma P, Steingrimsdottir H, Garcia L et al (2005) An
Edwards EA, Keyes AB (1980) Quality-control extremely rapid and simple DNA-release
in tuberculosis bacteriology. 1. Laboratory method for detection of M. tuberculosis from
studies on isolated positive cultures and the clinical specimens. J Microbiol Methods 63:
efficiency of direct smear examination. Tubercle 95–98
61:123–133 16. Choi YJ, Hu Y, Mahmood A (1996) Clinical
6. Boyd JC, Marr JJ (1975) Decreasing reliability significance of a polymerase chain reaction
of acid-fast smear techniques for detection of assay for the detection of Mycobacterium
tuberculosis. Ann Intern Med 82:489–492 tuberculosis. Am J Clin Pathol 105:200–204
17. Durmaz R, Aydin A, Durmaz B, Aydin NE,
7. Steingart KR, Ramsay A, Pai M (2007)
Akbasak BS, Gunal S (1997) Sensitivity of
Optimizing sputum smear microscopy for the
two-stage PCR amplification for detection of
diagnosis of pulmonary tuberculosis. Expert
Mycobacterium tuberculosis in paraffin-
Rev Anti Infect Ther 5:327–331
embedded tissues. J Microbiol Methods 29:
8. Perkins MD, Small PM (2006) Partnering for 69–75
better microbial diagnostics. Nat Biotechnol 18. Krambovitis E, Lock PE, Mcillmurray MB,
24:919–921 Hendrickse W, Holzel H (1984) Rapid diagno-
9. Keeler E, Perkins MD, Small P, Hanson C, sis of tuberculous meningitis by latex particle
Reed S, Cunningham J et al (2006) Reducing agglutination. Lancet 2:1229–1231
the global burden of tuberculosis: the contri- 19. Tamminen M, Joutsjoki T, Sjoblom M, Joutsen
bution of improved diagnostics. Nature M, Palva A, Ryhanen EL et al (2004) Screening
444(Suppl 1):49–57 of lactic acid bacteria from fermented vegeta-
10. Lazcka O, Del Campo FJ, Munoz FX (2007) bles by carbohydrate profiling and PCR-
Pathogen detection: a perspective of traditional ELISA. Lett Appl Microbiol 39:439–444
methods and biosensors. Biosens Bioelectron 20. Nassau E, Parsons ER, Johnson GD (1976)
22:1205–1217 The detection of antibodies to Mycobacterium
11. Torres-Chavolla E, Alocilja EC (2009) tuberculosis by microplate enzyme-linked
Aptasensors for detection of microbial and viral immunosorbent assay (ELISA). Tubercle 57:
pathogens. Biosens Bioelectron 24:3175–3182 67–70
 Microtip Immunosensor 69

21. Delacourt C, Gobin J, Gaillard JL, Deblic J, 32. Huang Y, Joo S, Duhon M, Heller M, Wallace
Veron M, Scheinmann P (1993) Value of B, Xu X (2002) Dielectrophoretic cell separation
ELISA using antigen-60 for the diagnosis of and gene expression profiling on microelectronic
tuberculosis in children. Chest 104:393–398 chip arrays. Anal Chem 74:3362–3371
22. Middlebrook G, Reggiardo Z, Tigertt WD 33. Yeo WH, Chung JH, Liu YL, Lee KH (2009)
(1977) Automatable radiometric detection of Size-specific concentration of DNA to a nano-
growth of Mycobacterium-tuberculosis in selec- structured tip using dielectrophoresis and capil-
tive media. Am Rev Respir Dis 115:1066–1069 lary action. J Phys Chem B 113:10849–10858
23. Gamboa F, Manterola JM, Lonca J, Matas L, 34. Yeo WH, Liu S, Chung JH, Liu YL, Lee KH
Vinado B, Gimenez M et al (1997) Detection (2009) Rapid detection of Mycobacterium
and identification of mycobacteria by amplifica- tuberculosis cells by using microtip-based
tion of RNA and DNA in pretreated blood and immunoassay. Anal Bioanal Chem 393:
bone marrow aspirates by a simple lysis method. 1593–1600
J Clin Microbiol 35:2124–2128 35. Yeo WH, Chou FL, Fotouhi G, Oh K, Stevens
24. Cambau E, Wichlacz C, Truffot-Pernot C, BT, Tseng HY et al (2010) Size-selective
Jarlier V (1999) Evaluation of the new MB immunofluorescence of Mycobacterium tuber-
Redox system for detection of growth of myco- culosis cells by capillary- and viscous forces.
bacteria. J Clin Microbiol 37:2013–2015 Lab Chip 10:3178–3181
25. Qin DL, He XX, Wang KM, Tan WH (2008) 36. Wei Q, Qi H, Luo W, Tseng D, Ki SJ, Wan Z
Using fluorescent nanoparticles and SYBR et al (2013) Fluorescent imaging of single
Green I based two-color flow cytometry to nanoparticles and viruses on a smart phone.
determine Mycobacterium tuberculosis avoid- ACS Nano 7:9147–9155
ing false positives. Biosens Bioelectron 24: 37. Miller AR, Davis GL, Oden ZM, Razavi MR,
626–631 Fateh A, Ghazanfari M et al (2010) Portable,
26. Griffiths D, Hall G (1993) Biosensors—what battery-operated, low-cost, bright field and
real progress is being made. Trends Biotechnol fluorescence microscope. PLoS One 5:4
11:122–130 38. Minion J, Pai M, Ramsay A, Menzies D,
27. Owen VM (1994) Market requirements for Greenaway C (2011) Comparison of LED and
advanced biosensors in healthcare. Biosens conventional fluorescence microscopy for
Bioelectron 9:xxix–xxxv detection of acid fast bacilli in a low-incidence
28. Bown MR, Meinhart CD (2006) AC electro- setting. PLoS One 6:e22495
osmotic flow in a DNA concentrator. 39. Trusov A, Bumgarner R, Valijev R, Chestnova
Microfluid Nanofluidics 2:513–523 R, Talevski S, Vragoterova C et al (2009)
29. Wong PK, Chen CY, Wang TH, Ho CM Comparison of Lumin LED fluorescent attach-
(2004) Electrokinetic bioprocessor for concen- ment, fluorescent microscopy and Ziehl-
trating cells and molecules. Anal Chem Neelsen for AFB diagnosis. Int J Tuberc Lung
76:6908–6914 Dis 13:836–841
30. Asbury CL, van den Engh G (1998) Trapping 40. Wayne LG (1976) Dynamics of submerged
of DNA in nonuniform oscillating electric growth of Mycobacterium tuberculosis under
fields. Biophys J 74:1024–1030 aerobic and microaerophilic conditions. Am
31. Gimsa J (2001) A comprehensive approach to Rev Respir Dis 114:807–811
electro-orientation, electrodeformation, dielec- 41. Kim J-H, Yeo W-H, Shu Z, Soelberg SD,
trophoresis, and electrorotation of ellipsoidal Inoue S, Kalyanasundaram D et al (2012)
particles and biological cells. Bioelectrochemistry Immunosensor towards low-cost, rapid diagnosis
54:23–31 of tuberculosis. Lab Chip 12:1437–1440
 Chapter 5

Improving Lateral-Flow Immunoassay (LFIA) Diagnostics

via Biomarker Enrichment for mHealth
James J. Lai and Patrick S. Stayton

Abstract
Optical detection technologies based on mobile devices can be utilized to enable many mHealth applica-
tions, including a reader for lateral-flow immunoassay (LFIA). However, an intrinsic challenge associated
with LFIA for clinical diagnostics is the limitation in sensitivity. Therefore, rapid and simple specimen
processing strategies can directly enable more sensitive LFIA by purifying and concentrating biomarkers.
Here, a binary reagent system is presented for concentrating analytes from a larger volume specimen to
improve the malaria LFIA’s limit of detection (LOD). The biomarker enrichment process utilizes
temperature-­responsive gold–streptavidin conjugates, biotinylated antibodies, and temperature-responsive
magnetic nanoparticles. The temperature-responsive gold colloids were synthesized by modifying the
citrate-stabilized gold colloids with a diblock copolymer, containing a thermally responsive poly(N-­
isopropylacrylamide) (pNIPAAm) segment and a gold-binding block composed of NIPAAm-co-N,N-­
dimethylaminoethylacrylamide. The gold–streptavidin conjugates were synthesized by conjugating
temperature-responsive gold colloids with streptavidin via covalent linkages using carbodiimide chemistry
chemistry. The gold conjugates formed half-sandwiches, gold labeled biomarker, by complexing with bio-
tinylated antibodies that were bound to Plasmodium falciparum histidine-rich protein 2 (PfHRP2), a
malaria antigen. When a thermal stimulus was applied in conjunction with a magnetic field, the half-
sandwiches and temperature-­responsive magnetic nanoparticles that were both decorated with pNIPAAm
formed large aggregates that were efficiently magnetically separated from human plasma. The binary
reagent system was applied to a large volume (500 μL) specimen for concentrating biomarker 50-fold into
a small volume and applied directly to an off-the-shelf malaria LFIA to improve the signal-to-noise ratio.

Key words Magnetic, Gold, Nanoparticles, Diagnostics, Lateral-flow, Stimuli-responsive, Polymer

1  Introduction

Clinical laboratory assays that can detect disease specific protein

biomarkers in human bodily fluids have been providing both diag-
nostic and prognostic value [1]. In addition to the laboratory-­based
clinical assays, rapid tests such as OraQuick are now also available as
the over-the-counter products for home-use d ­ iagnostics [2]. Point-
of-care diagnostics can improve the overall health care by providing
better medical access to meet the needs of underserved populations

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_5, © Springer Science+Business Media New York 2015

71
72 James J. Lai and Patrick S. Stayton

in remote locations where current techniques designed for conventional

medical settings are not viable [3–6]. Mobile devices such as cell
phones have the potential to provide clinical utilities that can address
the need of medical diagnostics in telemedicine, in resource poor
areas, and in global health settings—mHealth [6, 7]. For example,
optical detection technologies based on mobile devices have been
developed for mHealth [6], including a reader for lateral-flow
immunoassays (LFIAs) [8].
LFIA, a current dominant point-of-care diagnostic test format,
is utilized to expand clinical diagnostic tests from centralized hos-
pital laboratories to more distributed settings [9–16]. LFIA enables
non-instrumented assays by utilizing gold colloids as the detection
reagent [17] and capillary wicking via porous nitrocellulose strips
for fluidic transport. The current LFIA exhibits advantages such as
rapid, inexpensive, portable, and easy to use. Portable readers can
enable quantitative LFIA. However, an intrinsic challenge associ-
ated with LFIA for point-of-care diagnostics is the limitation in
sensitivity related to the small specimen volume that limits the bio-
marker repertoire to those at relatively higher blood and plasma
concentrations [18]. Therefore, there is a critical need for novel
specimen processing technologies that are rapid and amenable to
low-resource settings for purifying and concentrating biomarkers
in a form that could then be applied directly to the existing LFIA.
Magnetic affinity reagents are widely utilized for biomolecule
separation and purification due to their simplicity [19–23]. We
have previously developed a stimuli-responsive magnetic nanopar-
ticles that reversibly aggregate via temperature or pH changes to
overcome challenges associated with the traditional magnetic
microbeads and nanoparticles [24, 25]. Stimuli-responsive poly-
mers such as poly(N-isopropylacrylamide) (pNIPAAm) with revers-
ible phase-transition properties have been shown to enhance the
capture/separation capabilities of immunoassays [26]. Nanoparticles
or proteins, grafted with pNIPAAm, form aggregates together via
the polymer–polymer interaction [27–32]. Our stimuli-responsive
magnetic nanoparticles enabled both efficient analyte binding using
highly diffusive small (ca. 10 nm) particles, and rapid magnetopho-
retic separation by thermally aggregating nanoparticle–analyte
complexes [24, 25]. pNIPAAm magnetic nanoparticles can facili-
tate rapid magnetic separation of non-­magnetic pNIPAAm conju-
gates via a coaggregation mechanism [27, 33].
We recently presented a binary reagent system for concentrat-
ing analytes from a larger volume specimen to improve the off-the-­
shelf malaria LFIA’s limit of detection (LOD) via the coaggregation
of two stimuli-responsive reagents, gold colloid and magnetic
nanoparticle [33]. Gold colloids decorated with antibodies have
been widely utilized as optical labels in immunoassay and biodetec-
tion because gold colloids exhibit large extinction coefficients
(>109 M−1 cm−1) in the visible spectrum and enhanced electric
 Biomarker Enrichment 73

near-fields at the particle surface [34–38]. For example, half-­

sandwiches, the active detection species for disease diagnosis in con-
ventional LFIA, are formed via the complexation between target
analytes and gold–antibody conjugates. LFIA utilizes the capture
antibodies immobilized at the test line of membrane surface for rec-
ognizing half-sandwiches, which results in visible color. The gold–
antibody conjugates modified with pNIPAAm can complex with
target analytes and coaggregate with pNIPAAm magnetic nanopar-
ticles via heating to achieve rapid magnetic separation [27, 33].
Therefore, the binary reagent system can enable rapid magnetic
enrichment of half-sandwiches, gold-labeled antigens, that can
accommodate binding to the capture antibody at the test line of the
LFIA (Fig. 1) [33]. The approach is attractive for point-of-­care
diagnostic in low-resource settings because the temperature-­
responsive aggregation can be enabled using a portable battery-
powered heater and simple tube rack with magnets. Here we present
the volumetric enrichment method via the recently developed
binary reagent system to achieve higher analytical s­ensitivity in an
off-the-shelf multiplexed LFIA using PfHRP, a malaria antigen, as
the model assay.

Fig. 1 Depiction of the magnetic enrichment lateral-flow immunoassay. A biotinylated antibody is added to a
plasma sample containing the target biomarker(s). An equal volume of buffer containing temperature-­
responsive gold–streptavidin conjugates, pNIPAAm magnetic nanoparticles, and free pNIPAAm polymer is
added. Upon heating, the mixed gold–magnetic nanoparticle aggregates are separated by a magnet. After the
supernatant is discarded, the captured aggregates are redissolved into a smaller volume of cool buffer, result-
ing in particle disaggregation and 50-fold enrichment. The enriched mixture is then applied directly to an
immunochromatographic assay membrane with functionalized test and control line antibody regions.
Reproduced from ACS Nano 2012 with permission from ACS Publications
74 James J. Lai and Patrick S. Stayton

2  Materials

1. N-isopropylacrylamide, NIPAAm (415324, Sigma-Aldrich).

2. 4-Cyano-4-[(dodecylsulfanylthiocarbonyl)sulfanyl] pentanoic
acid, DCT (723274, Sigma-Aldrich).
3. 2,2′-Azobis(2-methylpropionitrile), AIBN (755745,
Sigma-Aldrich).
4. 1,4-Dioxane (360481, Sigma-Aldrich).
5. Tetrahydrofuran, THF (TX0282, EMD Millipore).
6. Pentane (PX0167, EMD Millipore).
7. MWCO 6–8 kDa dialysis membrane (132660, Spectrum
Laboratories, Inc.).
8. N,N-dimethylaminoethylacrylamide, DMAEAm (Catalog
No. 9573, Monomer-Polymer and Dajac Labs).
9. Methanol (34860, Sigma-Aldrich).
10. PD-10 desalting column (Catalog No. 17-0851-01, GE

Healthcare Life Sciences).
11. HAuCl4·3H2O (520918, Sigma-Aldrich).
12. Sodium citrate (W302600, Sigma-Aldrich).
13. 0.1 N NaOH (SS276, Fisher Scientific).
14. Amicon stirred cells (5124, Millipore).
15.
Ultracel regenerated cellulose membranes (14422AM,
Millipore).
16. 0.1 M MES buffered saline (28390, Thermo Scientific).
17. N-hydroxysulfosuccinimide, sulfo-NHS (24510, Thermo
Scientific).
18. 1-Ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochlo-
ride, EDC (22980, Thermo Scientific).
19. Zeba spin desalting column (89891, Thermo Scientific).
20. PBS (P4417, Sigma-Aldrich).
21. Streptavidin (21135, Thermo Scientific).
22. 2-(Dodecylthiocarbonothioylthio)-2-methylpropionic acid,
DMP (723010, Sigma-Aldrich).
23. Tetraethylene glycol dimethyl ether (172405, Sigma-Aldrich).
24. Iron(0) pentacarbonyl (481718, Sigma-Aldrich).
25. Millex-FG syringe filter (SLFG04NL, Millipore).
26. MWCO 100 kDa dialysis membrane (131420, Spectrum

Laboratories, Inc.).
27. Monoclonal mouse anti-PfHRP2 IgG (MPFG-55A, Immunology
Consultants Laboratory).
 Biomarker Enrichment 75

28. NHS-chromogenic-biotin (21325, Thermo Scientific).

29. Malaria home test kit (Sanitoets, Sallamander Concepts).
30. Adhesive-coated mylar (Fralock Inc.).
31. Anhydrous dimethylformamide (227056, Sigma-Aldrich).
32. Cellulose/synthetic blend fiber (grade 8301, Ahlstrom).
33. Cellulose absorbent pad (CFSP002000, Millipore).
34. DynaMag-2 Magnet (12321D, Life Technologies).

3  Methods

The biomarker enrichment process (Fig. 1) utilizes temperature-­

responsive gold–streptavidin conjugates, biotinylated antibodies,
and temperature-responsive magnetic nanoparticles. In addition to
the reagents, the assay also requires the modification of the off-the-­
shelf lateral-flow test. The methods for the reagent synthesis and
device modification are described here.

3.1  Synthesis The conjugates (Fig. 2) were synthesized by decorating temperature-­

of Temperature-­ responsive gold colloids with streptavidin via covalent linkages, car-
Responsive Gold– bodiimide chemistry. The temperature-responsive gold colloids
Streptavidin were synthesized by modifying the citrate-stabilized gold colloids
Conjugates with a diblock copolymer, poly(N-isopropylacrylamide-­block-N,N-
dimethylaminoethylacrylamide) (p(NIPAAm-b-­DMAEAm))
[27, 33, 39]. The polymer synthesis was carried out using reversible
addition-fragmentation chain transfer (RAFT) polymerization with
a macro chain transfer agent (mCTA), poly(N-­isopropylacrylamide)
(pNIPAAm) [27, 33, 40].

Fig. 2 Targeted bioconjugate gold colloid design. Gold colloids were modified with a diblock copolymer pro-
duced via two-step RAFT polymerization. The polymer’s semi-telechelic carboxyl group was conjugated to
lysine groups on streptavidin, enabling linkage to a biotinylated antibody. The conjugate design allowed facile
multiplexed detection by simply mixing different biotinylated antibodies with the sample before addition of the
universal streptavidin-gold detection reagent. Reproduced from ACS Nano 2012 with permission from ACS
Publications
76 James J. Lai and Patrick S. Stayton

3.1.1  Synthesis of mCTA, 1. Polymer targeted ca. 200 degree of polymerization (DP) was
pNIPAAm prepared by mixing in a round-bottom flask 3.11 g
(27.5 mmol) of recrystallized NIPAAm (see Note 1), 85 mg
(0.134 mmol) of DCT, and 3.4 mg (13.4 μmol) of AIBN in
6 g 1,4-dioxane.
2. The flask was sealed with a rubber septum and the solution
was raised to ca. 30–40 °C using a warm tap water bath, while
purging with N2 for ≥20 min prior to the polymerization.
Then, the polymerization was carried out by heating the solu-
tion in the sealed flask via an oil bath at 60 °C for 18 h.
3. The resulting polymers were diluted in 15 mL THF, and then
isolated by precipitation into 200 mL pentane. The precipitate
was filtered, dried in vacuo, dialyzed against DI water at 4 °C
for 72 h with MWCO 6–8 kDa dialysis membrane, and
lyophilized.
4. The polymer molecular weight was characterized using size
exclusion chromatography with multiangle light scattering
detection using dn/dc = 0.076.

3.1.2  Synthesis 1. Polymer targeted ca. 40 degree of polymerization was pre-

of Diblock Copolymer, pared by mixing in a round-bottom flask 0.2258 g (1.59 mmol)
p(NIPAAm-b-­DMAEAm) of DMAEAm, 0.18 g (1.59 mmol) of recrystallized NIPAAm,
1.32 g (83 μmol) of the mCTA, and 1.4 mg (8.3 μmol) of
AIBN in 8 mL of methanol.
2. The flask was sealed with a rubber septum and the polymeriza-
tion solution was purged with N2 for ≥20 min prior to the
polymerization. Then, the polymerization was carried out by
heating the solution in the sealed flask via an oil bath at 60 °C
for 18 h.
3. After the polymerization the solution was cooled down to
room temperature, the methanol was removed by rotary evap-
oration, and the product was dissolved in 5 mL THF and pre-
cipitated thrice into 100 mL pentane. The precipitate was
dried in vacuo, dissolved in DI water, purified by PD-10
desalting column, and lyophilized.
4. The polymer molecular weight was characterized using size
exclusion chromatography with multiangle light scattering
detection using dn/dc = 0.071. The polymer composition was
characterized using 1H NMR (see Note 2).

3.1.3  Synthesis 1. Citrate-stabilized colloidal gold was prepared according to the

of Temperature-­ literature [41]. The reaction solution was prepared by bring-
Responsive Gold Colloids ing 150 mL of 0.1 mg/mL HAuCl4·3H2O in a round-bottom
flask (see Note 3) to a boil, and then added 1.76 mL of 10 mg/
mL sodium citrate.
2. The solution was boiled under reflux with a condenser for
30 min, and then cooled to room temperature.
 Biomarker Enrichment 77

3. The solution was titrated to pH 8.2 using 0.1 N NaOH, and then
1.2 mL of 10 mg/mL diblock copolymer solution was added.
4. After the polymer addition, the flask was purged with N2 for
45 min and stirred at room temperature (ca. 22 °C) for 24 h
in darkness.
5. The solution was added with 1 g of NaCl and stirred for an
additional 24 h at room temperature.
6. The particles were then concentrated using Amicon stirred
cells with 100 kDa NMWL Ultracel regenerated cellulose
membranes under 35–40 psi of N2.
7. The temperature-responsive gold colloids on the membrane
were collected by washing with 2 mL of 0.1 M MES buffered
saline at pH 5.0.

3.1.4  Streptavidin 1. The temperature-responsive gold colloid solution (2 mL, ca.

Conjugation 70 nM in MES buffered saline at pH 5.0, see Note 4) was
to Temperature-­ added to the flask, containing 13.1 mg of sulfo-NHS and
Responsive Gold Colloids 14.8 mg of EDC in dry form (see Note 5).
2. After 40 min of orbital shaking at room temperature the solu-
tion was purified and buffer exchanged using a Zeba spin
desalting column, primed with PBS, and then immediately
mixed with 20 mg of streptavidin.
3. The conjugation was allowed to run 2 h at room temperature
with orbital shaking.
4. The resulting particles were purified and concentrated three
times using Amicon stirred cells with 100 kDa NMWL Ultracel
regenerated cellulose membranes under 35–40 psi of N2.
5. The resulting temperature-responsive gold colloid–streptavi-
din conjugates were rinsed off the membrane with PBS and
were stored (~80 nM) at 4 °C under N2 for up to 3 months.

3.2  Synthesis The magnetic nanoparticle synthesis will utilize micelles comprised
of Temperature-­ of DMP-pNIPAAm, pNIPAAm with a dodecyl (hydrophobic)
Responsive Magnetic chain end, for dimensional confinement as previously described
Nanoparticles [24]. The polymer synthesis was carried out using RAFT polymer-
ization with DMP CTAs.

3.2.1  Synthesis 1. Polymer targeted ca. 45 degree of polymerization was pre-

of DMP-pNIPAAm pared by mixing in a round-bottom flask 2 g (17.7 mmol) of
NIPAAm, 143.3 mg (0.4 mmol) of DMP, and 6.66 mg
(0.04 mmol) of AIBN, in 4 g of 1,4-dioxane.
2. The flask was sealed with a rubber septum and the polymeriza-
tion solution was purged with N2 for ≥20 min at ca. 30 °C
prior to the polymerization. Then, the polymerization was
carried out by heating the solution in the sealed flask via an oil
bath at 60 °C for 12 h.
78 James J. Lai and Patrick S. Stayton

3. The resulting polymers were diluted in 15 mL THF, and then

isolated by precipitation into 200 mL pentane. The precipitate
was filtered, dried in vacuo, dialyzed against DI water at 4 °C
for 72 h with MWCO 6–8 kDa dialysis membrane, and
lyophilized.
4. The polymer molecular weight was characterized using size
exclusion chromatography with multiangle light scattering
detection using dn/dc = 0.076.

3.2.2  Synthesis 1. The magnetic nanoparticle synthesis solution was prepared by

of Magnetic Nanoparticles mixing in a round-bottom flask 0.9 g (180 μmol) of 5 kDa
DMP-pNiPAAm in 50 mL of tetraethylene glycol dimethyl
ether.
2. The solution temperature was raised to 100 °C with constant
stirring, and then 200 μL of iron(0) pentacarbonyl (see Note 6)
was added into the solution. Then, the solution temperature
was further increased and maintained at 190 °C for 6 h.
3. Once the solution was cooled down to room temperature, the
resulting magnetic nanoparticles were isolated by precipitation
into pentane, dried in vacuo, dialyzed against DI water at 4 °C
for 72 h with MWCO 100 kDa dialysis membrane, and lyoph-
ilized. The dried product was then dissolved in DI water at
50 mg/mL and stored at 4 °C.

3.3  Synthesis Biotinylation of the IgG antibodies was performed using an NHS-­
of Biotinylated activated biotin containing a chromophore linker.
Anti-PfHRP2 IgG
1. The NHS-chromogenic-biotin dissolved at 10 mg/mL in anhy-
drous DMF was added in 7.5 M excess (8.1 μL) to 2 mL of the
IgG antibodies at 1 mg/mL in PBS, at room temperature.
2. After 3 h incubation at room temperature, the unreacted
NHS-chromogenic-­ biotin was removed using a Zeba spin
desalting column.
3. The degree of biotinylation was estimated by measuring the
ratio of the biotin-chromophore extinction (ε = 29,000 M−1 cm−1
at 354 nm) to the IgG extinction (ε = 210,000  M−1 cm−1 at
280 nm) (see Note 7).

3.4  Lateral-Flow 1. To remove the flow strip, plastic cassette of the lateral-flow test
Device Modification was opened using a utility knife (Fig. 3a).
2. The dried gold conjugate pad (pink) and absorbent pad
(Fig. 3b) were removed and discarded, and then the flow strip
was removed from the plastic cassette and remounted onto a
strip of adhesive-coated mylar (4 mm wide × 38 mm long)
using a pair of tweezers (Fig. 3b).
3. A porous rectangular sheet of cellulose fiber was placed on the
assay membrane upstream of the capture lines (Fig. 3c) to
 Biomarker Enrichment 79

Fig. 3 Photographs for lateral-flow device modification. (a) A lateral-flow device

with the plastic cassette before the modification. (b) An opened lateral-flow
device that shows the flow strip with dry gold conjugate pad (pink) and absor-
bent pad at the bottom of the strip. (c) The flow strip that is mounted on the
adhesive mylar layer with cellulose fiber to serve as a filter and liquid reservoir
for the nanoparticle mixtures and a cellulose absorbent pad to serve as a rinse
buffer reservoir (Color figure online)

serve as a filter and liquid reservoir for the nanoparticle mix-

tures, while the sample liquid was imbibed by the nitrocellu-
lose membrane. A cellulose absorbent pad (4 × 10 mm) was
placed just upstream of the nanoparticle cellulose fiber to serve
as a rinse buffer reservoir (Fig. 3c).

3.5  Magnetic PfHRP2 enrichment utilized the recently developed binary reagent
Enrichment of PfHRP2 system, including temperature-responsive magnetic nanoparticles
for Lateral-Flow and temperature-responsive gold–streptavidin conjugate, in con-
Immunoassays junction with the biotinylated antibodies. While the enrichment
protocol described here utilized a lab-based incubator for heating,
the process is suitable is suitable for low-resource setting environ-
ment by using battery-powered heater.
1. Biotinylated anti-PfHPR2 IgG (50 μL, 10 nM) was added
into the tube containing 250 μL human plasma specimen,
and then followed sequentially by 50 μL PBS, 50 μL
temperature-­responsive gold colloid–streptavidin conjugates
(20 nM), 50 μL temperature-responsive magnetic nanoparti-
cles (10 mg/mL), and 50 μL mCTA (15 mg/mL).
80 James J. Lai and Patrick S. Stayton

Fig. 4 Photographs for biomarker enrichment. (a) A tube, containing biotinyl-

ated antibody, temperature-responsive gold colloid–streptavidin conjugates,
temperature-responsive magnetic nanoparticles, and free pNIPAAm, was
­incubated in an aluminum heat block on the orbital shaker. The heat block was
equilibrated inside an incubator to 40 °C for aggregating the temperature-
responsive reagents. (b) To magnetically capture the analyte the tube was
­incubated at 40 °C for an additional 15 min in DynaMag-2 Magnet. (c) The cap-
tured temperature-­responsive reagent aggregates were redissolved and applied
to the cellulose fiber pad (brown) sitting atop the modified lateral-flow strip
(Color figure online)

2. To aggregate the temperature-responsive reagents and the

bound analytes, the tube was incubated 15 min with orbital
shaking in a 40 °C aluminum heat block equilibrated inside an
incubator after the reagent addition (Fig. 4a).
3. To magnetically capture the analyte the tube was incubated at
40 °C for an additional 15 min in close contact with a Nd
magnet, using DynaMag-2 Magnet (Fig. 4b).
4. The supernatant was carefully removed and discarded with a
pipet, and then 10 μL 4 °C PBS was added into the tube to
redissolve the temperature-responsive reagent aggregates cap-
tured along the wall of the tube, resulting in 50-fold volumet-
ric enrichment.
5. Seven microliters of the enriched temperature-responsive
reagent was then applied onto the cellulose fiber pad sitting
 Biomarker Enrichment 81

Fig. 5 Comparison of magnetic enrichment and commercial assay. Flow strip images from a 50-fold magnetic
enrichment immunoassay (top row) and from the unmodified commercial assay performed with no enrichment
(bottom row). Reproduced from ACS Nano 2012 with permission from ACS Publications

atop the lateral-flow assay membrane. The liquid was allowed

to wick into the strip for 60 s, after which, 60 μL of the rinse
buffer included in the commercial kit was applied to chase the
temperature-­responsive reagents (Fig. 4c).
6. The test was allowed to develop for 6–7 min in total, followed
by removal of the absorbent pads, and air drying. Figure 5
shows scanned images obtained after performing a 50-fold
enrichment assay (top row) or a conventional non-enriched
assay (bottom row) with the commercial gold conjugate
included in the flow strip cassette. All test strips utilized 7 μL
specimen volume. Compare to the non-enriched assays, the
assays with the enriched specimens show darker bands of gold
colloid absorbance. The enrichment enabled the assay to
detect 10 ng/mL PfHRP2, while the non-enriched assay
could only detect 25 ng/mL PfHRP2.

4  Notes

1. Recrystallization of NIPAAm: 20 % (v/v) NIPAAm was dis-

solved in n-hexane via gentle heating, ca. 30 °C. The NIPAAm
monomer was recrystallized by cooling to room temperature,
ca. 22 °C. The recrystallized monomer was filtered and dried
in vacuo for 16 h prior to the polymerization.
2. 1H NMR results from diblock copolymer preparations showed
that the chemical shift of the DMAEAm methyl group pro-
tons was dependent on the protonation state of the tertiary
amine. The four protons located between the amide and ter-
tiary amine groups of DMAEAm were assigned to the peak at
δ (ppm) 3.1, which represented 10 DMAEAm protons.
Therefore, the degree of polymerization of DMAEAm in the
diblock was determined using integrated peak area ratio of the
82 James J. Lai and Patrick S. Stayton

single NIPAAm hydrogen at δ (ppm) 4.0 to the 10 DMAEAm

protons at δ (ppm) 3.1.
3. All glassware was cleaned with aqua regia, thoroughly rinsed
with DI water, and dried before use.
4. The extinction coefficient of the 23 nm gold colloids was esti-
mated to be 2 × 109 M−1 cm−1 from sizing data and the litera-
ture [42, 43]. Therefore, the concentration of gold colloids
can be determined using the estimated extinction coefficient.
5. Carboxylates (–COOH) may be reacted to sulfo-NHS in the
presence of EDC, resulting in a semi-stable sulfo-NHS ester,
which may then be reacted with primary amines (–NH2) to
from amide covalent bond. However, sulfo-NHS ester will
hydrolyze within hours or minutes, depending on water-­
content and pH of the conjugation solution.
6. Iron(0) pentacarbonyl is easily oxidized, which form precipi-
tates. Therefore, the sealed bottle needed to be purged with
N2 for 5 min before and after the iron(0) pentacarbonyl was
transferred to the reaction flask. Additionally, the iron(0) pen-
tacarbonyl solution was filtered using Millex-FG syringe filter
before adding to the reaction flask.
7. The antibody molar concentration, Mantibody, can be calculated
using a UV–vis spectrophotometer.
Ac
M antibody = 280 , where ε is the antibody molar extinction
e ´l
coefficient, 210,000 M−1 cm−1, l is the cuvette path length, and
Ac280 is corrected absorbance at 280 nm, Ac280 = A280 −
(A354 × 0.23). The chromogenic biotin molar concentration,
A
Mbiotin, can be estimated using M biotin = 354 , where the biotin
e ´l
molar extinction coefficient is 29,000 M−1 cm−1. Therefore,
M biotin
the degree of biotinylation can be estimated by .
M antibody

References
1. Anderson NL (2010) The clinical plasma pro- 5. Yager P, Domingo GJ, Gerdes J (2008) Point-
teome: a survey of clinical assays for proteins in of-­care diagnostics for global health. Annu Rev
plasma and serum. Clin Chem Biomed Eng 10:107–144
56(2):177–185 6. Balsam J et al (2014) Thousand-fold fluores-
2. Ng OT et al (2012) Accuracy and user-­ cent signal amplification for mHealth diagnos-
acceptability of HIV self-testing using an oral tics. Biosens Bioelectron 51:1–7
fluid-based HIV rapid test. PLoS One 7. Istepanian R, Jovanov E, Zhang YT (2004)
7(9):e45168 Introduction to the special section on
3. Hay Burgess DC, Wasserman J, Dahl CA M-Health: beyond seamless mobility and
(2006) Global health diagnostics. Nature global wireless health-care connectivity. IEEE
444(Suppl 1):1–2 Trans Inf Technol Biomed 8(4):405–414
4. Urdea M et al (2006) Requirements for high 8. Mudanyali O et al (2012) Integrated rapid-
impact diagnostics in the developing world. diagnostic-­test reader platform on a cellphone.
Nature 444(Suppl 1):73–79 Lab Chip 12(15):2678–2686
 Biomarker Enrichment 83

9. Giljohann DA, Mirkin CA (2009) Drivers of desorption/ionization time-of-flight mass

biodiagnostic development. Nature 462(7272): spectrometry. Clin Chem 53(3):421–428
461–464 24. Lai JJ et al (2007) Dual magnetic-/
10. Moody A (2002) Rapid diagnostic tests for temperature-­ responsive nanoparticles for
malaria parasites. Clin Microbiol Rev 15(1): microfluidic separations and assays. Langmuir
66–78 23(13):7385–7391
11. Posthuma-Trumpie GA, Korf J, van 25. Lai JJ et al (2009) Dynamic bioprocessing and
Amerongen A (2009) Lateral flow (immuno) microfluidic transport control with smart mag-
assay: its strengths, weaknesses, opportunities netic nanoparticles in laminar-flow devices.
and threats. A literature survey. Anal Bioanal Lab Chip 9(14):1997–2002
Chem 393(2):569–582 26. Hoffman AS (2000) Bioconjugates of intelli-
12. Beadle C et al (1994) Diagnosis of malaria by gent polymers and recognition proteins for use
detection of Plasmodium falciparum HRP-2 in diagnostics and affinity separations. Clin
antigen with a rapid dipstick antigen-capture Chem 46(9):1478–1486
assay. Lancet 343(8897):564–568 27. Nash MA et al (2010) Mixed stimuli-­responsive
13. Bell D, Peeling RW (2006) Evaluation of rapid magnetic and gold nanoparticle system for rapid
diagnostic tests: malaria. Nat Rev Microbiol purification, enrichment, and detection of bio-
4(9 Suppl):S34–S38 markers. Bioconjug Chem 21(12):2197–2204
14. Laderman EI et al (2008) Rapid, sensitive, and 28. Malmstadt N, Hoffman AS, Stayton PS (2004)
specific lateral-flow immunochromatographic “Smart” mobile affinity matrix for microfluidic
point-of-care device for detection of herpes immunoassays. Lab Chip 4(4):412–415
simplex virus type 2-specific immunoglobulin 29. Malmstadt N et al (2003) A smart microfluidic
G antibodies in serum and whole blood. Clin affinity chromatography matrix composed of
Vaccine Immunol 15(1):159–163 poly(N-isopropylacrylamide)-coated beads.
15. Yager P et al (2006) Microfluidic diagnostic Anal Chem 75(13):2943–2949
technologies for global public health. Nature 30. Malmstadt N et al (2003) Affinity
442(7101):412–418 ­thermoprecipitation and recovery of biotinyl-
16. Hauck TS et al (2010) Nanotechnology diag- ated biomolecules via a mutant streptavidin-
nostics for infectious diseases prevalent in smart polymer conjugate. Bioconjug Chem
developing countries. Adv Drug Deliv Rev 14(3):575–580
62(4–5):438–448 31. Ebara M et al (2013) A photoinduced
17. Wilson SM et al (1992) Adaptation of a radio- nanoparticle separation in microchannels via
active L. donovani complex DNA probe to a pH-­sensitive surface traps. Langmuir 29(18):
chemiluminescent detection system gives 5388–5393
enhanced sensitivity for diagnostic and epide- 32. Ebara M et al (2006) Switchable surface traps
miological applications. Parasitology 104(Pt for injectable bead-based chromatography in
3):421–426 PDMS microfluidic channels. Lab Chip
18. Gordon J, Michel G (2008) Analytical sensitiv- 6(7):843–848
ity limits for lateral flow immunoassays. Clin 33. Nash MA et al (2012) Multiplexed enrichment
Chem 54(7):1250–1251 and detection of malarial biomarkers using a
19. Horak D et al (2007) Preparation and proper- stimuli-responsive iron oxide and gold
ties of magnetic nano- and microsized particles nanoparticle reagent system. ACS Nano
for biological and environmental separations. J 6(8):6776–6785
Sep Sci 30(11):1751–1772 34. Zhao W, Brook MA, Li Y (2008) Periodic
20. el Ibrahim SA et al (2012) Development and assembly of nanospecies on repetitive DNA
characterization of thermosensitive pluronic-­ sequences generated on gold nanoparticles by
based metronidazole in situ gelling formula- rolling circle amplification. Methods Mol Biol
tions for vaginal application. Acta Pharm 474:79–90
62(1):59–70 35. Wilson R (2008) The use of gold nanoparticles
21. Lin YS et al (2005) Affinity capture using in diagnostics and detection. Chem Soc Rev
vancomycin-­bound magnetic nanoparticles for 37(9):2028–2045
the MALDI-MS analysis of bacteria. Anal 36. Dar VS, Ghosh S, Broor S (1994) Rapid detec-
Chem 77(6):1753–1760 tion of rotavirus by using colloidal gold parti-
22. Tong XD, Xue B, Sun Y (2001) A novel mag- cles labeled with monoclonal antibody. J Virol
netic affinity support for protein adsorption and Methods 47(1–2):51–58
purification. Biotechnol Prog 17(1):134–139 37. Kunakorn M et al (1991) Gold blot for detec-
23. Fiedler GM et al (2007) Standardized pepti- tion of immunoglobulin M (IgM)- and IgG-­
dome profiling of human urine by magnetic specific antibodies for rapid serodiagnosis of
bead separation and matrix-assisted laser melioidosis. J Clin Microbiol 29(9):2065–2067
84 James J. Lai and Patrick S. Stayton

38. Hsu YH (1984) Immunogold for detection of 41. Frens G (1973) Controlled nucleation for reg-
antigen on nitrocellulose paper. Anal Biochem ulation of particle-size in monodisperse gold
142(1):221–225 suspensions. Nat Phys Sci 241(105):20–22
39. Nash MA et al (2010) “Smart” diblock copo- 42. Khlebtsov NG (2008) Determination of size
lymers as templates for magnetic-core gold- and concentration of gold nanoparticles
shell nanoparticle synthesis. Nano Lett from extinction spectra. Anal Chem 80(17):
10(1):85–91 6620–6625
40. Nash MA et al (2010) Laboratory-scale protein 43. Haiss W et al (2007) Determination of size
striping system for patterning biomolecules and concentration of gold nanoparticles
onto paper-based immunochromatographic from UV-vis spectra. Anal Chem 79(11):
test strips. Lab Chip 10(17):2279–2282 4215–4221
 Chapter 6

Microfluidic Toner-Based Analytical Devices: Disposable,

Lightweight, and Portable Platforms for Point-of-Care
Diagnostics with Colorimetric Detection
Karoliny Almeida Oliveira, Fabrício Ribeiro de Souza,
Cristina Rodrigues de Oliveira, Lucimeire Antonelli da Silveira,
and Wendell Karlos Tomazelli Coltro

Abstract
This chapter describes the development of microfluidic toner-based analytical devices (μTADs) to perform
clinical diagnostics using a scanner or cell-phone camera. μTADs have been produced in a platform com-
posed of polyester and toner by the direct-printing technology (DPT) in a matter of minutes. This technol-
ogy offers simplicity and versatility, and it does not require any sophisticated instrumentation. Toner-based
devices integrate the current generation of disposable analytical devices along paper-based chips. The cost
of one μTAD has been estimated to be lower than $0.10. In addition, these platforms are lightweight and
portable thus enabling their use for point-of-care applications. In the last 5 years, great efforts have been
dedicated to spread out the use of μTADs in bioassays. The current chapter reports the fabrication of
printed microplates and integrated microfluidic toner-based devices for dengue diagnostics and rapid colo-
rimetric assays with clinically relevant analytes including cholesterol, triglycerides, total proteins, and glu-
cose. The use of μTADs associated with cell-phone camera may contribute to the health care, in special, to
people housed in developing regions or with limited access to clinics and hospitals.

Key words Capillary-driven microfluidics, Cell-phone camera, Clinical assays, Enzyme-linked

immunosorbent assay, Immunoassays, Telemedicine, Tropical diseases

1 Introduction

In the last 25 years, the field related to the analytical instrumen-

tation has been revolutionized by the advances promoted with
the miniaturization trends. The use of miniaturized analytical
devices offers attractive advantages which include low sample
consumption, short analysis time and capability of integrating
multiple conventional analytical steps on a single chip.
Furthermore, the reduced size of the devices enables their use in

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_6, © Springer Science+Business Media New York 2015

85
86 Karoliny Almeida Oliveira et al.

point-of-care (PoC) applications due to the inherent portability,

easiness of use and, most importantly, sample-in-answer-out
capability [1].
Miniaturized devices may be fabricated by conventional or
alternative microfabrication techniques. Standard photolithogra-
phy is often explored to produce analytical devices in micro scale
once it is well established in semiconductor industry and it also
presents excellent resolution to define structures with dimensions
below 10 μm. However, it requires sophisticated instrumentation
which restricts its access to a few research groups due to the limited-
resources funding [2].
In this scenario, do Lago and coworkers proposed in 2003 the
simplest, the fastest, and the cheapest technique to produce minia-
turized devices based on a direct laser-printing process [3]. This
alternative fabrication process requires simple instruments easily
found in office stores like computers (desktops or laptops), laser
printers, and laminator or silk machines. In addition, this direct-
printing technology (DPT) requires low-cost consumables which
include toner cartridge and transparency sheets. The basic princi-
ple involved in the DPT comprises three simple steps: (1) the lay-
out drawing in a graphic software, (2) the laser printing on
transparency (polyester) films, and (3) the thermal lamination to
provide the channel sealing. During the laser printing, a toner layer
(ca. 7 μm thick) is deposited on a polyester film. This printed layer
can be directly used to delimit microzones or thermally laminated
against multiple printed polyester films forming a sandwiched
device for capillary-driven microfluidic applications, for example.
Toner is a chemical powder containing mostly a mixture of poly-
mer and iron or silicon oxide. This material is used in laser printers
and photocopiers to form an image on a smooth surface based on
a layout previously designed in specific software. The complete
composition depends on each manufacturer; however, the poly-
meric base is often composed of styrene and acrylate copolymers.
In addition, monochromatic and polychromatic laser printers con-
tain, respectively, iron and silicon oxide which are responsible to
promote the electrostatic interaction between the toner and the
desirable surface, thus creating the image during the laser-printing
step [4]. For microfluidic applications, the laser-printed image may
include spot zones defined by toner lines for chemical assays of
channels limited by toner walls for analytical separations as well as
flow-based assays.
This technology has been explored by different groups for the
development of microfluidic toner-based analytical devices
(μTADs) dedicated to a wide range of applications [5, 6]. Some
examples include electrospray tips [3], electrophoresis microchips
[4, 7], micromixers [8], preconcentrators [9], and devices for
genetics analysis [10, 11]. More recently, μTADs have been applied
to carry out clinical diagnostics [12] and immunoassays based on
 Disposable Devices for Clinical Diagnostics 87

enzyme-linked immunosorbent assay (ELISA) [13] or detection of

C-reactive protein (CRP) [14].
In this chapter, we describe the development of μTADs for
clinical diagnostics. The feasibility of clinical applications have
been demonstrated with the use of (1) printed microplates con-
taining 96 microzones and (2) integrated microfluidic devices for
quick distribution of sample through microchannels. Printed
microplates were evaluated for the detection of immunoglobulins
M (IgM) and G (IgG) in serum samples from patients infected
with dengue virus based on enzyme linked immunosorbent assay
(ELISA) technique. The integrated μTADs were explored to
investigate the feasibility of performing simultaneous assays with
clinically important analytes including cholesterol, total proteins,
glucose, and triglycerides. The first version of μTADs for this pur-
pose comprised four detection zones integrated with a sample
inlet central zone by printed microchannels. In both applications,
colorimetric reactions were employed to allow the color capture
with popular electronic devices, like scanner or cell-phone camera.
The captured images were converted to a color scale and analyzed
in a graphic software.

2 Materials

2.1 Toner-Based 1. Computer equipped with graphic software.

Analytical Devices 2. Office LaserJet printer.
3. Polyester sheets (3 M, Austin, TX, USA).
4. Laminator machine.
5. Paper punch.
6. CO2 laser ablation system.

2.2 Colorimetric 1. Deskjet multifunction printer.

Detection 2. Cell-phone camera.

2.3 ELISA 1. Commercial kit (Bioeasy) for Dengue IgM Capture ELISA
Procedures and anti-dengue IgG solution for indirect ELISA were kept at
4–8 °C.
2. Anti-human IgG: 20 μg/mL IgG solution dissolved in carbon-
ate buffer at pH 9.6.
3. Dengue antigen solution: dengue antigen diluted in carbonate
buffer (1:16 v:v) at pH 9.6.
4. Serum samples from patients infected and non-infected with
dengue virus were supplied by a local Tropical Diseases
Hospital (HDT, Goiânia, GO, Brazil). The serum samples
were kept at −80 °C.
88 Karoliny Almeida Oliveira et al.

5. Washing solution: phosphate buffered saline solution (10 mM

phosphate buffer, 140 mM NaCl, 3 mM KCl) and 0.05 %
(v/v) of Tween-20 (PBS-T). The solution was stored at 4 °C.
6. Blocking solution (5 %): 5 g of skimmed milk powder diluted
in 10 mL of PBS-T solution 0.05 %.
7. The control sera, calibrator and biological samples were diluted
(1:10 v:v) in sample diluent solution of commercial kits.
8. Anti-mouse IgG antibody was diluted (1:10,000) in buffer
PBS-T containing 0.5 % skimmed milk prior to use.
9. Anti-human IgG antibody (1.1 mg/mL) was diluted (1:5,000)
in buffer PBS-T containing 0.5 % skimmed milk prior to use.
10. For Dengue IgM capture assay the reagents of the commercial
kit were mixed. The lyophilized dengue antigen was diluted
with 1.5 mL of the conjugated diluent and blended with con-
jugated anti-dengue in the proportion of 1:2.
11. Tetramethylbenzidine (TMB) chromogen and stopping solu-
tion were used according to instructions provided by the
manufacturer.

2.4 Clinical 1. Cellulose paste: cellulose powder and ultrapure water (1:4 m:m).
Diagnostics 2. Solution 1: 0.6 M potassium iodide and 0.3 M trehalose dis-
solved in 100 mM phosphate buffer at pH 6.0.
3. Solution 2: five parts of 181 U/mg glucose oxidase and one
part of 73 U/mg horseradish peroxidase.
4. Solution 3: 0.5 mM sodium cholate, 0.5 mM 4-aminoantipyrine,
300 U of cholesterol esterase, 204 U cholesterol oxidase,
828 U of horseradish peroxidase, and 35 mM piperazine-
N,N′-bis(2-ethanesulfonic acid) (PIPES) buffer at pH 7.0.
5. Solution 4: 0.12 mM Coomassie brilliant blue dye.
6. Solution 5: 6 mM 4-chlorophenol, 5 mM magnesium chloride,
0.75 mM 4-aminoantipyrine, 0.9 mM ATP, 132 kU of lipase,
1.5 kU glycerokinase, 4.02 kU glycerol 3-phosphate oxidase,
282 U of peroxidase and, 50 mM PIPES buffer at pH 7.5.

3 Methods

3.1 Fabrication 1. Toner microzone plates were fabricated by using the

of Printed Microplates DPT. The first step requires the layout drawing using the
graphic software. Microzones were designed with 7 mm
diameter. As shown in Fig. 1a, b, the layout of printed
microplates is quite similar to the standard polymeric micro-
plates and consists of 96 wells arranged into 12 columns
with 8 wells each (see Note 1).
 Disposable Devices for Clinical Diagnostics 89

b
a

c 8
d
6
Height (mm)

2
toner barrier width
0

-2
0.0 0.5 1.0 1.5 2.0
Distance (mm)

Fig. 1 Representation of the (a) fabrication scheme of microplates based on laser-printing step, (b) layout of a
toner-based 96-zone microplate for immunoassays, (c) toner barrier profile, and (d) typical printed microplate
with sample added in test zones. (Reproduced with permission from ref. [13]. Copyright 2013The Royal Society
of Chemistry)

The layout of microplates was directly printed on the transpar-

ency (polyester) surface by using a laser printer with 1,200-dpi reso-
lution. Toner-based microzones were created by the laser printing
of toner barriers (1-mm width) defined on graphic software.
2. The toner layer laser-printed on the polyester surface presents
a thickness of 5.0 ± 0.4 μm (Fig. 1c) and due to its hydropho-
bic nature it is enough to a spot test for bioanalytical assays
(Fig. 1d).

3.2 Fabrication 1. The fabrication of μTADs basically involves three steps related
of μTADs to (a) the drawing step, where the μTAD layout is designed,
(b) the laser printing which deposits a toner layer on the poly-
ester surface to form the channels, and (c) the sealing stage to
promote the bonding of two or more pieces of printed
channels.
2. During the drawing stage in a graphic software (see Note 1),
the colors black and white are often used to define the areas to
represent the microfluidic walls and channels, respectively.
When the file is sent to the printer, the black area will form a
toner layer with thickness about 7 μm. Oppositely, the laser
90 Karoliny Almeida Oliveira et al.

printer ideally does not deposit toner for the white lines or
reservoirs (also called of detection zones) previously defined in
the desirable layout. Consequently, the microchannels and
detection zones are easily created with a simple laser printing
step.
3. Besides the printed layout, its mirror image is also printed dur-
ing the same step to increase the channel depth.
4. A cut-through polyester film, with the same microfluidic struc-
ture from the layout, has been prepared by using a CO2 laser
ablation machine (see Note 2) and added between both printed
images in order to enhance the channel depth. This strategy
has been adopted to ensure the fluidic transport by capillary
action through printed channels.
5. Before laminating the polyester pieces, the detection zones in
the upper polyester piece were perforated with a paper punch
to have access the microfluidic channels.
6. Afterwards, the three polyester pieces were aligned and ther-
mally laminated at 150 °C in a laminator machine (see Note 3).
Figure 2 shows the 3D and cross-section views of a μTAD.
7. Finally, a paste made of cellulose powder and ultrapure water
was added to each test zone and allowed to dry at room tem-
perature during 30 min (see Note 4). The layout of the pro-
posed device (35 mm × 35 mm) consisted of four test zones
interconnected by microfluidic channels and one central inlet
zone to sample distribution (Fig. 2). All channels were 10-mm
long, 1-mm wide and ca. 100-μm deep. Figure 3 displays a real
image of a μTAD in comparison with the size of a quarter dol-
lar coin.

3.3 Determination 1. The determination of flow rate induced by capillary force on

of Flow Rate on μTADs μTADs has been performed with a capacitively coupled con-
tactless conductivity detector (C4D), which measures the con-
ductivity difference inside microchannels [15]. This system
usually requires two planar electrodes which are fixed outside
the channel (Fig. 4a).
2. In practice, a high frequency sinusoidal wave (400 kHz) is
applied to an excitation electrode (E0) and the output signal is
captured by a receiver electrode (E1) (Fig. 4a). Afterwards, the
resulting current is filtered, amplified, and converted in digital
signal. The recorded signal is monitored in real time as a func-
tion of the analysis time (Fig. 4b).
3. The flow rate magnitude was estimated based on the time
required for an aqueous solution to move the distance from
the injection point (label 1 in Fig. 4b) to the detector (label
2 in Fig. 4b). Taking into account the effective length and the
analysis time, the flow rate was successfully determined.
 Disposable Devices for Clinical Diagnostics 91

Fig. 2 Simplified fabrication process of toner-based microfluidic devices for clinical assay in (a) 3D and (b)
cross-sectional views and (c) layout of a typical device used to demonstrate the capability of performing colo-
rimetric assays. Reprinted with permission from ref. [12]. Copyright 2012 American Chemical Society

Fig. 3 Optical micrograph showing a real image of a μTAD in comparison with the
size of a quarter dollar coin
92 Karoliny Almeida Oliveira et al.

b 5

C4D Signal (V)

3

0
(1) (2)

0 5 10 15 20 25 30 35 40
Time (s)

Fig. 4 Examples of (a) the experimental arrangement for the determination of the flow rate magnitude and
(b) typical conductivity signal monitored over time. In (a), the labels E0 and E1 mean the excitation and
received electrodes, respectively. In (b), the labels (1) and (2) indicate the injection time and the time required
for an aqueous solution move from the injection point to the detector, respectively

Fig. 5 Simplified scheme of capture ELISA procedure performed on toner-based

microzones. In (a), the anti-human IgM was immobilized on the microzone; In
(b), the sera and samples were added to each zone; In (c), the mixture of lyophi-
lized dengue antigen solution and conjugated solution was incubated; In (d), the
TMB was added to provide the color change during the binding event

3.4 ELISA The ELISA procedures were performed on printed microplates to

Procedures on Toner- detect IgG and IgM in biological samples with colorimetric detec-
Based Microplates tion. Figure 5 depicts the simplified scheme that was used to carry
out capture ELISA procedures on printed microplates.
 Disposable Devices for Clinical Diagnostics 93

3.4.1 Capture ELISA 1. The anti-human IgM antibodies were provided by a local labo-
ratory of monoclonal antibodies production (CEPRACO,
Goiânia, GO, Brazil). First, anti-human IgM was immobilized
on microzones using 10-μL aliquots. The microplate was kept
overnight at 4–8 °C (see Note 5).
2. Afterwards, the microplate was washed three times with PBS-T
solution and dried at room temperature.
3. Each printed zone was individually blocked by adding 20 μL of
blocking buffer during 2 h at 37 °C and washed again with
PBS-T solution.
4. Then, 10-μL aliquots of diluted samples and controls were
added to each zone. The microplate was incubated at 37 °C
during 1 h and washed with buffer.
5. After the drying step, 10-μL aliquots of the lyophilized mix-
ture containing dengue antigen and conjugated anti-dengue
were added to each zone and kept during 1 h at 37 °C.
6. Lastly, the microplate was washed and aliquots of 10 μL of the
substrate solution were added to each zone. After keeping
10 min at room temperature, the assay was stopped by the
addition of 10 μL/zone of stop solution.

3.4.2 Indirect ELISA 1. Firstly, the dengue antigen was immobilized on microzones
Procedures using 10-μL aliquots and kept overnight at 2–8 °C. The stages
associated with the washing, blocking and addition of sam-
ples/controls were performed similarly to the steps described
for the capture ELISA assay (Subheading 3.4.1).
2. Afterwards, the anti-human IgG (10-μL aliquots) was added
on each zone and kept at 37 °C during 1 h.
3. Then, 10-μL aliquots containing the peroxidise-conjugated sec-
ondary antibody (rabbit anti-mouse IgG) diluted in blocking
buffer were added to each zone and kept at 37 °C during 1 h.
4. Finally, the zones were washed and aliquots of 10 μL of the
substrate solution were added to each zone as described in the
capture ELISA procedure.

3.4.3 Simultaneous The procedure to carry out simultaneous colorimetric tests for
Colorimetric Tests clinical glucose, cholesterol, proteins, and triglycerides were quite
on μTADs similar. The μTADs were prepared to perform the bioassays by
adding the specific color reagents in different detection zones, as
described below
1. For glucose assay, the detection zone was first spotted with
8 μL of solution 1 (Subheading 2.4) and dried at room tem-
perature during 10 min. Afterwards, the zone was spotted with
8 μL of solution 2 (Subheading 2.4) and dried for 10 min dur-
ing additional 10 min.
94 Karoliny Almeida Oliveira et al.

2. For cholesterol assay, the detection zone was spotted with 8 μL

of solution 3 (Subheading 2.4) and dried for 10 min.
3. For total protein assay, the detection zone was spotted with
8 μL of solution 4 (Subheading 2.4) and dried at room tem-
perature during 10 min.
4. For triglycerides assay, the detection zone was spotted with
8 μL of solution 5 (Subheading 2.4) and dried at room tem-
perature by 10 min.

3.5 Detection A cell-phone camera was used to measure the resulting colorimetric
System information related to the detection of IgM and IgG in samples
from patients infected with dengue virus on printed microplates. On
the other hand, the colorimetric detection of glucose, cholesterol,
proteins, and triglycerides on integrated μTADs was performed with
the scanner mode of a Deskjet multifunction printer.
1. The images captured with the cell-phone camera may be digi-
tally filtered using software to minimize interferences associ-
ated light brightness and scattering.
2. For the detection of cholesterol, glucose, proteins, and triglyc-
erides using the scanner, μTADs were placed and fixed on the
glass of scanner. In this arrangement, the image was captured
in the bottom part of the device.
3. The images recorded with both cell-phone camera and scanner
were converted to a CMYK color scale in graphic software. For
the color developed inside each zone, the arithmetic mean of
pixel intensity (Fig. 6) was used to quantify the concentration
of the analytes (see Note 6).

Fig. 6 Example of the graphic tool usually explored to extract the mean color intensity. This information is
required to quantify the analyte concentration in samples
 Disposable Devices for Clinical Diagnostics 95

3.6 Application 1. The ELISA test is a solid-phase enzyme immunoassay which

of μTADs for Point-of- allows the detection of antigens in a solution based on their
Care Diagnostics interactions with specific antigens usually immobilized on a
surface solid. The TMB substrate added on the reaction is a
3.6.1 Dengue
chromogenic compound for horseradish peroxidase (HRP)
Diagnostics
yielding a blue color that changes to yellow upon addition of
an acid stop solution. The color of this reaction is developed
when the HRP binds to the substrate in positive samples.
2. We have tested sera samples collected from patients, a blank
(B) sample (blocking buffer) as well as negative (NC) and posi-
tive (PC) controls. Positive (PS) and negative (NS) samples
were collected from infected and non-infected patients with
Dengue virus, respectively. PC samples were prepared to con-
tain IgG or IgM and thus to be a reference for the proposed
diagnostic on printed zones.
3. The presence of high levels of IgM and IgG in serum sample is
an indicative of Dengue infection (Fig. 7). On the other hand,

a
70
60
Color Intensity

50
40
30
20
10
0
B NC PC PS NS
Test samples
b
120

100
Color Intensity

80

60

40

20

B NC PC PS NS
Test samples

Fig. 7 ELISA data recorded with cell phone camera showing the detection of
(a) IgM and (b) IgG based on IgM capture and indirect ELISA procedures, respec-
tively. The labels B, NC, and PC mean blank sample, negative control, and posi-
tive control, respectively. Positive (PS) and negative (NS) samples are related to
infected and non-infected patients. The captured images are inset in the figures
96 Karoliny Almeida Oliveira et al.

the analysis of the serum sample from the non-infected patient

(NS) has provided absorbance intensity similar to those
recorded for B and NC samples. This result was expected once
the non-infected patient does not contain IgM or IgG pro-
duced by its organism as immunological defense against the
viral infection (Fig. 7).
4. The images captured with cell-phone camera for the Dengue
assay are shown inset Fig. 7. The mean pixel intensity was
determined in triplicate using three independent zones for
each assay. The average value for control and biological sam-
ples are displayed in Fig. 7, where the bars mean the respective
standard deviation values. The results found with cell-phone
camera are in agreement to the values found by conventional
microplate reader (data not shown). Due to the global acces-
sibility and also the large number of applications (apps) avail-
able on modern cell phones (e.g., Bluetooth, Multimedia
Messaging Service, and Wireless connection), the captured
images can be easily transmitted to a specific place where a
technician or a medical staff would have conditions to analyze
it. Consequently, a prescription or a suitable treatment could
be sent back to the patient by the same way (cell phone). This
possibility is called of telemedicine and it can be helpful to
provide fast diagnostics of diseases at early stages.

3.6.2 Clinical Assays 1. μTADs have been explored to detect glucose, total proteins,
with Artificial Sera cholesterol, and triglycerides in artificial sera samples. Glucose
Samples is the main indicator of diabetes in biological fluids like urine,
serum and blood. The levels of total protein can be helpful to
diagnose renal diseases or disorders. On the other hand, the
concentration levels of cholesterol and triglycerides can be
indicative factors of coronary risks. Based on the high impor-
tance of all analytes mentioned, the capability of μTADs per-
form simultaneous assays can contribute to an early diagnostic
of the patient
2. Before demonstrating the simultaneous analysis, each bioassay
was separately performed in a single device. In this case, all
four detection zones were prepared according to the proce-
dures described in Subheading 3.2.
3. The use of four detection zones has allowed investigating the
zone-to-zone repeatability for each assay. The standard relative
deviation (RSD) values for each assay have been lower than
8 %. Based on the reaction time optimization, the images
should be captured after 10 min to obtain reliable analytical
concentrations.
4. Once demonstrated good reproducibility and high accuracy
for individual assays, the simultaneous tests for cholesterol,
 Disposable Devices for Clinical Diagnostics 97

Fig. 8 Typical example of simultaneous colorimetric assays for cholesterol, glu-

cose, total protein, and triglycerides on μTADs based on capillary-driven
microfluidics

glucose, total proteins, and triglycerides were performed with

a mixture of standard solutions. In this case, each zone of the
μTAD was spotted with specific reagents for individual bioas-
says, as described earlier in Subheading 3.4.3.
5. A 40-μL of the standard solution was then added to the sample
inlet central zone with a manual micropipette. Due to the
improved aspect-to-ratio promoted by the use of an intermedi-
ary transparency film, the fluidic channels ensured an effective
transport of the sample by capillary action, i.e., without exter-
nal instruments like pumps or high-voltage power supplies.
The spontaneous transport towards the detection zones
induces the capability of performing the assays in regions of
restricted access.
6. The colorimetric detection with a scanner has ensured great
reproducibility and reliability for the determination of all ana-
lytes in biological samples. Figure 8 depicts a typical example
of simultaneous assay for glucose, cholesterol, total proteins,
and triglycerides on μTADs.
7. The determination of the concentration levels of total proteins,
cholesterol, and glucose in artificial sera samples have demon-
strated good correlation with the values reported by the sup-
plier. The estimated error has been ca. 10 %.

4 Notes

1. The layout of devices can be defined in several dimensions

since it fits the polyester sheet size.
2. If you do not have access to a laser cutting machine, the cut-
ting of the intermediary transparency piece may be performed
98 Karoliny Almeida Oliveira et al.

with a scalpel. However, this procedure can lead to observation

of many imperfections along the channel.
3. Alternatively, the lamination step can be done with a silk screen
machine. This equipment allows the sealing of multiple μTADs
in a single step.
4. Once dried, the cellulose paste forms a porous structure that
retains the color reagent for each bioassay inside each zone.
This strategy ensures the color development just in the detec-
tion zone and the assay-to-assay reproducibility.
5. The aliquot volume can be modified according to the detec-
tion zone diameter.
6. In our experiments, we used the CMYK color channel; how-
ever, other color channel (RGB, gray scale) or models (HIS,
for example) can also be employed for the same purpose.

References
1. Arora A, Simone G, Salieb-Beugelaar GB, Kim fabrication of passive micromixers microfluidic
JT, Manz A (2010) Latest developments in devices using a direct-printing process. Lab
micro total analysis systems. Anal Chem Chip 5:974–978
82:4830–4847 9. Yu H, Lu Y, Zhou YG, Wang FB, Hi FY, Xia
2. Chen Y, Pépin A (2001) Nanofabrication: XH (2008) A simple disposable microfluidic
conventional and nonconventional methods. devices for rapid protein concentration and
Electrophoresis 22:187–207 purification via direct-printing. Lab Chip
3. do Lago CL, da Silva HDT, Neves CA, Brito- 8:1496–1501
Neto JGA, da Silva JAF (2003) A dry process 10. Duarte GRM, Coltro WKT, Borba JC, Price
for production of microfluidic devices based CW, Landers JP (2012) Disposable polyester-
on the lamination of laser-printed polyester. toner electrophoresis microchips for DNA
Anal Chem 75:3853–3858 analysis. Analyst 137:2692–2698
4. Gabriel EFM, do Lago CL, Gobbi AL, Carrilho 11. Duarte GRM, Price CW, Augustine BH,
E, Coltro WKT (2013) Characterization of Carrilho E, Landers JP (2011) Dynamic solid
microchip electrophoresis devices fabricated by phase DNA extraction and PCR amplification
direct-printing process with colored toner. in polyester-toner based microchip. Anal
Electrophoresis 34:2169–2176 Chem 83:5182–5189
5. Coltro WKT, de Jesus DP, da Silva JAF, Do 12. de Souza FR, Alves GL, Coltro WKT (2012)
Lago CL, Carrilho E (2010) Toner and paper- Capillary-driven toner-based microfluidic
based fabrication techniques for microfluidic devices for clinical diagnostics with colorimet-
applications. Electrophoresis 31:2487–2498 ric detection. Anal Chem 84:9002–9007
6. Coltro WKT, Carrilho E (2008) Low-cost 13. Oliveira KA, de Oliveira CR, da Silveira LA,
technologies for microfluidics applications. In: Coltro WKT (2013) Laser-printing of toner-
Landers JP (ed) Handbook of capillary and based 96-microzone plates for immunoassays.
microchips electrophoresis and associated Analyst 138:1114–1121
microtechniques, 3rd edn. New York: CRC 14. Kim AR, Kim JY, Choi K, Chung DS (2013)
Press, pp 1169–1184 On-chip immunoassay of a cardiac biomarker
7. Gabriel EFM, Duarte GF, Garcia PT, de Jesus in serum using a polyester-toner microchip.
DP, Coltro WKT (2012) Polyester-toner elec- Talanta 109:20–25
trophoresis microchips with improved analytical 15. Coltro WKT, Lima RS, Segato TP, Carrilho E,
performance and extended lifetime. de Jesus DP, do Lago CL, da Silva JAF (2012)
Electrophoresis 33:2660–2667 Capacitively coupled contactless conductivity
8. Liu AL, He FY, Wang K, Zhou T, Lu Y, Xia detection on microfluidic systems-ten years of
XH (2005) Rapid method for design and development. Anal Methods 4:24–33
 Chapter 7

Detection of Protein Biomarker Using a Blood

Glucose Meter
Tian Lan, Yu Xiang, and Yi Lu

Abstract
mHeath technologies are recognized to play important roles in the future of personal care and medicine.
However, their full potentials have not been reached, as most of current technologies are restricted to
monitoring physical and behavioral parameters, such as body temperature, heart rate, blood pressure, and
physical movement, while direct monitoring of biomarkers in body fluids can provide much more accurate
and useful information for medical diagnostics. A major barrier to realizing the full potential of mHealth
is the high costs and long cycles of developing mHealth devices capable of monitoring biomarkers in body
fluids. To lower the costs and shorten the developmental cycle, we have demonstrated the leveraging of
the most successful portable medical monitoring device on the market, the blood glucose meter (BGM),
with FDA-approved smartphone technologies that allow for wireless transmission and remote monitoring
of a wide range of non-glucose targets. In this protocol, an aptamer-based assay for quantification of
interferon-γ (IFN-γ) using an off-the-shelf BGM is described. In this assay, an aptamer-based target rec-
ognition system is employed. When IFN-γ binds to the aptamer, it triggers the release of a reporter
enzyme, invertase, which can catalyze the conversion of sucrose (not detected by BGM) to glucose. The
glucose being produced is then detected using a BGM. The system mimics a competitive enzyme-linked
immunosorbent assay (ELISA), where the traditional immunoassay is replaced by an aptamer binding
assay; the reporter protein is replaced by invertase, and finally the optical or fluorescence detector is
replaced with widely available BGMs.

Key words Blood glucose meter (BGM), Aptamer, Biosensor, Point-of-care diagnostic, Electrochemical
sensor

1 Introduction

Recent advances in mobile health (mHealth) technologies have led

to significant changes in ways people can manage their health.
A large number of wearable devices have been developed to moni-
tor different parameters of a person’s physiological state, such as
blood pressure, heart rate, and body temperature, as well as behav-
ior, such as medication adherence and physical movement [1].
While these physical and behavioral results are important indica-
tors of human health, they are far from satisfactory in accurate

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_7, © Springer Science+Business Media New York 2015

99
100 Tian Lan et al.

medical diagnostics, which requires direct detection and monitor-

ing of biomarkers in body fluids such as blood, saliva, or urine [1].
Therefore, there is still a huge gap between clinical diagnosis in the
hospitals and point-of-care tests using mHealth devices. To bridge
such a gap, new technological advances have led to several new
products and approaches for remote biomarker monitoring. One
example is the mChip system, based on an ELISA assay housed in
a microfluidic disposable and an optical reader, which has been
developed and field-tested for HIV diagnosis in Rwanda. The diag-
nostic results are transmitted and stored in the cloud [2]. Despite
these advances, very few mHealth devices capable of detecting bio-
markers in body fluids are available on the market. One reason for
such a situation is the high costs and long cycles associated with
developing mHealth technologies and most such devices are dedi-
cated to only one or a few biomarker detections.
To lower the costs and shorten the cycles of developing new
mHealth technologies that can be generally applied to a wide range
of biomarkers, we proposed and demonstrated an alternative
approach of repurposing existing technologies and devices for a
much wider range of applications. The blood glucose meter (BGM)
is an excellent platform to leverage, because it has gone through
decades of research and development, making the current genera-
tion of BGMs accurate, well designed for simple operation, low
cost, and portable. More importantly, several network-connected
smartphone-compatible BGMs [3–5] are available and some of
them have already been approved by FDA. In fact, these network-
connected BGMs account for the majority of current mHealth
devices capable of monitoring biomarkers [1]. Furthermore, BGM
technologies are constantly being improved due to the growing
number of diabetes [6]. By taking advantage of this highly devel-
oped and widely available mHealth device, we and others have
developed novel methodologies to transform the binding of non-
glucose biomarkers by either aptamers or antibodies into glucose
so that network-connected smartphone-compatible BGMs can be
used to detect and monitor a wide range of targets, such as metal
ions, small organic molecules, protein markers, and nucleic acids
[7–12].
In this protocol, a method is described to measure a protein
biomarker, IFN-γ, using an unmodified, off-the-shelf BGM [7].
IFN-γ is a cytokine released by immune cells and its level can be a
general indicator for many infectious diseases [13]. An aptamer
[14–18] is used for the selective binding of IFN-γ. An aptamer
possesses similar ligand binding properties as an antibody. Aptamers
can be isolated via an in vitro selection process and they are a prom-
ising new technology for diagnostic assays in the near future. A
more detailed review of current aptamer technology can be found
elsewhere [19]. Binding of IFN-γ to the aptamer triggers the
release of an enzyme, invertase. As an enzyme used widely in the
 BGM for Protein Detection 101

Fig. 1 Design of the aptamer based assay for detection of IFN-γ using a BGM. The assay system is composed
of a DNA-modified invertase (blue DNA sequence), hybridized with an IFN-γ aptamer (black DNA sequence),
which also hybridizes with a biotin modified DNA (green DNA sequence) that is bound to streptavidin-coated
magnetic beads (brown sphere ). In the absence of IFN-γ, complementary DNA hybridization between the three
DNA sequences keeps the enzyme invertase bound on the magnetic beads. Upon addition of IFN-γ, the binding
of IFN-γ to aptamer results in the dehybridization of thiol-modified DNA, leading to the release of invertase
from the magnetic beads. The released invertase can be separated from the remaining bead-bound invertase
by magnetic separation. The released invertase is then used to produce glucose for BGM detection, from
sucrose (which is undetectable by BGM)

confectionery industry, invertase is capable of converting sucrose,

which cannot be detected by a glucose meter, into glucose. The
glucose produced by invertase then can be quantified by an unmod-
ified, off-the-shelf BGM. A scheme of the assay is shown in Fig. 1.
Three steps are performed to complete the assay: (1) conjugation of
thiol-modified DNA to invertase; (2) immobilization of invertase
on magnetic beads; and (3) detection of IFN-γ using a BGM.
In addition to aptamers, antibodies have also been used as
target recognition elements in BGM-based assays [9]. The
BGM-based immunoassay has been developed using a sandwich
type assay system, where instead of using reporter proteins to
generate a fluorescent or colorimetric signal, such as HRP or
NADH dehydrogenase, invertase is used to generate glucose, for
convenient detection by existing BGMs. This type of sandwich
102 Tian Lan et al.

assay has been developed for Prostate Specific Antigen. Simple

replacement of the reporter protein in the traditional ELISA
provided a more convenient way to convert today’s ELISA assay
to a BGM based immunoassay.
To apply the BGM-based biomarker detection in mHealth set-
tings, a more simple and elegant solution will need to be developed
to perform the different steps of the assay. The solution for mHealth
application would encompass two different components: a dispos-
able assay cartridge and a reader. The assay reagents can be stored
in the cartridge. The different steps of the assay, e.g., sample col-
lection, pretreatment, binding of biomarkers, release of invertase
and glucose production, can be driven by the cartridge. A number
of existing technologies can be used to design the cartridge, for
example, lateral flow device and various microfluidic designs.
Ultimately, the amount of a specific biomarker is correlated to a
concentration of glucose, which can be precisely quantified. As the
product of decades of engineering and refinement, today’s BGM
offers an excellent system for convenient and comfortable opera-
tion by millions of diabetes. These BGMs are developed with
mobile monitoring, cloud storage, Electronic Health Record inte-
gration, and remote analysis in mind, such as the iBG STAR® and
Telcare® blood glucose monitoring systems. Future generation
BGMs are focused on more accurate and sensitive detection, as
well as noninvasive and continuous monitoring. The reader for the
BGM-based biomarker assays can be designed or modified based
on these existing BGMs to leverage matured technologies. The
BGM-based biomarker assays being developed can potentially
expand the biomarkers for mHealth application greatly beyond
physical body parameters, such as body temperature, heart rate,
and exercise tracking.

2 Materials

1. The glucose meter used for the protocol is an ACCU-CHEK

Avia glucose meter which can be found in stores. Other glu-
cose meters, such as the iBG STAR® and Telcare® blood glu-
cose monitoring systems, can also be used for the same
application.
2. Streptavidin-coated magnetic beads (1 μm average diameter)
are purchased from Bangs Laboratories (Fishers, IN). The
magnetic rack used for separation is purchased from PIERCE
Biotechnology (Rockville, IL).
3. Amicon-10K and Amicon-100K (500 μL capacity) are pur-
chased from Millipore Corporation (Billerica, MA).
4. Grade VII invertase (S. cerevisiae) and human recombinant
interferon-γ (IFN-γ) are purchased from Sigma-Aldrich
(St. Louis, MO).
 BGM for Protein Detection 103

5. General chemicals and human serum are purchased from

Sigma-Aldrich (St. Louis, MO) and are used as received unless
otherwise specified.
6. Three different oligonucleotides are used and custom-
synthesized by Integrated DNA Technologies, Inc. (Coralville, IA).
The list of oligonucleotides is shown in Table 1. Oligos are
HPLC purified by the vendor and they are used without fur-
ther purification.
7. Buffers and solutions used in this protocol:
Buffer A: 100 mM sodium phosphate (NaPi), 100 mM NaCl,
0.05 % Tween-20, pH 7.3.
Buffer B: 100 mM NaPi, 100 mM NaCl, pH 7.3.
Buffer C: 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesul-
fonic acid (HEPES), 100 mM KCl, 1 mM MgCl2, 0.05 %
Tween-20, pH 7.4.
Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) solu-
tion: 30 mM freshly prepared in Millipore water.

3 Methods

1. A 1 mM solution of thiol modified DNA (Table 1) is prepared

in Millipore water. To 30 μL of thiol-modified DNA, 2 μL of
3.1 Conjugation 1 M NaPi (pH 5.5) and 2 μL of TCEP solution are added to
of Thiol Modified activate the thiol moiety on the DNA. After 1 h of reaction at
DNA to Invertase room temperature, the unreacted TCEP is removed by wash-
(See Fig. 2) ing with Buffer B at least four times using an Amicon-10K. The
final volume should be approximately 30 μL.
2. A 20 mg/mL invertase solution is prepared in Buffer B. To
400 μL of 20 mg/mL invertase solution, 1 mg of Sulfo-SMCC
(Sigma-Aldrich, St. Louis, MO) is added (see Note 1). The mix-
ture is vortexed for 5 min and placed on a shaker for at least 1 h
at room temperature. After 1 h, any insoluble matter is removed

Table 1
DNA sequences used in the study

IFN-γ aptamer 5′-TGGGGTTGGTTGTGTTGGGTGTT

GTGTAAAAAAAAAAAAAACTA
CTCATCTGTGA-3′
Biotin-modified DNA 5′-Biotin-AAAAAAAAAAAATCA
CAGATGAGTAGT-3′
Thiol-modified DNA 5′-HS-AAAAAAAAAAACAACCA
ACCCCA-3′
Underlined sequence is the actual functional sequence of the aptamer
104 Tian Lan et al.

Fig. 2 Conjugation chemistry for covalently attachment of a thiol-modified DNA to invertase. The primary group
on the surface of invertase is first reacted with a bifunctional linker (Sulfo-SMCC), which contains an amine-
reactive N-hydroxysuccinimide ester moiety and a thiol-reactive maleimide group. The invertase-SMCC is then
reacted with the thiol-modified DNA to complete this step of conjugation

by briefly centrifuging the mixture (15,000 × g) and retaining

the supernatant. The unreacted Sulfo-SMCC in the supernatant
is removed by washing with Buffer B six times using Amicon-
100K. The final volume should be approximately 30 μL.
3. The thiol modified DNA and SMCC activated invertase
obtained from steps 1 and 2 are mixed and incubated at room
temperature for at least 24 h to form the DNA-modified inver-
tase. After the reaction, the unreacted thiol-modified DNA is
removed by washing with Buffer B six times using an
Amicon-100K.

3.2 Immobilization 1. The DNA-modified invertase is washed with Buffer C twice to

of Invertase exchange the buffer. The final concentration of invertase is
on Magnetic Beads around 20 mg/mL.
2. A 1 mL suspension of streptavidin-coated magnetic beads
(1 mg/mL) in a microcentrifuge tube is placed on the mag-
netic rack. The magnetic beads are allowed to collect at the
 BGM for Protein Detection 105

bottom of the tube for 2 min. The supernatant is pipetted off

and discarded. The magnetic beads are resuspended in 1 mL of
Buffer C.
3. A 12 μL aliquot of 0.5 mM biotin-modified DNA solution is
prepared in Millipore water. The biotin-modified DNA is
added to the streptavidin-coated magnetic beads from step 2
and mixed by vortexing for 30 min at room temperature. After
incubation, the magnetic beads are washed three times with
1 mL of Buffer C and resuspended in 1 mL of Buffer C.
4. A 12 μL aliquot of a 0.5 mM IFN-γ aptamer solution is pre-
pared in Millipore water. The aptamer solution is added to the
magnetic bead solution from step 3 and vortexed for 30 min
at room temperature. After the incubation, the magnetic beads
are washed three times with 1 mL of Buffer C and resuspended
in 1 mL of Buffer C
5. A 1 mL aliquot of DNA-modified invertase (containing around
5 mg/mL invertase) is added to the magnetic beads prepared
in the previous step and vortexed at room temperature for
30 min. Excess invertase is removed by washing the magnetic
beads with 1 mL of Buffer C five times (see Notes 2–5). These
DNA-invertase-coated magnetic beads are suspended in 1 mL
of Buffer C. For detection of IFN-γ, 60 μL of DNA- and inver-
tase-coated magnetic beads are used for each sample.

3.3 Detection 1. IFN-γ solutions of different concentrations are prepared in

of IFN-γ Using BGM Buffer C. Then 20 μL of an IFN-γ solution is added to
60 μL of DNA- and invertase-coated magnetic beads and
vortexed for 15 min at room temperature in a 1.5 mL
microcentrifuge tube.
2. After incubation with the IFN-γ solution, the tubes are placed
on a magnetic rack for 2 min to collect the magnetic beads at
the bottom of the tube. Then, 10 μL of supernatant is with-
drawn and added to 3.3 μL of 2 M sucrose in Buffer A at room
temperature.
3. After 30 min of incubation, 5 μL of the sucrose solution is mea-
sured for glucose content using an ACCU-CHEK Avia glucose
meter (see Note 6). To use the glucose meter, a strip is first inserted
into the glucose meter. After a beep, which indicates the strip is
ready for measurement, the sample can be applied directly to the
sampling area of the glucose strip. A numeric result will be dis-
played on the glucose meter after several seconds (see Notes 7–8).
4. For a negative control, different concentrations of human
serum albumin solutions are tested using the same protocol as
described in steps 1–3 above. For detection in 20 % human
serum, the same procedures in steps 1–3 may be used without
modification.
106 Tian Lan et al.

5. Representative data for IFN-γ detection in buffer C using BGM

is shown in Fig. 3a. An IFN-γ dependent glucose response can
be observed, and the glucose readout reached a maximum at
400 nM of IFN-γ (see Note 9). The limit of detection (LoD) is
calculated to be 2.6 nM (based on 3σ), which is similar to the
affinity of the aptamer [14]. Existing ELISA assays for IFN-γ
typically have a LoD in the sub-picomolar range [20]. The
higher LoD obtained from the BGM assay was mainly due to
lower affinity of the aptamer, as compared to antibodies.
Adoption of antibodies for the BGM assay can improve the
assay’s sensitivity to a similar level achieved by immunoassays.
For example, a BGM-based sandwich assay for Prostate Specific
Antigen has been demonstrated with a LoD at 400 pg/mL [9],
which is closer to the LoD of existing ELISA assay (typically in
the 10–100 pg/mL range) [21–23] (see Note 10). Negative
controls using human serum albumin result in negligibly low
readings from the BGM under otherwise identical conditions.
IFN-γ detection in 20 % human serum has also been tested
and a representative result is shown in Fig. 3b. A slight decrease
in the maximum glucose signal can be observed due to the
presence of serum components; however, the effect on the limit
of detection is small (3.4 nM in 20 % human serum). Signal
saturation was observed with a lower IFN-γ concentration
compared to the assay in buffer. This earlier signal saturation
can be a result of a weaker hybridization of the DNA and upon
IFN-γ binding to the aptamer, as release of the DNA-invertase
conjugate is easier. In addition, serum proteins may block the
nonspecific binding of IFN-γ to magnetic beads.

Fig. 3 Representative data of IFN-γ detection in buffer, with human serum albumin as negative control (a); and
in 20 % human serum (b)
 BGM for Protein Detection 107

4 Notes

1. Sulfo-SMCC may not be completely soluble in buffer during

the conjugation of thiol-modified DNA to invertase. This is
normal and does not affect the end result of the conjugation.
Using DMSO or DMF (for example, 20 % volume of buffer)
to increase the solubility of Sulfo-SMCC is not recommended
here, because yeast invertase is potentially deactivated in solu-
tions containing high contents of DMSO or DMF and Amicon
centrifuges are not able to tolerate solutions with such high
DMSO or DMF content.
2. After incubating the DNA-modified invertase with magnetic
beads, excess DNA modified invertase can be recycled for fur-
ther usage by using an Amicon-100K filter and washing with
Buffer B.
3. Invertase is an enzyme widely used in industrial applications.
It can withstand multiple cycles of heating and cooling, long
periods of elevated temperature (65 °C), and low pH. Its opti-
mal activity is observed in acidic pH and elevated temperatures.
4. To increase assay performance, Grade X invertase (Sigma
Aldrich, catalog # I4753) from Candida utilis can be used.
The same procedures described in the protocol can be directly
applied to the Grade X invertase. We have observed an activity
enhancement of at least 100 % when using Candida utilis
invertase versus yeast invertase.
5. To scale up production of the aptamer/invertase coated mag-
netic beads, the amount of materials can be increased, while
maintaining the same mass ratio.
6. The ACCU-CHEK Avia BGM requires a code chip to be
inserted to calibrate different batches of glucose strips—otherwise,
the BGM will not perform a measurement. This code chip is
obtained from each container of glucose strips. Other brands
of BGM may not require a code chip for operation.
7. The BGM will only report glucose levels between 10 mg/dL
and 600 mg/dL. If a measurement is outside this range, either
a “LO” (<10 mg/dL) or a “HI” (>600 mg/dL) signal will be
displayed, respectively.
8. Other brands of BGM may also be used; however, due to dif-
ferent chemistry used and calibration, glucose readings will be
different for the same sample when different BGM is used
from this study. This difference can be minimized if the buffer
conditions (pH, ionic strength, viscosity, etc.) are more similar
to those of human blood.
108 Tian Lan et al.

9. The assay has been designed to perform optimally at room tem-

perature (around 23 °C); significant temperature deviation can
result in faulty assay results.
10. The principles described in the protocol have been also applied
to antibody and nucleic acid hybridization-based assays for the
quantification of protein biomarkers and nucleic acid sequences
using a BGM [8, 9].

Acknowledgements

This material is based upon work supported by the US National

Institutes of Health (RDA035524A and RDK100213A and
ES16865) and National Science Foundation (IIP-1330934).

References
1. mHealth G. Device Listing (2012) A global list 10. Nie Z, Deiss F, Liu X, Akbulut O, Whitesides
of commercially available devices in the health- GM (2010) Integration of paper-based micro-
care sector, updated fortnightly 2012. http:// fluidic devices with commercial electrochemi-
www.gsma.com/connectedliving/wp-content/ cal readers. Lab Chip 10(22):3163–3169,
uploads/2012/04/22-November12_ PMCID: 3060706
MobileHealth_Device_Listing.pdf 11. Su J, Xu J, Chen Y, Xiang Y, Yuan R, Chai Y
2. Chin CD, Cheung YK, Laksanasopin T, (2012) Personal glucose sensor for point-of-
Modena MM, Chin SY, Sridhara AA et al care early cancer diagnosis. Chem Commun
(2013) Mobile device for disease diagnosis and (Camb) 48(55):6909–6911
data tracking in resource-limited settings. Clin 12. Yan L, Zhu Z, Zou Y, Huang Y, Liu D, Jia S
Chem 59(4):629–640 et al (2013) Target-responsive “sweet” hydro-
3. iBG STAR, 510(k) Substantial Equivalence gel with glucometer readout for portable and
Determination Decision Summary (510(k) quantitative detection of non-glucose targets.
number: k103544) J Am Chem Soc 135(10):3748–3751
4. iBGSTAR (2011) iBGStar® blood glucose 13. Boehm U, Klamp T, Groot M, Howard JC
meter. Sanofi (updated 2011; cited 2011 Dec (1997) Cellular responses to interferon-
1) http://www.bgstar.com/web/ibgstar gamma. Annu Rev Immunol 15:749–795
5. Telcare (2013) Telcare (updated 2013; cited 14. Liu Y, Kwa T, Revzin A (2012) Simultaneous
2013 09/09) https://www.telcare.com/ detection of cell-secreted TNF-alpha and IFN-
6. Hones J, Muller P, Surridge N (2008) The gamma using micropatterned aptamer-modified
technology behind glucose meters: test strips. electrodes. Biomaterials 33(30):7347–7355
Diabetes Technol Ther 10:S10–S26 15. Liu Y, Yan J, Howland MC, Kwa T, Revzin A
7. Xiang Y, Lu Y (2011) Using personal glucose (2011) Micropatterned aptasensors for con-
meters and functional DNA sensors to quan- tinuous monitoring of cytokine release from
tify a variety of analytical targets. Nature Chem human leukocytes. Anal Chem 83(21):8286–
3(9):697–703, PMCID: 3299819 8292, PMCID: 3235337
8. Xiang Y, Lu Y (2012) Using commercially 16. Lee PP, Ramanathan M, Hunt CA, Garovoy
available personal glucose meters for portable MR (1996) An oligonucleotide blocks
quantification of DNA. Anal Chem interferon-gamma signal transduction.
84(4):1975–1980, PMCID: 3302979 Transplantation 62(9):1297–1301
9. Xiang Y, Lu Y (2012) Portable and quantita- 17. Balasubramanian V, Nguyen LT,
tive detection of protein biomarkers and small Balasubramanian SV, Ramanathan M (1998)
molecular toxins using antibodies and ubiqui- Interferon-gamma-inhibitory oligodeoxynu-
tous personal glucose meters. Anal Chem cleotides alter the conformation of interferon-
84(9):4174–4178, PMCID: 3341531 gamma. Mol Pharmacol 53(5):926–932
 BGM for Protein Detection 109

18. Tuleuova N, Jones CN, Yan J, Ramanculov 21. Acevedo B, Perera Y, Ruiz M, Rojas G, Benitez
E, Yokobayashi Y, Revzin A (2010) J, Ayala M et al (2002) Development and vali-
Development of an aptamer beacon for dation of a quantitative ELISA for the mea-
detection of interferon-gamma. Anal Chem surement of PSA concentration. Clin Chim
82(5):1851–1857 Acta 317(1–2):55–63
19. Stoltenburg R, Reinemann C, Strehlitz B 22. Vessella RL (1993) Trends in immunoassays of
(2007) SELEX–a (r)evolutionary method to prostate-specific antigen: serum complexes
generate high-affinity nucleic acid ligands. and ultrasensitivity. Clin Chem 39(10):
Biomol Eng 24(4):381–403 2035–2039
20. Wilson AB, McHugh SM, Deighton J, Ewan 23. Matsumoto K, Konishi N, Hiasa Y, Kimura E,
PW, Lachmann PJ (1993) A competitive inhi- Takahashi Y, Shinohara K et al (1999) A highly
bition ELISA for the quantification of human sensitive enzyme-linked immunoassay for
interferon-gamma. J Immunol Methods serum free prostate specific antigen (f-PSA).
162(2):247–255 Clin Chim Acta 281(1–2):57–69
 Chapter 8

Microchip ELISA Coupled with Cell Phone to Detect Ovarian

Cancer HE4 Biomarker in Urine
ShuQi Wang, Ragip Akbas, and Utkan Demirci

Abstract
Ovarian cancer is a leading cause of death from gynecologic cancers in the USA, and early diagnosis can
potentially increase 5-year survival rate. Detection of biomarkers derived from hyperplasia of epithelial tis-
sue by enzyme-linked immunosorbent assay (ELISA) proves to be a practical way of early diagnosis of
ovarian cancer. However, ELISA is commonly performed in a laboratory setting, and it cannot be used in
a clinical setting for on-site consultation. We have shown a microchip ELISA that detects HE4, an ovarian
cancer biomarker, from urine using a cell phone integrated with a mobile application for imaging and data
analysis. In microchip ELISA, HE4 from urine was first absorbed on the surface; the primary and second-
ary antibodies were subsequently anchored on the surface via immuno-reaction; and addition of substrate
led to color development because of enzymatic labeling. The microchip after color development was
imaged using a cell phone, and the color intensity was analyzed by an integrated mobile application. By
comparing with an ELISA standard curve, the concentration of HE4 was reported on the cell phone
screen. The presented microchip ELISA coupled with a cell phone is portable as opposed to traditional
ELISA, and this method can facilitate the detection of ovarian cancer at the point-of-care (POC).

Key words Microchip ELISA • Mobile application • Diagnostics • HE4 biomarker

1 Introduction

Ovarian cancer patients often present with advanced disease (stage

III/IV) in the clinic, which results in a 5-year survival rate of 33 %
compared to 90 % at stage I [1]. To maximize the clinical benefit, a
variety of approaches such as biopsy, medical imaging, and genetic
analysis have been used to achieve early diagnosis. Practically, these
approaches cannot be used for routine screening in the clinic due to
the need for sophisticated instruments and well-trained operators. In
contrast, detection of serum CA125 and HE4 by enzyme-linked
immunosorbent assay (ELISA) has shown moderate success in
detection of ovarian cancer. Serum CA125 ELISA has a sensitivity of
72 % at specificity 95 % [2]; HE4 can be detected at a sensitivity
of 86.6 and 89.0 % for patients at stage I/II and III/IV, respectively.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_8, © Springer Science+Business Media New York 2015

111
112 ShuQi Wang et al.

However, these two ELISA methods require a 96-well plate washer

and a spectrophotometer that inherently limited their utility in
clinical diagnosis outside a laboratory setting. Therefore, there is
an unmet need to develop a deliverable approach to detect ovarian
cancer biomarkers inexpensively at the point-of-care (POC).
Considering the challenges associated with traditional ELISA
[3, 4], we developed microchip ELISA coupled with a cell phone
[5], which eliminates the reliance on the use of a plate washer and a
plate reader, offering an inexpensive and easy-to-use approach for
POC testing (Fig. 1). The detection of HE4 from urine was based
on an indirect ELISA. First, a urine sample was added into a micro-
channel, and HE4 was immobilized on the surface via passive
adsorption. Second, a wash buffer was flowed into the microchan-
nel to remove unbound protein. Third, an antibody (primary anti-
body) specific for HE4 was added, which was followed by a wash.

Fig. 1 Workflow of microchip ELISA coupled with cell phone detection (Reproduced from ref. 5 with permis-
sion from The Royal Society of Chemistry). (a) 100 µL of urine sample containing HE4 was pipetted into a
microchannel. (b) Formation of an indirect ELISA immune-complex on a microchip. (c) Blue color with varying
intensities was developed in microchannels and imaged using a cell phone. (d) The blue color (marked with
rectangles) developed in microchannels was auto-recognized by a mobile application, and R pixel values were
reported. The R pixel value of background was used for normalization. The concentration of HE4 in urine was
calculated with a reference to the standard curve and then reported on the screen
 Microchip ELISA Coupled with Cell Phone to Detect Ovarian Cancer HE4 Biomarker… 113

Fourth, a horseradish peroxidase (HRP) labeled secondary antibody

targeting the primary antibody was added, and the unbound sec-
ondary antibody was again washed away. Fifth, addition of TMB, a
substrate of HRP, led to blue color development. Lastly, the color
intensity was imaged using a cell phone, and the concentration of
HE4 was analyzed using a mobile application, enabling immediate
result report without referring to peripheral equipment.

2 Materials

2.1 Microchip 1. Polymethyl-methacrylate (PMMA) (McMaster Carr, Atlanta, GA).

2. Double-sided adhesive film (iTapstore, Scotch Plains, NJ).
3. Laser cutter (VersaLaser™, Scottsdale AZ).
4. Polystyrene petri dish (BD Biosciences, San Jose, California).

2.2 ELISA 1. Urinary peptides (CSLPNDKEGSCPQVNINFPQL) derived

from human protein HE4 were synthesized and used to gener-
ate rabbit polyclonal antibodies (21st Century Biochemicals,
Inc. Marlborough, MA).
2. Rabbit polyclonal antibody (21st Century Biochemicals, Inc.
Marlborough, MA).
3. ELISA washing buffer, 50 mM Tris–HCl, 150 mM NaCl, and
0.05 % Tween-20.
4. 3 % bovine serum albumin (BSA, m/v, Fischer Scientific,
Pittsburgh, PA).
5. Anti-rabbit-HRP (1 mg/mL, Abcam, Cambridge, MA).
6. One-Step ultra TMB (Thermo Fisher Scientific Inc., Waltham,
MA).
7. Stop solution (0.5 M H2SO4).
8. Microplate reader (BioTek, Winooski, VT).

2.3 Cell Phone 1. Cell phone (Sony Ericson i790).

Imaging 2. 5 mm High-Brightness White LED (Cat. 276–017, Raidoshack,
Fort Worth, TX).
3. Black acrylic box (Entegris, Penang, Malaysia).

3 Methods

3.1 Microchip 1. A microchip with three straight microchannels was designed

Design and (Fig. 2). Laser cutting was used to fabricate microchips as we
Fabrication previously published [6, 7].
2. Two PMMA layers with dimensions of 3 × 24 × 40 mm3 were
cut using a laser cutter (see Note 1).
114 ShuQi Wang et al.

Fig. 2 (a) Assembly of a microchip with three straight microchannels on a polystyrene petri dish. (b) Actual
image of an assembled microchip with color development. Adapted from ref. [5] with permission from The
Royal Society of Chemistry

3. A circular inlet and outlet with a diameter of 0.375 mm were

cut on the top of the PMMA layer.
4. The bottom PMMA layer had three straight microchannels
with dimensions of 4 × 7.5 mm2.
5. Two layers of double-sided adhesive film with dimensions of
24 × 40 mm2 were also cut using the laser cutter (see Note 2).
6. Two layers of PMMA were bonded via one layer of double-
sided adhesive film.
7. At the bottom layer of PMMA, the second layer of double-
sided adhesive film was attached.
8. The device was assembled onto a polystyrene petri dish, form-
ing three microchannels.

3.2 Construction To avoid the interference of external light, we built a cell phone-
of a Black Box for Cell imaging instrument (Fig. 3). This setup consisted of one black
Phone Imaging acrylic box, one imaging stage, and an imaging hole on the top.
1. To provide internal light, two LEDs were installed under the
imaging stage.
2. One layer of PMMA (1.5 mm thick) was mounted as the top of
the stage.
3. One piece of white print paper was attached to the bottom of
the PMMA layer to provide a white background.

3.3 Microchip ELISA 1. Pure HE4 peptide antigen was serially twofold diluted in
sodium bicarbonate (0.1 M, pH 9.7) to obtain quantification
standards (1,250, 625.0, 312.5, 156.3, 78.1, 39.1, and
19.5 ng/mL).
2. 100 μL of each HE4 quantification standard was injected into
a microchannel and incubated at room temperature for 1 h
(see Note 3).
 Microchip ELISA Coupled with Cell Phone to Detect Ovarian Cancer HE4 Biomarker… 115

Fig. 3 (a) Design of a black box for cell phone imaging. (b) Actual image of the black box. Adapted from ref.
[5] with permission from The Royal Society of Chemistry

3. 100 μL of urine sample was also manually pipetted into a

different microchannel and incubated at room temperature
for 1 h (see Note 3).
4. The microchannels were washed three times by pipetting
200 μL of ELISA washing buffer. The wash buffer was col-
lected at the outlets using Kimwipes™ Delicate Task Wipers.
5. Microchannels were blocked with 100 μL of 3 % BSA at 37 °C
for 1 h (see Note 3).
6. The wash step was repeated as in step 4.
7. Anti-HE4-rabbit primary antibody (0.61 mg/mL) was diluted
in 1:50,000 in 3 % BSA blocking buffer, and 100 μL of this
secondary antibody solution was loaded into the microchan-
nels. The incubation took place at 37 °C for 1 h (see Note 3).
8. The wash step was repeated as in step 4.
9. Anti-rabbit-HRP (1 mg/mL), the secondary antibody, was
diluted in 1:3,000 in Tris-buffered saline and Tween-20
(0.05 %). 100 μL of this secondary antibody solution was
pipetted into microchannels and incubated at 37 °C for 1 h
(see Note 3).
10. The wash step was repeated as in step 4.
11. 100 μL of one-Step ultra TMB was flowed into microchan-
nels, and incubated at room temperature in the dark for 9 min.
12. A blue color was developed in microchannels, and the solution
was gently mixed (see Note 4).
13. The microchip after color development was rapidly imaged
using a cell phone (see Note 5).
14. To compare with microplate ELISA, the blue solution gener-
ated on microchips was collected and analyzed using a spec-
trophotometer (see Note 6) (Fig. 4a).
116 ShuQi Wang et al.

Fig. 4 Validation of on-chip HE4 ELISA compared to conventional 96-well microplate ELISA (Adapted from
ref. 5 with permission from The Royal Society of Chemistry). (a) The standard curve obtained from microchip
ELISA was compared with that obtained from a 96-well microplate ELISA. The absorbance of the resultant
color solution obtained from both methods was measured using a spectrophotometer (BioTek, Winooski, VT) at
a wavelength of 450 nm. (b) The standard curve of HE4 microchip ELISA. R pixel values were measured using
a mobile application, which was correlated with the color intensity obtained at varying concentrations. Data
are presented as mean ± standard deviation (n = 8)

3.4 Microplate ELISA As a standard method, conventional 96-well microplate ELISA was
for Detection of HE4 also performed. The testing procedure was the same except for the
follow steps:
1. The washing steps were performed using a plate-washer.
2. The incubation of TMB in microwells was 15 min prior to addi-
tion of 100 μL of stop solution.
3. The readout of absorption (optical density, OD) was measured
using a microplate reader at a wavelength of 450 nm.

3.5 Quantitative 1. The microchip with blue color development at the end of 9-min
Image Processing by incubation was placed in the black box as shown in Fig. 3.
Matlab 2. A cell phone (Sony Ericson i790) with a 3.2 megapixel camera
was used to image the on-chip color development through the
hole on top of the black box.
3. The images were exported to a computer, and the manually
cropped image regions with color development were analyzed
using a customized MATLAB (MathWorks, Natick, MA) code
(Table 1). Red, green and blue pixel values were reported within
seconds as mean value ± standard deviation.
4. Red (R) pixel values were used for data analysis, since they cor-
related well with the concentrations of HE4 loaded to micro-
channels (Fig. 4b).
 Microchip ELISA Coupled with Cell Phone to Detect Ovarian Cancer HE4 Biomarker… 117

Table 1
The Matlab code for image analysis

for i=1:34
str=strcat(int2str(i),'.jpeg');
I=imread(str);
Ired=I;
Igreen=I;
Iblue=I;
Ired(:,:,2)=[];
Ired(:,:,2)=[];
Igreen(:,:,1)=[];
Igreen(:,:,2)=[];
Iblue(:,:,1)=[];
Iblue(:,:,1)=[];
r=mean2(Ired);
g=mean2(Igreen);
b=mean2(Iblue);
std=std2(Iblue);
rgb(i,1)=r;
rgb(i,2)=g;
rgb(i,3)=b;
rgb(i,4)= std;
end
rgb
xlswrite('rgb.xls',rgb);

5. Red (R) pixel values obtained from quantification standards

were used to construct a standard curve, which was used to
quantify HE4 in unknown urine samples.

3.6 Quantification Microchip ELISA was first validated by comparing to 96-well plate
of Clinical Samples ELISA. The blue color solution from microchannels was trans-
with Unknown ferred to a 96-well plate using a pipet. The optical density of each
Concentrations of HE4 well was measured using a spectrophotometer and the readouts
were plotted as shown in Fig. 4a. The result indicated that the
microchip ELISA can be used for HE4 quantification. To elimi-
nate the use of a spectrophotometer, we developed a cell-phone
based imaging and analysis system. By measuring the red pixel
118 ShuQi Wang et al.

Fig. 5 Mobile application for image processing and data report. (a) Automatic selection of channel region for
image processing. (b) Batch processing to report R pixel values versus the concentrations of HE4 (quantifica-
tion standards). (c) Plot of standard curve for microchip ELISA

value and substituting it into the inverse of the standard curve

function (x = exp((y − 0.3547)/0.08861) in this case) (Fig. 4b), we
can obtain the concentration of HE4 in clinical samples.

3.7 Quantitative A mobile application was developed to facilitate image processing

Image Processing by and to report the levels of HE4 on the cell phone screen (Fig. 5).
a Mobile Application The mobile application was written to run on a Windows Phone 7
and 8 operating systems, and the cell phone code algorithm can be
adapted for use in other mobile operating systems. To obtain the
quantification data, the following steps are carried out on the cell
phone.
1. Import new images or previously saved images for processing.
2. Calculate automatically rectangular regions within the channels
for data analysis.
3. Convert color intensity into R values.
4. Normalize R values using the background.
5. Calculate, store, and edit the standard curve regression param-
eters (see Note 7).
6. Display standard curve.
7. Load the images of clinical samples and report the concentra-
tion of HE4 (ng/mL) (see Notes 8–10).
The mobile application can yield R pixel values comparable to
the desktop application (Table 2).
Microchip ELISA Coupled with Cell Phone to Detect Ovarian Cancer HE4 Biomarker… 119

Table 2
Comparison of Red pixel values obtained by MATLAB and mobile application

HE4 concentration Average R values Average R values

(ng/mL) by MATLAB by mobile application
1,250.0 2.662 2.165
625.0 14.737 14.708
312.5 35.597 35.610
156.3 58.823 57.865
78.1 69.038 68.213
39.1 82.520 81.805
19.5 93.251 92.723
0.0 121.014 121.090

Note: The detection limit was 19.5 ng/mL. For each standard point, we performed
three replicates. Reproduced from ref. [5] with permission from The Royal Society of
Chemistry

Currently, there is no clearly defined clinical cutoff for the level

of HE4 concentrations in urine in determining the status of ovar-
ian cancer. The common strategy to evaluate the clinical utility of
HE4 is to compare the level of HE4 between ovarian cancer
patients and health individuals [5, 8]. It has been reported that
HE4 can be used to detect ovarian cancer patients at a sensitivity of
86.6 and 89.0 % at stages I/II and III/IV, respectively [8]. In our
study, the microchip ELISA coupled with cell phone detection
showed a sensitivity of 89.5 % and a specificity of 90 % in differen-
tiating ovarian cancer patients from healthy individuals [5]. Our
method represents the synergistic efforts from clinicians and engi-
neers to provide a POC diagnostic tool for early detection of ovar-
ian cancer. Considering the simplicity of microchip ELISA coupled
with cell phone detection, this method can potentially be imple-
mented in decentralized laboratory settings for cancer screening by
health care workers with minimum training. Nevertheless, we fur-
ther streamlined the microchip ELISA approach using a micro-a-
fluidic strategy for an automated sample-in-result-out capability
within 10 min [9]. We believe this technology advance can signifi-
cantly improve the clinical acceptability of microchip ELISA for
disease diagnosis and treatment monitoring at the POC. However,
there is an unmet need to discover and validate reliable biomarkers,
which remains the major obstacle for establishing an ovarian
cancer-screening program due to the low prevalence of ovarian
cancer [10, 11]. Therefore, the integration of vigorously validated
biomarker panels in the microchip ELISA platform holds great
promise for ovarian cancer screening with an optimal sensitivity
and specificity.
120 ShuQi Wang et al.

4 Notes

1. The power of the laser cutter used for cutting PMMA was
75 % at a speed of 7 %.
2. The power of the laser cutter used for cutting double-sided
adhesive was 10 % at a speed of 6 %.
3. Microchips were wrapped with Parafilm during incubation
steps wherever applied to avoid evaporation.
4. Homogeneous color distribution was achieved by pipetting
the blue solution inside the microchannel back and forth for a
couple of times. Generation of air bubbles should be avoided.
5. Homogeneous color distribution was obtained prior to cell
phone imaging.
6. One hundred microliters of sample or reagent was a standard
volume as a comparison to the control method, i.e., 96-well
plate ELISA. Compared to smaller volumes, this volume was
more tolerant to sampling errors and allowed us to collect
more reliable blue color images.
7. The quantification standards were serially diluted in sodium
bicarbonate to construct the standard curve for quantifying
the concentration of HE4 in urine. Although this sample pro-
cessing may cause a systemic variation in terms of absolute
quantification, it would not affect the comparison results
between healthy and cancerous urine samples using a defined
threshold. Nevertheless, a pooled negative urine sample would
be ideal to dilute quantification standards to reduce the varia-
tion of quantification.
8. For validation, the blue solution was transferred to an
Eppendorf tube using a pipette. The blue solution was stopped
using 100 μL of 1 M H2SO4 and subsequently measured using
a microplate reader at a wavelength of 450 nm.
9. Standard images must to be loaded in an order starting from
higher concentrations to lower ones for regions to be assigned
to the right concentration values.
10. The application reports R values and concentrations for each
selected image.

Acknowledgments

We would like to acknowledge the W.H. Coulter Foundation

Young Investigator Award, RO1 A1081534, R21 AI087107. This
work was supported by the Center for Integration of Medicine and
Innovative Technology (CIMIT) under US Army Medical Research
 Microchip ELISA Coupled with Cell Phone to Detect Ovarian Cancer HE4 Biomarker… 121

Acquisition Activity Cooperative Agreements DAMD17-02-2-

0006, W81XWH-07-2-0011, and W81XWH-09-2-0001. And
this work was made possible by a research grant that was awarded
and administered by the US Army Medical Research & Materiel
Command (USAMRMC) and the Telemedicine & Advanced
Technology Research Center (TATRC), at Fort Detrick, MD. We
also acknowledge NIH U01 HL065899-08.
Utkan Demirci (UD) is a founder of, and has an equity interest
in, DXNow, a company that is developing microfluidic and imag-
ing technologies for point-of-care diagnostic solutions. UD’s inter-
ests were reviewed and are managed by the Brigham and Women’s
Hospital and Partners HealthCare in accordance with their conflict
of interest policies.

References
1. Clarke-Pearson DL (2009) Clinical practice. 7. Moon S et al (2009) Integrating microfluid-
Screening for ovarian cancer. N Engl J Med ics and lensless imaging for point-of-care
361(2):170–177 testing. Biosens Bioelectron 24(11):
2. Visintin I et al (2008) Diagnostic markers for 3208–3214
early detection of ovarian cancer. Clin Cancer 8. Hellstrom I et al (2010) Detection of the HE4
Res 14(4):1065–1072 protein in urine as a biomarker for ovarian neo-
3. Wang S, Xu F, Demirci U (2010) Advances in plasms. Cancer Lett 296(1):43–48
developing HIV-1 viral load assays for resource- 9. Wang S et al (2014) Micro-a-fluidics ELISA
limited settings. Biotechnol Adv 28(6):770–781 for rapid CD4 cell count at the point-of-care.
4. Wang S et al (2013) Point-of-care assays for Sci Rep 4:3796
tuberculosis: role of nanotechnology/micro- 10. Cramer DW et al (2011) Ovarian cancer bio-
fluidics. Biotechnol Adv 31(4):438–449 marker performance in prostate, lung, colorec-
5. Wang S et al (2011) Integration of cell phone tal, and ovarian cancer screening trial
imaging with microchip ELISA to detect ovar- specimens. Cancer Prev Res (Phila) 4(3):
ian cancer HE4 biomarker in urine at the 365–374
point-of-care. Lab Chip 11(20):3411–3418 11. Zhu CS et al (2011) A framework for evaluat-
6. Kim YG et al (2009) Quantum dot-based HIV ing biomarkers for early detection: validation
capture and imaging in a microfluidic channel. of biomarker panels for ovarian cancer. Cancer
Biosens Bioelectron 25(1):253–258 Prev Res (Phila) 4(3):375–383
 Chapter 9

Point-of-Care Rare Cell Cancer Diagnostics

David Issadore

Abstract
The sparse cells that are shed from tumors into peripheral circulation are an increasingly promising resource
for noninvasive monitoring of cancer progression, early diagnosis of disease, and serve as a tool for improv-
ing our understanding of cancer metastasis. However, the extremely sparse concentration of circulating
tumor cells (CTCs) in blood (~1–100 CTC in 7.5 mL of blood) as well as their heterogeneous biomarker
expression has limited their detection using conventional laboratory techniques. To overcome these chal-
lenges, we have developed a microfluidic chip-based micro-Hall detector (μHD), which can directly mea-
sure single, immunomagnetically tagged cells in whole blood. The μHD can detect individual cells even in
the presence of vast numbers of blood cells and unbound reactants, and does not require any washing or
purification steps. Furthermore, this cost-effective, single-cell analytical technique is well suited for minia-
turization into a mobile platform for low-cost point-of-care use. In this chapter, we describe the methodol-
ogy used to design, fabricate, and apply these chips to cancer diagnostics.

Key words Hybrid microfluidic, Microelectronic chips, BioMEMS, Sensors, Microfluidics, Magnetic
nanomaterials

1 Introduction

Tumor cells are often localized in difficult to access parts of the

body, making molecular diagnostics on cancer cells reliant on inva-
sive procedures. The analyses of the sparse molecular markers that
are shed from tumor cells into peripheral circulation have great
potential to address this challenge [1]. These markers, including
soluble proteins and nucleic acids [2, 3] as well as circulating tumor
cells (CTCs) [1] (Table 1), have been shown to contain valuable
information on the molecular state of the cancer that can be used
for disease diagnostics and monitoring [1, 4, 5].
Engineers have devised many ingenious strategies to isolate
and measure both rare CTCs and sparse soluble molecules in
blood. However, fundamental technical challenges of these detection
modalities have impeded their clinical application. The detection
of soluble protein based biomarkers, such as prostate specific
antigen (PSA), has been limited by issues of specificity [6].

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_9, © Springer Science+Business Media New York 2015

123
124 David Issadore

Table 1
Comparison of soluble proteins, circulating nucleic acid, and circulating tumor
cells (CTCs) for cancer monitoring

Because the biogenesis of soluble protein found in blood cannot

be determined, diagnostics based on protein detection suffer from
high false-positive rates. Circulating genetic material can have
improved specificity to proteins, but are limited by low concentra-
tions due to their short half-life in the blood (~10 min) [2].
Circulating tumor cells (CTCs) carry with them a great deal of
information about the cancer that they come from—protein
expression, RNA expression, mutations—and as such their biogen-
esis can be determined with high specificity [1]. Additionally,
because each CTC carries with it multiple biomarkers, their mea-
surement allows cellular heterogeneity to be resolved and thus
reveals a richer picture of the molecular state of the cancer than is
not possible with soluble protein detection [1, 7, 8]. However,
using conventional techniques, the detection of CTCs has been
limited by their sparse concentrations in blood (~1–100 CTC in
7.5 mL of blood), resulting in long run-times, excessive consump-
tion of valuable clinical samples, and Poisson counting error [1].
A promising methodology to overcome these challenges is
immunomagnetic detection, in which molecular biomarkers iden-
tifying CTCs are labeled using magnetic particles. The inherently
negligible magnetic susceptibility of biological material enables
CTCs to be isolated and detected with little interference from the
complex background regardless of pH, salinity, or turbidity. High-
contrast detection can thus be performed with minimum sample
preparation and short incubation times, which avoids sample loss
and simplifies clinical use. Furthermore, magnetic sensing and
sorting is amenable to miniaturization and integration into mono-
lithic chips for portable use [9].
 Rare Cell Cancer Diagnostics 125

Fig. 1 Summary of our detection method. (a) Our hybrid chip consists of a GaAs substrate with microfabricated
Hall sensors. Built directly on top of the semiconductor chip is a microfluidic network to controllably deliver
cells to the sensors. The microfluidics are fabricated using soft-lithography and are permanently bonded to the
GaAs. (b) Circulating tumor cells are magnetically labeled based on a panel of relevant cancer biomarkers
(EGFR, HER2/neu, etc.) and magnetic nanoparticles (MNPs). The cells are passed one-by-one via hydrody-
namic flow structures over μHall sensors for highly sensitive detection

We harness the small feature size and high sensitivity of

microelectronics and combine it with the biocompatibility of micro-
fluidics to create a highly sensitive and miniaturized diagnostic
(Fig. 1a). Our proposed design consists of an array of sensors
fabricated on a semiconductor substrate with a microfluidic net-
work built directly on top of it (Fig. 1b). The microfluidics serve
as an interface between the sensors and the sample, preparing the
complex sample for measurement and guiding potential CTCs
over the array of sensors for detection. Because the chip detects
cells individually, it is able to ignore biological objects with inad-
vertently attached MNPs using software-defined gating, as is
done in fluorescence flow cytometry.
Additionally, the chip can ignore unbound MNPs, eliminating
the need to wash away unbound reagents [10]. The μHall sensors
measure the magnetic moment arising from a very small volume
(∼10 pL) directly above the sensor. In an unwashed sample con-
taining a particle concentration of 108/mL, there is on average less
than 1 unbound MNP in the detection volume at any given time.
Targeted tumor cells, in contrast, can have 105–107 MNPs per cell,
creating a much larger signal than the background.
Our micro-Hall Detector (μHD) was specifically designed to
detect magnetically labeled cellular targets. On our chip, cells
that have been labeled with MNPs are magnetized with an exter-
nal permanent magnet, and assume a magnetic moment m. The
magnetic moment of each cell m is directly proportional to both
the number of biomarkers on the cell N and the magnetic
moment of the MNPs mp (m = N × mp). The demagnetization
field of each individual cell is measured as it passes over an array
microfabricated Hall sensors (Fig. 1b). The total number of bio-
markers N for each cell can thus be calculated by the measurement
126 David Issadore

of the demagnetization field from individual cells as they are

delivered one-by-one by a microfluidic network built over the
micro-Hall sensors.
Magnetic sensors based on both the giant magnetoresistance
(GMR) and the Hall effect have been developed to sensitively
detect magnetically labeled biomarkers in complex biological
media. The Hall effect refers to the production of a voltage
VH = RH × I × B⟂ across an electrical conductor proportional to the
magnetic field normal to the plane of the conductor B⟂, its bias
current I, and its characteristic Hall resistance RH [11]. Sensors
that use GMR have inherently greater sensitivity than the Hall
effect at low fields [12]. However, due to the nonlinear response
of GMR, these sensors saturate at large fields (B > 0.01 T). We
selected the Hall effect for cell detection, rather than magnetore-
sistance because it allows us to apply the large magnetic fields
(B > 0.1T) necessary to fully magnetize the superparamagnetic
MNPs [13], without saturating the sensors. Additionally, due to
the linear response of the μHall sensors, cells with nonspecifically
bound MNPs can be accurately excluded by gating the measured
signals above a particular threshold value. Additionally, Hall sen-
sors are fully compatible with standard semiconductor processing,
which enables low-cost production and integration with auxiliary
electronics onto a monolithic chip. Because of this integration, the
entire detection system can be realized as a self-contained, cost-
effective lab-on-chip for mobile health applications.
The detection of rare cells using magnetic nanomaterials and
micro-Hall detectors is particularly well suited for mobile diagnos-
tics. Because the diagnostic is fully automated and the sample
requires minimal processing, it can be used in practical clinical
environments by minimally trained personnel. Additionally,
because the chip uses sensors built on semiconductor substrates
with electronic output, the chip can be easily interfaced with
mobile devices for remote clinical data sharing and epidemiological
surveillance [10, 14]. The microchip’s input is a raw clinical sample
and its output is quantitative, electronic data that can be immedi-
ately transmitted, shared, and stored [4, 15].
We have described in two recent papers the development of
these μHall chips for the detection of CTCs [4] and bacteria [15].
In the sections below, we focus with greater depth than has previ-
ously been published on the methodology used to design and fab-
ricate these chips.

2 Materials

2.1 μHall Chip The sensors were built on a pseudomorphic high electron mobility
Farbrication GaAs wafer using standard semiconductor processing (Fig. 2a).
A mesa was etched onto the GaAs wafer heterostructure (Fig. 2b).
The mesa was defined using photolithography and followed by an
 Rare Cell Cancer Diagnostics 127

Fig. 2 Summary of our fabrication method. The cartoons on the left are side-view
schematics of the chip in the various stages of the fabrication (a–g). The images
on the right are schematics of the top-view. (g) A micrograph of the finished
micro-Hall chip. The scale bar is 250 μm. (h) A cross-sectional schematic of the
instrumentation that connect to the μHall chip (i) A photograph of the slot-
connector/magnet holder, where the μHD chip sits. A shielded ribbon cable
connects the output of the μHD to the custom printed circuit board
128 David Issadore

anisotropic reactive ion etch. Electrodes were photolithographically

patterned and metal layers were deposited in the following order:
Nickel/Ni (50 Å), Gold/Au (50 Å), Germanium/Ge (250 Å), Au
(400 Å), Ni (100 Å), and Au (400 Å) (Fig. 2c). To make ohmic
contact to the two-dimensional electron gas, an eutectic alloy was
formed. To create this alloy, the electrodes were subsequently
annealed at 480 °C for 90 s using a rapid thermal annealer (Fig. 2d).
The Hall sensors were protected from the biological solutions
using three layers of oxide. The first layer was Al2O3 layer (30 nm)
grown by atomic layer deposition (ALD) to ensure conformal
coverage (Fig. 2e). Next, a Si3N4 layer (100 nm) was grown by
chemical vapor deposition (CVD) to protect against the diffusion
of ions. Finally, an SiO2 layer (100 nm) was grown by CVD to
form a layer that could be activated to make a permanent bond
with the PDMS microfluidics (Fig. 2f).
The PDMS based microfluidics were fabricated using soft
lithography. A two layer SU-8 (MicroChem) mold was fabricated
with two-step photolithography. The PDMS was poured onto the
mold and cured at 65 °C for 3 h. The PDMS microfluidics and
GaAs chip were treated with O2 plasma, aligned using a modified
mask aligner (ABM), and then permanently bonded (Fig. 2g).

2.2 Electronics We electrically characterized the bandwidth, noise, and sensitivity

of the μHall sensors that we fabricated. The magnetic field sensi-
tivity of the Hall sensors (78 Ω/T) was measured using a known
magnetic field. The magnetic field was created using a water
cooled electromagnet (HV-4H, Walker LDJ Scientific) and was
independently measured using a commercial magnetometer
(THM 7025, MetroLab) (Fig. 3a). The bandwidth 150 MHz and
input-referred noise 1.3 nV/√Hz were measured with a spec-
trum analyzer (Fig. 3b).
The μHall sensors were connected to custom electronics that
drive the μHall sensors and condition their output before they are
sent to a mobile device or computer for analysis. The electronic
scheme to read out the μHall sensors is shown in Fig. 3c. Howland
voltage to current converters were used to bias the Hall sensors
with a current from −10 to 10 mA. The Hall sensors were AC cou-
pled to the preamplifier through a high pass filter, with a cutoff
frequency of ƒ3dB = 500 Hz. AC coupling allows the large signal
that arises from the static field (B ~ 0.5T) be completely removed,
such that the much smaller signal that arises from a passing cell to
be resolved. The preamplifier and amplifier (THS4131, Texas
Instruments) had a gain of 30 × 30 (900). The conditioned signal
was digitized (PCI6133, National Instruments) and sent to either
a computer or a mobile device for digital analysis. The chip fits into
a slot connector, which holds a 2″ diameter, 1″ thick NdFeB mag-
net underneath it (K and J magnets) (Fig. 2h, i).
While the μHD was designed with mobile diagnostics in mind,
several improvements to the peripheral instrumentation are needed
 Rare Cell Cancer Diagnostics 129

Fig. 3 Electronics to process the micro-Hall signal. (a) The hall resistance R was measured versus the applied
magnetic field B, and a sensitivity of 78 Ω/T was calculated. (b) The input-referred noise of the μHall sensors
as measured with a spectrum analyzer. (c) A schematic of the electronics to condition the μHall signal prior to
software analysis. Each pair of μHall sensors are driven with a current source. The differential output of each
μHall sensor is AC coupled to a two-stage amplifier (gain = 30 × 30) before analog-to-digital conversion
ADC. Once digitized the signal is sent to either a computer or mobile device for software analysis

to make the platform fully mobile. One important design choice

for the μHD, which make it well suited for mobile applications, is
that it only requires a single source of negative pressure (at the
outlet). Because this pressure source does not need to be stable,
since the flow focusing is sensitive only to the relative velocities
between the sheath and sample flow, the flow can be driven by an
inexpensive spring-loaded syringe or vacuum pack [16].
Additionally, rather than the PCI card based analog-to-digital-
converted used in this demonstration, a USB based platform would
allow the PC used in this study to be replaced with a tablet or
smartphone. In previous work, we demonstrated that a similar
device could be interfaced with an iPhone and an iPad using a
custom programmed app [14].

2.3 Magnetic The μHD is designed to detect magnetically labeled cellular tar-
Nanomaterials gets. By targeting cells with MNPs and subjecting them to an
external magnetic field B, each cell acquired a magnetic moment m
that is directly proportional to both the number of biomarkers
N and the magnetic moment of the MNPs mp (m = N × mp).
The induced magnetic fields of these cells are then measured using
the microfabricated Hall sensors.
130 David Issadore

To efficiently label cellular biomarkers with MNPs, we used a

two-step bio-orthogonal procedure that is based on the cycloaddi-
tion between a 1,2,4,5-tetrazine (Tz) and a trans-cyclooctene
(TCO) [17]. In this approach, cells were first targeted with TCO-
conjugated antibodies. Once bound to their cellular targets, the
TCO antibodies then acted as scaffolds onto which multiple
Tz-modified MNPs could be coupled. This technique results
in much higher (>300 %) nanoparticle loading onto target cells
(~106 MNPs per cell) than direct antibody–MNP conjugates [17].
Amine-terminated, cross-linked iron oxides (CLIOs) were
prepared as described previously [17]. These CLIOs nanoparticles
have a core size of 7 nm and a hydrodynamic diameter of 30 nm.
Modification with 2,5-dioxopyrrolidin-1-yl 5-(4-(1,2,4,5-
tetrazin-3-yl)benzylamino)-5-oxopentanoate (TZ-NHS) create
CLIO-TZ. Briefly, excess TZ-NHS was reacted with amino-CLIO
in phosphate-buffered saline (PBS) containing 0.1 M sodium
bicarbonate, for 3 h at room temperature. TZ-CLIO was purified
using Sephadex G-50 columns (GE Healthcare).
Antibodies were modified with (E)-cyclooct-4-enyl 2,5-
dioxopyrrolidin-1-yl carbonate (TCO-NHS) as previously
reported [2]. Briefly, purified antibody was reacted with TCO-
NHS in 10 % dimethylformamide for 3 h at room temperature.
TCO-conjugated antibodies were subsequently buffer-
exchanged into PBS and their concentrations determined by
absorbance measurements. The following monoclonal antibod-
ies were modified with TCO: Herceptin (anti-HER2/neu),
Cetuximab (anti-EGFR), anti-EpCAM (R&D Systems), and
anti-Mucin-1 (Fitzgerald Industries).
The sample was labeled with TCO-conjugated antibodies
(10 μg/mL) in PBS with 0.5 % bovine serum albumin (BSA,
Sigma) for 45 min at 4 °C. Following washing and centrifugation,
cells were labeled with CLIO-TZ at room temperature for 30 min.
The sample was subsequently measured using the μHD sensor.

2.4 List of Materials Here, we list the materials used in the construction of the μHD. The
list is organized into the following sections: micro-Hall (μHall)
chip, magnetic labeling reagents, and auxiliary electronics.

2.4.1 μHall Chip 1. Substrate: epitaxially grown pseudomorphic high-electron

mobility transistor (PHEMT) GaAs (IntelliEpi).
2. Metals: Ni, Au, Ge.
3. Photoresists: SU-8 2000 (Microchem), Shipley 1813 (Microchem).
4. Poly(dimethylsiloxane) (PDMS) (KR Anderson Company,
Sylgard/Elastomer Kit). (All fabrication performed at Harvard
Center for Nanoscale Systems).
5. 2″ × 2″ × 1″ NdFeB magnet (K and J).
 Rare Cell Cancer Diagnostics 131

2.4.2 Magnetic Labeling 1. Cross-linked iron oxide nanoparticles: Courtesy of N. Sergeyev,

Center for Systems Biology, Massachusetts General Hospital
[17]. The particles were modified with 2,5-dioxopyrrolidin-1-yl
5-(4-(1,2,4,5-tetrazin-3-yl)benzylamino)-5-oxopentanoate
(TZ-NHS) to create CLIO-TZ.
2. The following monoclonal antibodies were modified with
TCO: Herceptin (anti-HER2/neu), Cetuximab (anti-EGFR),
anti-EpCAM (R&D Systems), and anti-Mucin-1 (Fitzgerald
Industries).

2.4.3 Auxiliary 1. Custom printed circuit board (described in Subheading 2.2).

Electronics 2. Power supply, 12 V, 100 mA.
3. National Instruments Analog to Digital Converter (NI-PCI-633).
4. USB cable.
5. Computer, tablet, or smartphone.

2.4.4 Auxiliary Fluidics 1. Syringe pump (Braintree Scientific).

2. Tygon non-DEHP microbore tubing, 0.020″ × 0.060″OD
(Cole Parmer).

3 Methods

3.1 Finite Element To aid in the design and characterization of our microfabricated
Simulations: Magnetic Hall sensors, we constructed a numerical model to describe the
spatial response of the sensors to magnetically labeled cells
(Matlab). The model treats each cell as a magnetic dipole moment
mtot located at the center of the cell, aligned with the external mag-
netic field in the z direction (Fig. 4a). The Hall voltage
(VH = I × RH × B⊥), where I is the input current to the sensor and RH
is the characteristic Hall resistance of the device, is calculated by
analytically solving for the demagnetization field B, and averaging
B⊥ over the sensing area of the Hall element. This numerical model
was used to determine the optimum sensor size that maximizes
signal-to-noise ratio (SNR) (Fig. 4b). We found that for cells with
an average diameter of 12 μm, the SNR was maximal when the
sensors had a detection area similar to that of the cells (8 × 8 μm2).
The spatial response of the Hall voltage VH was also simulated.
As the dipole was moved away from the sensor surface, the VH
signal steeply declined (Vh∝d−3) (Fig. 4c) Our numerical simula-
tion showed that the VH was >100-fold larger for a cell brought to
the sensor surface (d = 4 μm) than for a cell placed at the center of
the microfluidic channel (d = 7.5 μm), thus motivating accurate
positioning of the cell to the bottom of the channel with hydrody-
namic focusing.
132 David Issadore

Fig. 4 Summary of our detection method. (a) A computer model was used to design and optimize the μHall
chip. The software calculates the magnetic field B⊥ normal to the chip’s surface, which is produced by a
magnetic dipole at a specific position (x, y, z). (b) VH is a function of the sensor size (w, l) for a magnetic dipole
at a height d = 4 μm above the chip surface. (c) Simulation of the normalized VH as a function of the height of
a magnetic dipole at (x, y) = (0, 0). The signal strength was found to decay rapidly with distance (~d−3) from the
sensor surface

3.2 Finite Element To controllably stream cells over the μHall sensors, we design a
Simulations: Fluidic two-stage flow focusing structure (Fig. 5a). Cells are directed lat-
erally towards the center of the fluid channel via coplanar sheath
flows. Cells are pushed vertically toward the bottom of the chan-
nel using hydrodynamic forces generated with Chevron patterned
cut-out of the roof of the channel. The physical channel was
500 μm wide and 30 μm high, and designed to operate under
very large flow rates (φ ~ 1 mL/h). The use of hydrodynamic
focusing allows the physical channel to be much larger than the
cells, which in turn lowers the fluidic resistance and reduce the
risk of channel clogging.
To design this two-stage flow focusing structure we utilized an
iterative finite element analysis (Comsol). A cross section of the
microfluidic channel after lateral flow focusing is plotted (Fig. 5a,
inset). The sample, which includes the cells, is focused laterally
towards the center of the channel. After the chevron pattern the
sample is pushed vertically towards the bottom of the channel
(Fig. 5a, inset). The initial design for the chevron pattern was
modeled after a previously reported design [18]. We modified this
design for our application by including chevrons only on the top of
the channel rather than on the top and bottom as was previously
done. In this way, our chip drives cells to the bottom of the chan-
nel rather than towards the center as was done by Howell et al.
After a process of iterative simulation and redesign, which led
to a design that brought the cells to the bottom-center of the
channel, experimental verification was done. The device was built
using standard two-layer SU-8 soft lithography. The sample flow
was stained red with rhodamine and the sheath flow green with
fluorescein, and flow was analyzed using a fluorescence microscope
(Fig. 5b). As the sheath flow rate increased relative to the sample
 Rare Cell Cancer Diagnostics 133

Fig. 5 To enhance the μHall signal, a hydrodynamic focusing structure was used to bring individual cells close
to the μHall sensors. Cells entering the chip though the sample input are pushed towards the center of the
channel by the lateral sheaths and are pushed vertically to the bottom of the channel by the chevron patterns.
(a) The results of a finite element computer simulation. The sheath fluid is colored blue and the sample fluid in
colored red. (b) Fluorescence micrographs, demonstrating flow focusing. The sheath fluid was labeled with
fluorescein and the sample was labeled with rhodamine. As the sheath/sample flow rate ratio increased, the
sample flow was focused to an increasingly narrow region in the center of the chip

flow rate, the width of sample stream decreased. The ratio of the
flow rate between the sheath flow and the sample streams was con-
trolled using gravitational flow, with negative pressure provided by
a syringe pump on the output (BS8000, Braintree Scientific).
Typical total flow rates were 0.1–1 mL/h.

3.3 Circulating o demonstrate clinical use of the μHall chip, we detected rare cir-
Tumor Cell Detection culating tumor cells (CTCs) in whole blood. We conducted a small
study on a cohort of ovarian cancer patients (n = 20). Each of the
patients selected had advanced disease (Stage IIIc or IV), such that
we expected to find CTCs in each of their blood samples. As a
negative control, blood samples were obtained from healthy vol-
unteers (n = 15). To measure the performance of the μHall chip,
we directly compared its results against the clinical gold standard,
the CellSearch system.
134 David Issadore

Each patient sample we divided into two aliquots. One aliquot

was magnetically labeled for a panel of four markers (EpCAM,
HER2/neu, EGFR, and MUC1) and measured using the μHD
(Fig. 6a). The use of a panel of biomarkers enables a heteroge-
neous population of CTCs to be labeled, including those that
may have gone through an epithelial–mesenchymal transition and
do not express EpCAM [19]. The other aliquot was measured
using CellSearch. CA-125, a clinical biomarker for ovarian can-
cer, was also measured for each patient. The average CA-125
level for each of the cancer patients was 640 U/mL (with a range
of 141–1,142 U/mL), whereas a level for a healthy patient is
35 U/mL. In addition to the n = 20 ovarian cancer patients,
blood samples from n = 15 healthy volunteers were also measured
as a negative control.
Figure 6b compares CTC counts reported by the μHall Chip
and by CellSearch for each patient (n = 20). CellSearch detected
CTCs in only 5 of 20 ovarian cancer cases, resulting in a diagnos-
tic accuracy of 25 %. The μHall chip counted more CTCs in every
patient, and the cell counts were greater for patients with either
advanced disease who were no longer undergoing therapy or with
aggressive cancer types [4]. The μHall chip successfully identified
CTCs in 100 % of patients with evidence of clinical progression
or stage IV disease, where only 18 % of cases were detected with
CellSearch. Samples from patients with cancer showed signifi-
cantly higher number of CTC counts than the control group
(P < 0.01, two-tailed t-test) with wide margins (Fig. 6c). We con-
structed a receiver operating characteristic (ROC) curve using
the micro-Hall chip results from 20 patient and 15 healthy volun-
teers (Fig. 6d) [4]. Cutoff values were determined with input
from control cohort data. The diagnostic accuracy of the mHD
reached 94 %.

3.4 Drug Response In addition to the detection of cancer via the enumeration of rare
Measurements tumor cells in blood, the μHall chip can also be used to sensitively
measure biomarker expression on individual cells. This ability
allows the μHall chip to profile biomarker expression on samples
with limited cell numbers, such as fine needle aspirates (FNAs), for
applications such as the monitoring of disease progression during
treatment. The single cell measurements enabled by the μHall chip
obviate the need for separate experiments to normalize against an
independent cell count. To demonstrate this capability, we used
the μHall chip to monitor drug treatment efficacy on human A431
epidermoid cancer cells xenografted onto mice. Tumor samples
were obtained through fine-needle aspiration (~1,800 tumor cells
per aspirate). The aspirates were then labeled with EGFR-specific
MNPs and processed for micro-Hall measurement (Fig. 7a).
For a proof-of-concept experiment, three mice with xeno-
grafted tumors received treatment with geldanamycin, a heat shock
 Rare Cell Cancer Diagnostics 135

Fig. 6 Circulating tumor detection on a μHall chip. (a) The μHall chip was used to detect CTCs in blood samples
from patients with late stage ovarian cancer. Samples were magnetically labeled using four known cancer
biomarkers: EpCAM, HER2/neu, EGFR, and MUC1. (b) Patient samples (n = 20) were split into two and profiled
using both the μHall chip (top ) and CellSearch (bottom ). (c) Additionally, n = 15 healthy controls were screened
by the μHD. The mean values of cell counts are shown as a dashed line. (d) A receiver operating curve
(ROC) for the μHD was generated from the data in (c)
136 David Issadore

Fig. 7 Molecular analysis of fine needle aspirates on a μHall chip. (a) The μHall
chip was used to profile EGFR expression of cells obtained by fine needle aspi-
rate from mice with a xenografted tumor. Mice bearing xenografted tumors were
treated with geldanamycin for 6 days or left untreated (n = 6 per group). (b) Rate
of tumor growth in untreated mice and mice treated with geldanamycin. The blue
and red data points represent untreated and treated mice, respectively. (c) Tumor
samples were screened by the μHall chip to monitor the changes in EGFR expres-
sion over the course of the drug treatment (Color figure online)

protein 90 (HSP90) inhibitor. As a negative control, three mice

received saline. Geldanamycin has been shown to decrease the
expression of growth factor receptors by promoting their degrada-
tion. As time progressed, the tumor size of the treated mice
decreased and the tumor size of the untreated continued to grow
(Fig. 7b). With the micro-Hall chip, we observed a progressive
decrease in EGFR expression in tumor cells from the treatment
group (P < 0.05, two-tailed t-test), whereas the expression level of
EGFR remained unchanged in untreated mice (Fig. 7c). The μHD
could therefore be used for minimally invasive longitudinal treat-
ment monitoring because it reports biomarker expression per cell
in a small number of cells.

References

1. Lang JM, Casavant BP, Beebe DJ (2012) 3. Alix-Panabieres C, Schwarzenbach H, Pantel

Circulating tumor cells: getting more from K (2012) Circulating tumor cells and circulat-
less. Sci Transl Med 4:141ps13 ing tumor DNA. Annu Rev Med 63:199–215
2. Wang G-K, Zhu J-Q, Zhang J-T, Li Q, Li Y, 4. Issadore D, Chung J, Shao H, Liong M,
He J, Qin Y-W, Jing Q (2010) Circulating Ghazani AA, Castro CM, Weissleder R, Lee H
microRNA: a novel potential biomarker for (2012) Ultrasensitive clinical enumeration of
early diagnosis of acute myocardial infarction rare cells ex vivo using a micro-hall detector.
in humans. Eur Heart J 31:659–666 Sci Transl Med 4:141ra92
 Rare Cell Cancer Diagnostics 137

5. Nagrath S, Sequist LV, Maheswaran S, Bell Liao JC, Gambhir SS et al (2009) Matrix-
DW, Irimia D, Ulkus L, Smith MR, Kwak EL, insensitive protein assays push the limits of
Digumarthy S, Muzikansky A, Ryan P, Balis biosensors in medicine. Nat Med 15:
UJ, Tompkins RG, Haber DA, Toner M 1327–1332
(2007) Isolation of rare circulating tumour 13. Lee JH, Huh YM, Jun YW, Seo JW, Jang JT,
cells in cancer patients by microchip technol- Song HT, Kim S, Cho EJ, Yoon HG, Suh JS,
ogy. Nature 450:1235 Cheon J (2007) Artificially engineered mag-
6. Mettlin C, Littrup PJ, Kane RA, Murphy GP, netic nanoparticles for ultra-sensitive molecu-
Lee F, Chesley A, Badalament R, Mostofi FK lar imaging. Nat Med 13:95–99
(1994) Relative sensitivity and specificity of 14. Issadore D, Min C, Liong M, Chung J,
serum prostate specific antigen (PSA) level Weissleder R, Lee H (2011) Miniature mag-
compared with age-referenced PSA, PSA den- netic resonance system for point-of-care diag-
sity, and PSA change. Cancer 74:1615–1620 nostics. Lab Chip 11:2282–2287
7. de B, Johann S, Scher HI, Montgomery RB, 15. Issadore D, Chung HJ, Chung J, Budin G,
Parker C, Miller MC, Tissing H, Doyle GV, Weissleder R, Lee H (2013) μHall chip for
Terstappen LWWM, Pienta KJ, Raghavan D sensitive detection of bacteria. Adv Healthc
(2008) Circulating tumor cells predict survival Mater 2(9):1224–1228. doi:10.1002/adhm.
benefit from treatment in metastatic castration- 201200380
resistant prostate cancer. Clin Cancer Res 16. Weigl B, Domingo G, LaBarre P, Gerlach J
14:6302–6309 (2008) Towards non-and minimally instru-
8. Ozkumur E, Shah AM, Ciciliano JC, Emmink mented, microfluidics-based diagnostic
BL, Miyamoto DT, Brachtel E, Yu M, Chen devices. Lab Chip 8:1999–2014
P-I, Morgan B, Trautwein J et al (2013) 17. Haun JB, Devaraj NK, Hilderbrand SA, Lee
Inertial focusing for tumor antigen – depen- H, Weissleder R (2010) Bioorthogonal chem-
dent and – independent sorting of rare circu- istry amplifies nanoparticle binding and
lating tumor cells. Sci Transl Med 5:179ra47 enhances the sensitivity of cell detection. Nat
9. Gaster R, Hall D, Wang S (2011) nanoLAB: Nanotechnol 5:660–665
an ultraportable, handheld diagnostic labora- 18. Howell J, Peter B, Mott DR, Fertig S, Kaplan
tory for global health. Lab Chip 11:950–956 CR, Golden JP, Oran ES, Ligler FS (2005) A
10. Issadore D, Westervelt RM (eds) (2013) Point microfluidic mixer with grooves placed on the
of care diagnostics on a chip. Springer, Berlin, top and bottom of the channel. Lab Chip
Heidelberg. 5:524–530
11. Ejsing L, Hansen MF, Menon AK, Ferreira HA, 19. Rhim AD, Mirek ET, Aiello NM, Maitra A,
Graham DL, Freitas PP (2004) Planar Hall Bailey JM, McAllister F, Reichert M, Beatty
effect sensor for magnetic micro- and nanobead GL, Rustgi AK, Vonderheide RH et al (2012)
detection. Appl Phys Lett 84:4729–4731 EMT and dissemination precede pancreatic
12. Gaster RS, Hall DA, Nielsen CH, Osterfeld tumor formation. Cell 148:349–361
SJ, Yu H, Mach KE, Wilson RJ, Murmann B,
 Chapter 10

Mobile Flow Cytometer for mHealth

Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly

Abstract
Flow cytometry is used for cell counting and analysis in numerous clinical and environmental applications.
However flow cytometry is not used in mHealth mainly because current flow cytometers are large, expensive,
power-intensive devices designed to operate in a laboratory. Their design results in a lack of portability and
makes them unsuitable for mHealth applications. Another limitation of current technology is the low volu-
metric throughput rates that are not suitable for rapid detection of rare cells.
To address these limitations, we describe here a novel, low-cost, mobile flow cytometer based on
wide-field imaging with a webcam for large volume and high throughput fluorescence detection of rare
cells as a simulation for circulating tumor cells (CTCs) detection. The mobile flow cytometer uses a com-
mercially available webcam capable of 187 frames per second video capture at a resolution of 320 × 240
pixels. For fluorescence detection, a 1 W 450 nm blue laser is used for excitation of Syto-9 fluorescently
stained cells detected at 535 nm. A wide-field flow cell was developed for large volume analysis that allows
for the linear velocity of target cells to be lower than in conventional hydrodynamic focusing flow cells
typically used in cytometry. The mobile flow cytometer was found to be capable of detecting low concen-
trations at flow rates of 500 μL/min, suitable for rare cell detection in large volumes. The simplicity and
low cost of this device suggests that it may have a potential clinical use for mHealth flow cytometry for
resource-poor settings associated with global health.

Key words mHealth, Webcam, Cytometry, CTC, Fluidics, Fluorescence detection

1 Introduction

Flow cytometers are used for many clinical applications, environ-

mental analysis and basic research. However, flow cytometry is not
used in mHealth, mainly because current flow cytometers are not
portable and are not suitable for resource-poor settings. The main
components of current flow cytometers consist of a fluidic system
for carrying fluorescently labeled biological material (e.g., cells)
and an optical system for fluorescence detection. Most current
flow cytometers utilize microfluidic sheathing to focus the cells
or particles to a channel whose width permits only a single cell
to pass to enable narrow-field detection using photomultipliers
or other narrow-field photodetectors. In a recent device, passive

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_10, © Springer Science+Business Media New York 2015

139
140 Joshua Balsam et al.

hydrodynamic focusing generated by chevron grooves imbedded

on the walls of the microchannel [1, 2] enables the sheath fluid to
completely surround the sample stream so that cells can be inter-
rogated one at a time by the narrow-field fluorescent detector.
However, one limitation of the focused stream is a low flow
rate due to the high hydrodynamic resistance and pressure
constraints of the cell, which ultimately limits the device to small
volumes or long analysis times. In addition, the commonly used
photomultipliers used for detection are delicate and expensive.
Current flow cytometers are large expensive devices which are
power-intensive and designed to operate in a laboratory environ-
ment, so they are not suitable for mHealth applications.
An alternative detection approach to photomultipliers are array
imaging sensors such as complimentary metal-oxide-semiconductor
(CMOS) or charge-coupled device (CCD) devices, which can be
used for imaging large numbers of cells in parallel. Because of the
inherent portability of these sensors, either CMOS or CCD cam-
eras can be employed as relatively simple, low cost, sensitive device
for optical detection over large areas, and have already been
employed in several array assays [3–6]. Their main advantage is
they can be used for analyzing light from a large enough area that
it can cover the entire surface of a lab-on-a-chip (LOC) or an array
[7–9]. This has made CMOS or CCD-based detectors an ideal
choice for mHealth detectors.
Portable cytometers have potential for Point-of-Care (POC)
clinical applications, and for use in low and middle-income coun-
tries. LOC fluidic technology provides a potential approach for
developing POC analytical tools in resource-poor settings
[10, 11]. Recently, optofluidic fluorescent imaging cytometry on a
cell phone with a spatial resolution of ~2 μm was described [12, 13].
While very mobile and versatile, the flow rate of this system is
~1 μL/min, which limits analysis to small volumes. Mobile phones
are used effectively for mobile POC devices [12–20], however the
cameras in these phones are often limited in their sampling rates
(e.g., many phones are limited for 30 fps) and are less versatile with
their optical systems (e.g., inability to change lenses or directly set
imaging parameters) than devices such as webcams. In this paper,
we describe a small, mobile, and low-cost flow cytometer based on
webcam imaging that overcomes these limitations and is capable of
large volume analysis and rare cell detection for a variety of mHealth
applications.
Rare cells including circulating tumor cells (CTCs) have
many promising clinical applications and there are many tech-
nologies for CTC detection and analysis, however flow cytometry
despite it great potential is not well suited for rare cells analysis.
To demonstrate the potential utility of mobile flow cytometer to
CTC detection, fluorescently stained THP-1 human monocytes
 Mobile Flow Cytometer for mHealth 141

were used as a model to simulate rare cells. Though monocytes

themselves are not rare, they were diluted to levels similar to
those of rare cells such as CTCs.

2 Materials

2.1 Optical 1. A Sony PlayStation® Eye webcam was used as an inexpensive

Components CMOS photodetector (Amazon, Seattle, WA).
for Mobile Flow 2. An alternative camera with improved sensitivity is a C mount
Cytometer 1.3 Mp grey scale CCD camera CMLN 13S2M-CS (Point
Grey Research, Richmond BC Canada).
3. A C-mount CCTV lens (Pentax 12 mm f/1.2) was used to
replace the webcam lens to improve imaging performance
(Spytown, Utopia, NY).
4. Green emission filter HQ535/50M (Chroma Technology
Corp. Rockingham, VT).
5. For fluorescent excitation illumination, a 1 W 450 nm laser
module was used (Hangzhou BrandNew Technology Co.,
Zhejiang, China).

2.2 Flow Cell 1. 3 mm clear acrylic sheet.

and Fluidic System 2. Glass microscope slide.
3. 3M 9770 Adhesive transfer Tape (Piedmont Plastics,
Beltsville, MD).
4. Fusion100 syringe pump was used for flow rate control
(Chemyx, Stafford, TX).
5. Epilog Legend CO2 65 W laser cutter (Epilog, Golden, CO).

2.3 Computer 1. Drivers and software allowing the webcam to be controlled on

Control and Data a personal computer were developed and freely distributed by
Analysis Code Laboratories, Inc. (Henderson, NV). Control software
(CL-Eye Test) was used to set camera parameters (exposure
time, frame rate, and gain) and to capture and save video in
uncompressed AVI format.
2. Video files were analyzed using ImageJ software (freely distrib-
uted by NIH, http://rsb.info.nih.gov/ij/download.htm).
3. Data analysis software: Excel (Microsoft, Redmond, WA).

2.4 Cell Culture 1. THP-1 human monocytes, ATCC #TIB-202 (ATCC Manassas, VA).
and Fluorescence 2. Culture media RPMI-1640 medium supplemented with 10 %
Staining (v/v) heat-inactivated FBS, 1 % antibiotic–antimycotic solu-
tion, and 2 mM glutamine.
3. SYTO-9 dye (Molecular Probes Eugene, OR).
142 Joshua Balsam et al.

3 Methods

3.1 Wide-Field Flow The flow cytometer for fluorescence detection (Fig. 1) consists of
Cytometer for mHealth four main elements: (1) a webcam used as an optical detector,
Applications (2) a laser excitation source for illumination, (3) a wide-field flow
cell, and (4) an optical stage to hold each module in alignment for
stable and clear imaging. The optical detector consists of the elec-
tronics of a webcam which was disassembled in order to remove
the original lens, a 12 mm f/1.2 CCTV lens which enables
improved light collection and image magnification (see Note 1), a
green emission filter for fluorescent imaging, and a computer to
collect and analyze data. The excitation source is a 1 W 450 nm
blue laser (see Note 2). The optical elements must be vertically
centered (see Notes 2 and 3). The fluidics module consists of a
flow cell and a programmable syringe pump. The flow cytometer
platform shown in Fig. 1a consists of a stationary platform and a
moveable stage for focusing (using a screw mechanism). The flow
cytometer platform was constructed using 0.5 in. thick clear acrylic
sheet and nylon rod (McMaster-Carr, Robbinsville, NJ) which was
used as a rail for focusing. The webcam electronic circuitry and
optics were attached to the stationary platform while the flow cell
was attached to the translating stage. It is important to enclose the
device to reduce background light (see Note 4).

a b

Flow Cell Flow Cell

Inlet

In from syringe Movable stage Out to Waste Field of View

pump
mm Laser Spot
CCTV
Lens
Emission
Filter
Computer
Webcam circuit board and Outlet
sensor with USB output

Fig. 1 Schematic of mobile wide-field flow cytometer for mHealth. (a) The flow cytometer consists of four
modules: (1) a webcam as an optical detector, (2) a blue 435 nm excitation source for illumination, (3) a wide-
field flow cell, and (4) a stage to focus the image ant to hold each module in alignment. The sensing element
(1) consists of the electronic elements of a webcam, a 12 mm f/1.2 CCTV lens, a green emission filter. It is
connected to a computer to collect and analyze data. The excitation source (2) is a 450 nm 1 W blue laser.
The sample handling module consists of the wide-field flow cell, a programmable syringe pump, and a waste
container. (b) A schematic of the wide-field flow cell with key elements labeled with the illuminated field of
detection
 Mobile Flow Cytometer for mHealth 143

3.2 Wide-Field Flow The flow cell (Fig. 1b) consists of a 4 mm wide imaging channel
Cell Fabrication through which fluorescent labeled samples are injected via an inlet,
excited by a laser, and then imaged by the webcam before being
collected at the outlet. The wide-field flow cell was fabricated using
an Epilog Legend CO2 65 W laser cutter (Epilog, Golden, CO) (see
Note 5) using similar techniques described in our previous work
[11, 21–25]. The flow cell consisted of three functional layers. The
upper layer fabricated of a 3 mm thick clear acrylic plate (75 × 25 mm)
with laser machined inlet and outlet ports into which 18GA needles
were pressed and bonded to allow for sample injection and waste
collection. The middle layer, which defines the shape of the fluid
flow, is a laser machined channel fabricated from 3M 9770 double-
sided adhesive transfer tape, which results in a wide-field flow cell
with width 4 mm, length 45 mm, and depth of approximately
0.075 mm. The bottom layer consisted of a single glass microscope
slide which was bonded to the acrylic with the double-sided adhe-
sive transfer tape micromachined flow channel (see Note 6).
A Fusion100 syringe pump connected to the flow cell was used
for flow rate control. The maximum flow rate achievable through
this flow cell is 10 mL/min. The adhesive may fail at higher flow
rate due to the dynamic pressures required. The flow rate in experi-
ments presented in this paper was limited to 500 μL/min due to
the maximum frame rate of the webcam employed for sensing. For
very faint cells the flow rate should be decreased in order to improve
sensitivity. Sensitivity also increases if exposure time is increased to
allow cell images to form short streaks, with length equal to the
length of the cell image plus one pixel (detailed in Subheading 3.7).

3.3 Optical System The optical system includes the blue laser for fluorophore excitation,
for Webcam-Based webcam detector, and excitation filter.
Cytometer Platform
Laser Excitation. For fluorescent imaging of a wide field, a 1 W
consumer laser was used to project an elliptical spot which covered
the width of the flow cell. The laser illuminates the flow cell at an
angle of incidence of approximately 45° (see Note 2). A laser such
as this is fairly expensive (~$300) for a device designed for use in a
low-resource setting. To reduce the cost of the laser, a lower power
laser with line generator optics could be used to further focus the
laser spot. This could allow the critical parameter of photon flux to
remain unchanged while reducing overall power. Alternatively,
high power LEDs could be utilized. This would require the addi-
tion of an excitation filter as well as collimating optics, leading to a
more complex and potentially more expensive configuration.
Webcam detector and emission filter. A Sony PlayStation® Eye web-
cam was used as the photodetector in this platform by converting
the webcam to a microscope. The device was disassembled and
the main circuit board (with attached image sensor and USB
cable) was removed. A C-mount CCTV lens (Pentax 12 mm f/1.2)
144 Joshua Balsam et al.

was used to replace the regular webcam lens in a distance of

approximately 20 mm from the CMOS. This distance allowed the
lens to focus at very close range (see Note 1). For fluorescence
detection, a green emission filter with center wavelength 535 nm
and bandwidth 50 nm (Chroma Technology Corp., Rockingham,
VT) was used for detecting fluorescent emission. To improve sen-
sitivity, other video devices which employ CCDs with electronics
for high frame rate video imaging, such as the Point Grey Research
CMLN 13S2M-CS, can be used.

3.4 Cell Culture Fluorescently stained THP-1 human monocytes were used as a
Staining and Dilution model to simulate rare CTC. Though monocytes themselves are
not rare, they were diluted to levels similar to those of rare cells
such as CTCs. Cells were cultured with RPMI-1640 medium sup-
plemented with 10 % (v/v) heat-inactivated FBS, 1 % antibiotic–
antimycotic solution, and 2 mM glutamine (see Note 7) in a 5 %
CO2 environment at 37 °C. Cells were removed from an active
culture, pelleted by centrifugation and resuspended in deionized
water at an approximate concentration of 106 cells/mL (staining
protocol for Syto-9 dye advises against the use of phosphate buffer
solutions). 10 μL of Syto-9 dye (3.34 mM stock concentration)
was added to 1 mL of suspended cells and allowed to rest at room
temperature in the dark for 20 min. Washing of cells to remove
excess dye is not necessary due to the low intrinsic quantum yield
when not bound to nucleic acids (<0.01), and because of the later
dilution of this stock solution by factors of 104–106.
One known factor which contributed substantially to the
decreasing counting efficiency at lower concentrations is the cell
staining protocol. The protocol recommended by the dye manu-
facturer was followed. However, dye which had entered the cell
but not yet bound to DNA was noted to diffuse out of cells over
time, leading to diminishing cell brightness. Because all dilutions
were prepared simultaneously and then analyzed sequentially from
highest to lowest concentration, lower dilutions exhibited cells
with lower fluorescent intensity. Allowing the cells to equilibrate at
a low concentration for several hours would allow this diffusion
process time to complete.
The stock solution of cells after staining was diluted to a level
of approximately 1–10 cell/μL (measured by microscopy) to allow
for accurate manual counting. Cell concentration was measured by
placing 3 μL sample droplets on a microscope slide and counting
cells in the droplet in real time under laser excitation using the
same imaging platform employed to image the flow cells in these
experiments. This was repeated many (N > 20) times. An average
cell concentration of 3.73 cells/μL with a standard error of 0.3 is
an example of typical population estimate resulting from these
measurements. From this relatively high concentration, lower con-
centration samples of 100, 10, and 1 cell/mL were generated by
 Mobile Flow Cytometer for mHealth 145

single-step dilution. For each dilution, 268 μL of stock solution

(with measured concentration of 3.73 cells/μL) was diluted into a
volume of deionized water to yield the final target concentration
(10 mL, 100 mL, and 1,000 mL of deionized water for concentra-
tions of 100, 10, and 1 cell/mL, respectively). Based on a normal
sampling distribution with standard deviation of 0.3 cells/μL, the
95 % confidence range for each concentration was 84–116,
8.4–11.6, and 0.84–1.16 cells/mL for concentrations of 100, 10,
and 1 cell/mL, respectively. Pipetting volume error was measured
to be less than 1 %. Results of measuring these concentrations of
rare cells are presented in Subheading 4.

3.5 Imaging Flow The webcam sensor was connected to a 32-bit Windows-based
Cytometry laptop computer via a USB2 port. The drivers and software con-
trolling the webcam were developed and freely distributed by Code
Laboratories, Inc. (Henderson, NV). The camera control software
(CL-Eye Test) enables setting several camera parameters (exposure
time, frame rate, and gain), and the capture of video in uncom-
pressed AVI format. The imaging parameters for the camera (expo-
sure time and frame rate) depend on the brightness of the
fluorescently labeled cells and the flow rate (see Note 8). A single
video frame of THP-1’s stained with SYTO-9 dye in wide-field
flow cell at a flow rate of 500 μL/min is shown in Fig. 2a. The flow
cell geometry and fluorescent detection optics allow for a high sig-
nal to noise ratio for easy detection of these cells (Fig. 2b).

3.6 Optimization In order to optimize the performance of a wide-field flow cytom-

of Device Dimensions eter, the following method for setting various device parameters
and Imaging can be used. This will result in maximized cell image SNR, maxi-
Conditions mized sample throughput, and accurate flow sampling.

Fig. 2 Cell imaged using wide-field flow cytometer and webcam-based flow focusing cytometry. (a) Single
video frame of fluorescent labeled THP-1 human monocytes in wide-field flow cell at 500 μL/min, and (b) 3D
visualization of a
146 Joshua Balsam et al.

First, the distance between the imaging lens and the flow cell
should be set such that images of cells are projected onto at most a
3 × 3 array of pixels. Distance should not be so great that a cell
image is less than one pixel in size. At those distances, photon flux
per pixel begins to diminish and SNR drops quickly. When cells are
imaged on more than one pixel, photon flux per pixel is a constant
and independent of the distance between flow cell and lens. The
largest possible distance should be chosen in order to maximize
field of view (FOV). Set the flow cell width so that it covers the
FOV of the image sensor at this lens-to-flow cell distance. This
allows the largest possible flow cell width to be used which mini-
mizes flow velocity and maximizes SNR at a given flow rate.
Next, the relationship between average cell velocity and flow
rate should be determined empirically. The maximum velocity
achievable will be determined later and is a function of cell image
signal to noise ratio (SNR) (e.g., higher SNR allows higher veloc-
ity). This in turn depends on the cell staining characteristics
(e.g., quantum yield of the dye, number of fluorophores bound,
and excitation field intensity), the camera used (e.g., camera sensi-
tivity, characteristic noise levels, and maximum frame rate), and the
lens used (e.g., focal ratio).
Exposure time should be set such that cell image brightness is
maximized. For a given average flow velocity and average cell
brightness (i.e., photon emission rate), there will be a maximum
cell image brightness that can be produced. This is based on the
number of photons than can strike a pixel as the cell image passes
over it. The minimum exposure time that will yield this maximum
image intensity will be the time required for a cell image to traverse
a number of pixels equal to its image length in pixels (L) plus one
pixel, thereby producing an image streak of length 2L + 1.
Finally, volumetric flow rate should be set such that the entire
distribution of cells is significantly above the noise background of
the system (this typically corresponds to an SNR > 5). At higher
flow rates image SNR will decrease. Above a certain flow rate, cells
at the faint end of the distribution will be indistinguishable from
background noise and the average cell count will decrease. In order
to set flow rate, a stock solution of fluorescent cells should be pre-
pared and counted beginning at a high flow rate and decreasing
until the average cell count per sample reaches a constant value.
As previously mentioned this maximum flow rate will depend on
several factors, including the inherent brightness of cell staining
and the excitation photon flux. The parameters of exposure time
and frame rate depend directly on flow rate, so these parameters
will need to be updated as flow rate is varied in order to maintain
similar levels of device sensitivity across various tests.

3.7 Computational Several computational approaches were used to improve detection:

Image Enhancement (1) color separation, (2) image stacking, (3) background substraction,
to Improve Detection and (4) cell streaking.
 Mobile Flow Cytometer for mHealth 147

1. Color separation. The images can be analyzed using any standard

image processing software capable of pixel value measurement
and pixel based calculations (i.e., averaging multiple frames).
For our system, we chose the freeware ImageJ developed at
NIH (see Note 7). If a color image sensor is used (typical of
cell-phone cameras and webcams), each color image can be
decomposed into three different monochrome images, each
one representing a color channel (red, green and blue) and
each pixel of each image having a value between 0 and 255 for
8-bit sensors. After splitting the RGB color channels apart,
only the channel corresponding to the wavelength in the spe-
cific emission spectrum of interest is analyzed, and the other
channels are discarded to reduce noise.
To perform color separation in ImageJ, with the image
open select from the menu Image > Color > Split Channels.
The original image will be split into its RGB components.
Knowing the emission wavelength of the fluorophore(s) being
used, discard those components which do not contain the
wavelengths of interest (i.e., for SYTO-9, a green dye, discard
the red and blue components as they contain mostly noise).
The final image, now monochrome, must be measured to
extract the assay data. For frame-by-frame analysis, using the
brightness and contrast control window, increase the screen
contrast until the individual cells can be distinguished from
one the background. Using an appropriate selection size and
shape, define regions of interest around each cell image. A suit-
able statistical value to use for measurements must be chosen
(e.g., mean, median, maximum). Data can then be exported
into analysis software (e.g., Excel, Minitab, or other suitable
analysis software).
2. Image stacking for rare cell detection. When using video mode,
to quickly find a rare event (e.g., very few cells in large volume)
without manually scanning each frame, an image stacking
approach was used (for more details on image stacking please
see Chapter 16). Image stacking will enable analysis of all
frames simultaneously and identifying rare cells presented in
few frames.
3. Background reduction. Frames may contain a signal of interest
(e.g., cell image), interfering signal (e.g., autofluorescence),
and random noise. To enhance detection the constant compo-
nent of the interfering signal can be subtracted (for more detail
on image stacking noise subtraction please see Chapter 16). For
background subtraction the median value and the maximum
value of each pixel in a video file of the relevant color channel
can be calculated. Figure 3 is the green channel video frames of
a video clip containing 2,000 frames (10.7 s of video) showing
a single cell passing through the flow cell (shown in different
148 Joshua Balsam et al.

Fig. 3 Background subtraction to improve imaging. (a) Median pixel value from 2,000 frames showing average
background autofluorescence from the flow cell, (b) maximum pixel value from 2,000 video frames showing a
single cell moving through the laser spot (marked with circled in yellow), (c) result of subtracting image b from
image c, allowing for improved visualization of cell movement and faint cell images

positions along its path). The median image of all the frames is
shown in Fig. 3a (this represents the average background
signal), and the maximum image is shown in Fig. 3b which
shows the highest signal recorded at each pixel during the video
with the cell position in each frame marked. In order to improve
cell identification, especially for lower SNR images than this
example, the median image shown in Fig. 3a, which contains
only background signal and no information from the cells, was
subtracted from the maximum image to produce Fig. 3c, which
allows for improved visualization of cell movement by removing
background autofluorescence from the flow cell.
4. Cell streaking. Cell streaking is imaging of moving cells at long
exposure times such that they produce line images, as seen in
Fig. 4a. This cell image was captured at a flow rate of 20 mL/
min with a frame rate of 20 fps. The primary benefit of cell
streaking is that it allows pixels to integrate over the entire
length of a cell as its moves image moves over them. Shorter
exposure times result in only a portion of the available photon
flux to be integrated by a pixel, reducing image SNR. Figure 4a
shows significant streaking, which is not necessary to realize
this benefit. The minimum streak length required such that at
least one pixel measures the maximum possible signal is equal
to twice the length of the cell image in pixels before streaking
(L) plus one pixel (i.e., for a cell image of 3 × 3 pixels, a streak
length of 7 is required). This effect is shown graphically in
Fig. 4b, where a cell of length 3 pixels is seen to move a dis-
tance of 4 pixels, thereby producing an image streak with
length 7 pixels with a single bright pixel in its center. An actual
cell streak image is shown in Fig. 4c for a similar cell image size
and similar flow conditions, along with a plot of actual pixel
brightness along the center of the cell streak in Fig. 4d. Cell
streaking such as this results in at least one pixel measuring the
maximum possible signal from the cell.
 Mobile Flow Cytometer for mHealth 149

d 20

15
Gray Value

10

0 2 4 6 8
Distance (pixels)

Fig. 4 Cells in a streaking mode. (a) An image of cells traveling in a flow rate of
10 mL/min captured in 20 frames per second imaged as lines. (b) Schematic of
a cell traversing a number of pixels equal to its length plus one pixel, showing a
maximum brightness achieved in the pixel at the image center. (c) An actual cell
image is shown under these approximate flow conditions (125 μL/min, 20 fps),
along with (d) a plot of its brightness along the center line of pixels
150 Joshua Balsam et al.

5. Rare cell counting using streaking mode. Counting of rare cell

events was simulated using dilutions of fluorescently labeled
THP-1 monocytes. Dilutions were prepared as described in
Subheading 3.4 and measured using the wide-field flow
cytometer depicted in Fig. 1a and described in Subheading 2.
Figure 5a shows results that were achieved using the flow cell
depicted in Fig. 1b. Dilutions of 100, 10, and 1 cell per mL
yielded average concentrations of 84, 7.9, and 0.56 cells/mL
with 95 % confidence bounds of 61–107, 6.9–9, and
0.16–0.96 cells/mL, respectively (error bars in Fig. 5 repre-
sent 95 % confidence interval).
Following this set of experiments, flow cell geometry and sys-
tem operating parameters (flow rate, frame rate) were optimized
based on the methods described in Subheading 3.6. This resulted in
the flow cell pictured in Fig. 5b with increased channel width and
an increased field of view. The excitation laser spot was modified to

a 1000
Before Optimization
Measured Cell Count (cells/mL)

100
b

10 Inlet
1

0.1

0.01
Field of View
100 10 1
c 1000 Laser
After Optimization
Measured Cell Count (cells/mL)

100

10

0.1 Outlet
0.01
100 10 1 0.1
Target Cell Count (cells/mL)

Fig. 5 Rare cell counting using streaking mode. (a) Results of counting dilutions of rare cells using the first
generation flow cell depicted in Fig. 1b. The large error bars evident at the level of 1 cell/mL prompted further
optimization of the flow cell geometry and operating parameters. (b) An updated flow cell design which
improves image SNR by reducing cell velocity. Field of view was also improved. (c) Following optimization,
performance of the wide-field flow cytometer improved substantially at the level of 1 cell/mL, and utility was
expanded to measurements as low as 0.1 cell/mL. In each case, measurements made via microscopy (sparsely
filled columns) and those made via wide-field flow cytometry (dense fill) were not significantly different at the
95 % confidence level
 Mobile Flow Cytometer for mHealth 151

form a line covering the width of the channel. The modifications

to the flow cell resulted in cells moving with lower velocity, which
translates to higher image SNR. This allowed for substantially
improved results at low cell dilutions as seen in Fig. 5c. Dilutions
of 1 and 0.1 cell/mL were prepared and measured via microscopy
(95 % confidence bounds: 0.84–1.16 and 0.094–0.106, respec-
tively). These dilutions were then measured in cell streaking mode
with the optimized wide-field cytometer, yielding average counts
of 0.91 and 0.083 cells/mL, each with a 95 % confidence bound
of 0.85–0.97 and 0.065–0.102 cells/mL, respectively. In both
cases the two means are not significantly different at the 95 % con-
fidence level.
The technology described here has many promising clinical
applications, including CTC detection. This may enable better
cancer prognosis, earlier detection of metastasis-capable malig-
nancy, guidance in selection of treatment, and evaluation of treat-
ment efficacy to help prevent patients from being exposed to a
treatment that is ineffective. The simplicity and the low cost of the
webcam-based wide-field flow cytometer presented here suggests
that this configuration may have the potential for developing POC
clinical flow cytometry for rare cell detection in resource-poor set-
tings associated with global health.

4 Notes

1. A fast focal ratio (e.g., f/1.2) will enable shorter exposure

time, but focusing will be more difficult as the depth of field
will be reduced. Extension tubes are added between the lens
and sensor to allow for close focusing and enlargement of cell
images. This can result in noticeable image field curvature
(i.e., image corners out of focus while image center is in focus),
so spacing and imaging distance must be optimized. Focal
ratio can be reduced to reduce field curvature.
2. Make sure the lens, filters, and flow cell are vertically centered
and aligned and that the laser illuminates the image area of the
flow cell.
3. Make sure the arrows on the coated filters are facing the cam-
era, and that if the filters being used have angular dependence
(e.g., interference filters) the excitation source beam is con-
fined to the correct range of angles.
4. The imager enclosure must be sealed to block external light
sources. A long exposure can be used to detect light leaks.
5. For flow cell fabrication, the laser power and speed for cutting
polymers has to be determined empirically. It is recommended
to use the minimum laser power to reduce overheating or
excessive burning of the material.
152 Joshua Balsam et al.

6. For strong bonding, all air bubbles between the surfaces and
the adhesive tape should be pressed out manually. The adhesive
layer should be attached to the acrylic side first to ensure that
proper alignment exists between the inlet and outlet ports.
Bubbles between these layers can be pressed out before the pro-
tective paper layer on the back of the adhesive tape is removed.
After this, with the protective paper removed, the glass layer
can be bonded. Bubbles between the glass and adhesive layer
can be removed by carefully applied pressure, or by use of a
heated press. Care must be taken to ensure even pressure
because of the risk of glass breakage. To avoid this, the glass
layer can be replaced by another layer of acrylic. However the
high autofluorescent emission rate of acrylic will result in
reduced SNR and diminished dynamic range of the sensor. For
improved sensitivity, two quartz microscope slides can be used
in place of the acrylic layer and the glass slide. Inlet and outlet
holes must be cut using a glass cutting drill bit or other appro-
priate method. High pressure ports can be constructed by
pressing an 18GA needle into a laser cut hole in a small square
of acrylic sheet, sealing with cyanoacrylate glue, and bonding
over the inlet and outlet holes in the quartz slide using double
sided adhesive transfer tape with a center hole cut out.
7. Cell cultures with very high density may result in cell anucle-
ation (ghost cells), cells with low fluorescence signal when
labeled with nucleic acid stain.
8. Imager Qualification: webcam performance varies depending
both on the device and the application used to collect the
images from the camera. Prior to preparing a sample for mea-
surement it is important to understand the performance of the
camera and application being used. First, block any light from
reaching the camera sensor by taping a layer of aluminum foil
over the lens aperture. Take two images and open them in
ImageJ or a comparable software package. If a color sensor is
used, use only the green channel for analysis. Convert the
images to 32-bit float format, and subtract one image from the
other. Finally, use the software package to produce a histogram
of all pixel values. This histogram should have an approxi-
mately Gaussian profile. If the original images had pixel values
that were predominantly zero valued, you must change the
settings of the camera capture software so that the black end of
the sensor histogram is not being artificially clipped. This clip-
ping behavior is typically set by an adjustment labeled bright-
ness. If the images captured by the sensor with no light arriving
cannot be made to show values above zero with an approxi-
mately Gaussian profile, it is likely that the device is not suit-
able for performing sensitive fluorescence measurements of
faint objects.
 Mobile Flow Cytometer for mHealth 153

References
1. Golden JP, Kim JS, Erickson JS, Hilliard LR, 13. Zhu H, Mavandadi S, Coskun AF, Yaglidere O,
Howell PB, Anderson GP et al (2009) Multi- Ozcan A (2011) Optofluidic fluorescent imaging
wavelength microflow cytometer using groove- cytometry on a cell phone. Anal Chem 83(17):
generated sheath flow. Lab Chip 9(13): 6641–6647, Epub 2011/07/22
1942–1950, Epub 2009/06/18 14. Wei Q, Qi H, Luo W, Tseng D, Ki SJ, Wan Z
2. Howell PB Jr, Golden JP, Hilliard LR, Erickson et al (2013) Fluorescent imaging of single
JS, Mott DR, Ligler FS (2008) Two simple and nanoparticles and viruses on a smart phone. ACS
rugged designs for creating microfluidic sheath Nano 7(10):9147–9155, Epub 2013/09/11
flow. Lab Chip 8(7):1097–1103, Epub 15. Coskun AF, Nagi R, Sadeghi K, Phillips S,
2008/06/28 Ozcan A (2013) Albumin testing in urine using
3. Taitt CR, Anderson GP, Ligler FS (2005) a smart-phone. Lab Chip 13(21):4231–4238,
Evanescent wave fluorescence biosensors. Epub 2013/09/03
Biosens Bioelectron 20(12):2470–2487, Epub 16. Navruz I, Coskun AF, Wong J, Mohammad S,
2005/04/28 Tseng D, Nagi R et al (2013) Smart-phone
4. Ngundi MM, Qadri SA, Wallace EV, Moore based computational microscopy using multi-
MH, Lassman ME, Shriver-Lake LC et al frame contact imaging on a fiber-optic array. Lab
(2006) Detection of deoxynivalenol in foods Chip 13(20):4015–4023, Epub 2013/08/14
and indoor air using an array biosensor. Environ 17. Zhu H, Sencan I, Wong J, Dimitrov S, Tseng
Sci Technol 40(7):2352–2356 D, Nagashima K et al (2013) Cost-effective
5. Moreno-Bondi MC, Taitt CR, Shriver-Lake and rapid blood analysis on a cell-phone. Lab
LC, Ligler FS (2006) Multiplexed mea- Chip 13(7):1282–1288, Epub 2013/02/09
surement of serum antibodies using an 18. Zhu H, Sikora U, Ozcan A (2012) Quantum
array biosensor. Biosens Bioelectron 21(10): dot enabled detection of Escherichia coli using
1880–1886 a cell-phone. Analyst 137(11):2541–2544,
6. Ligler FS, Sapsford KE, Golden JP, Shriver- Epub 2012/03/08
Lake LC, Taitt CR, Dyer MA et al (2007) The 19. Zhu H, Yaglidere O, Su TW, Tseng D, Ozcan A
array biosensor: portable, automated systems. (2011) Wide-field fluorescent microscopy on a
Anal Sci 23(1):5–10 cell-phone. Conference proceedings: Annual
7. Kostov Y, Sergeev N, Wilson S, Herold KE, international conference of the IEEE engineering
Rasooly A (2009) A simple portable electrolu- in medicine and biology society, IEEE Engineering
minescence illumination-based CCD detector. in Medicine and Biology Society Conference,
Methods Mol Biol 503:259–272, Epub 2011. pp 6801–6804. Epub 2012/01/19
2009/01/20 20. Zhu H, Yaglidere O, Su TW, Tseng D, Ozcan
8. Sapsford KE, Sun S, Francis J, Sharma S, A (2011) Cost-effective and compact wide-
Kostov Y, Rasooly A (2008) A fluorescence field fluorescent imaging on a cell-phone. Lab
detection platform using spatial electrolumi- Chip 11(2):315–322, Epub 2010/11/11
nescent excitation for measuring botulinum 21. Rasooly A, Bruck HA, Kostov Y (2013) An
neurotoxin A activity. Biosens Bioelectron ELISA lab-on-a-chip (ELISA-LOC). Methods
24(4):618–625, Epub 2008/07/23 Mol Biol 949:451–471, Epub 2013/01/19
9. Sun S, Francis J, Sapsford KE, Kostov Y, Rasooly 22. Rasooly A, Kostov Y, Bruck HA (2013)
A (2010) Multi-wavelength Spatial LED illumi- Charged-coupled device (CCD) detectors for
nation based detector for in vitro detection of lab-on-a chip (LOC) optical analysis. Methods
botulinum neurotoxin A activity. Sens Actuators Mol Biol 949:365–385, Epub 2013/01/19
B 146(1–8):297–306, Epub 2010/05/26 23. Balsam J, Bruck HA, Rasooly A (2013)
10. Sun S, Ossandon M, Kostov Y, Rasooly A Capillary array waveguide amplified fluores-
(2009) Lab-on-a-chip for botulinum neuro- cence detector for mHealth. Sens Actuators B
toxin a (BoNT-A) activity analysis. Lab Chip 186:711–717, Epub 2013/09/17
9(22):3275–3281, Epub 2009/10/30 24. Balsam J, Rasooly R, Bruck HA, Rasooly A
11. Sun S, Yang M, Kostov Y, Rasooly A (2010) (2014) Thousand-fold fluorescent signal
ELISA-LOC: lab-on-a-chip for enzyme-linked amplification for mHealth diagnostics. Biosens
immunodetection. Lab Chip 10(16):2093– Bioelectron 51:1–7, Epub 2013/08/10
2100, Epub 2010/06/15 25. Balsam J, Ossandon M, Bruck HA, Lubensky I,
12. Zhu H, Ozcan A (2013) Wide-field fluorescent Rasooly A (2013) Low-cost technologies for
microscopy and fluorescent imaging flow medical diagnostics in low-resource settings.
cytometry on a cell-phone. J Vis Exp. Expert Opin Med Diagn 7(3):243–255, Epub
doi:10.3791/50451, Epub 2013/04/23 2013/03/14
 Chapter 11

Mobile Fiber-Optic Sensor for Detection of Oral

and Cervical Cancer in the Developing World
Bing Yu, Vivek Krishna Nagarajan, and Daron G. Ferris

Abstract
Oral and cervical cancers are a growing global health problem that disproportionately impacts women and
men living in the developing world. The high death rate in developing countries is largely due to the fact
that these countries do not have the appropriate medical infrastructure and resources to support the orga-
nized screening and diagnostic programs that are available in the developed world. Diffuse reflectance
spectroscopy (DRS) with a fiber-optic probe can noninvasively quantify the optical properties of epithelial
tissues and has shown the potential as a cost-effective, easy-to-use, and sensitive tool for diagnosis of early
precancerous changes in the cervix and oral cavity. However, current fiber-optic DRS systems have not
been designed to be robust and reliable for use in developing countries. They are subject to various sources
of systematic or random errors, arising from the uncontrolled probe–tissue interface and lack of real-time
calibration, use bulky and expensive optical components, and require extensive training. This chapter
describes a portable DRS device that is specifically designed for detection of oral and cervical cancers in
resource-poor settings. The device uses an innovative smart fiber-optic probe to eliminate operator bias,
state-of-the-art photonics components to reduce size and power consumption, and automated software to
reduce the need of operator training. The size and cost of the smart fiber-optic DRS system may be further
reduced by incorporating a smartphone based spectrometer.

Key words Diffuse reflectance spectroscopy, Fiber-optic sensor, Smartphone, Oral and cervical
cancers

1  Introduction

Each year there are 528,000 cases of cervical cancer with 274,000
deaths worldwide [1]. The majority of these cases (approxi-
mately 85 %) occur in resource-poor countries [1, 2]. Cervical
cancer is the second most common cancer among women. The
highest incidence rates of cervical cancer are found in Central
and South America, Africa, and Asia where rates may exceed 50
cases per 100,000 women. Because of the lack of modern pre-
vention practices in these areas, the global mortality rate of

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_11, © Springer Science+Business Media New York 2015

155
156 Bing Yu et al.

cervical cancer is estimated to increase by 25 % during the next

10 years [2]. While the Pap test has substantially reduced the
rate of cervical cancer, a preventable disease, too many women
continue to suffer even in resource-rich countries.
Oral cancers are part of a group of cancers commonly referred
to as head and neck cancers. More than 650,000 new cases of
head and neck cancer are diagnosed annually worldwide, of which
about 85 % or more are oral cancers [3, 4]. Of these cases, nearly
two-­thirds occur in resource-poor countries where diagnostic and
therapeutic interventions are limited or nonexistent. Approximately
90 % of oral cancers are buried in the crypts of the tonsils or
located at the base of the tongue [5, 6]. Both of these sites are
difficult to screen clinically due to location and the tendency to
elicit an uncomfortable gag reflex in the patient. Virtually no
screening testing is performed for oral cancer, particularly in
resource-poor regions. Because of no screening test, oral cancers
are usually detected only when a large mass is noted. At this late
stage, therapeutic cure becomes exceedingly challenging. Despite
various treatment options, the 5-year survival rate is approximately
25 % [5, 6].
In resource-poor settings, barriers to oral and cervical cancer
screening include lack of access to health care, shortage of quality
medical centers and laboratories, few trained and experienced per-
sonnel, distance to health care clinics, irregular or inconvenient
hours of operation, long waiting times, and lack of affordable
options for follow-up care. Thus, oral and cervical cancer preven-
tion is thus extremely challenged in developing nations.
Diffuse reflectance spectroscopy in the visible wavelength
range (VIS-DRS) is sensitive to the absorption and scattering
properties of epithelial tissue and has shown promise for early diag-
nosis of cancers in the cervix and oral cavity [7–17]. The absorp-
tion and scattering coefficients of epithelial tissues reflect their
underlying physiological and morphological properties [18]. In
the visible band, dominant absorbers in oral and cervical tissues are
oxygenated (HbO2) and deoxygenated hemoglobin (Hb), arising
from blood vessels in the stroma. Light scattering is primarily
caused by cell nuclei and organelles in the epithelium and stroma,
as well as collagen fibers and cross-links in stroma. Neoplastic tis-
sue exhibit significant changes in their physiological and morpho-
logical characteristics that can be quantified optically: stromal layer
absorption is expected to increase with angiogenesis, whereas stro-
mal scattering is expected to go down with neoplastic progression
as extracellular collagen networks degrade [8, 18–22]. Epithelial
scattering has been shown to increase due to increased nuclear size,
increased DNA content, and hyperchromasia [18–20, 23]. VIS-­
DRS has a penetration depth that can be tuned to be comparable
to the thickness of the epithelial layer or deeper to probe both the
epithelial and stromal layers [13, 18, 24]. Therefore, VIS-DRS has
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 157

Fig. 1 Illustration of the principle of VIS-DRS with a fiber-optic probe and the spectrum analysis procedures.
(Reprinted with permission. Copyright Austin Publishing Group 2014)

a great potential to be used as a cost-effective, fast, and sensitive

tool for diagnosis of early precancerous changes in the oral cavity
and cervix.
In DRS, a beam of broadband light emitted from a light
source, such as thermal lamp, white LED or laser diodes, is
launched into a tissue, often through multimode optical fibers, as
illustrated in Fig. 1 (upper left). The photons propagating in the
tissue may experience various events, including elastic scattering,
Raman scattering, absorption and fluorescence. Some of these
photons escape from the tissue surface, after multiple elastic scat-
tering, as reflectance. A detector, often another multimode optical
fiber or fiber bundle, collects a portion of the reflectance and relay
it to an optical spectrometer, where the photons are converted to
a wavelength-dependent intensity distribution of electrons, termed
diffuse reflectance spectrum (upper right). The reflectance spec-
trum is analyzed using a model of photon propagation in tissue
(e.g., diffusion equation [25–27], Monte Carlo simulation [19,
28–30], or empirical model [31]) to extract the absorption and
reduced scattering coefficients (μa(λ) and μs’(λ)), as shown in Fig. 1
(lower right). From the absorption spectrum, the concentrations
of the absorbers (such as oxy-hemoglobin, deoxy-hemoglobin,
beta-carotene, and melanin) can be computed using the Beer–
Lambert law (also known as Beer’s law). The scattering reflects the
158 Bing Yu et al.

tissue morphological information, such as nuclear size and density.

Both tissue compositions and morphological information have
been identified as useful biomarkers for cancer diagnostics.
Current VIS-DRS systems typically consist of a broadband
source, a spectrometer for multispectral detection and a fiber-
optic probe for relaying light to and from the instrument [32].
However, these systems have not been specifically designed to be
robust, reliable, and cost-effective for use in developing countries.
First, uncontrolled probe-to-tissue coupling and pressure can
make it difficult to obtain a reproducible and intact tissue reflec-
tance spectrum by an inexperienced operator. The illumination
and detection efficiency of a fiber-optic probe is very sensitive to
the contact between the fibers and tissue, and thus certain force is
commonly applied to the probe during DRS measurements.
However, studies have found that the diffuse reflectance as well as
the extracted tissue absorption and scattering properties can sig-
nificantly change with varying probe pressure [33–35]. These
changes may be attributed to the compression of the blood vessels
which causes reduced blood flow and alterations in the metabo-
lism of the tissue as well as a change in the density of the scatter-
ers. It is therefore critical to measure and control the probe contact
pressure in order to obtain reproducible and intact tissue physio-
logical and morphological properties.
Second, the lack of a robust, real-time calibration technique
makes the DRS calibration process cumbersome, time-consuming,
and potentially inaccurate. To consistently yield accurate estima-
tion of tissue optical properties, calibration is required to compen-
sate for the wavelength-dependent instrument response, lamp
intensity fluctuations, and fiber bending losses [36, 37]. Current
calibration techniques typically rely on measurements from reflec-
tance standards and/or tissue mimicking phantoms, typically after
the clinical measurements are completed [14, 25, 29]. Because
the calibration is performed at the beginning or end of a study,
real-­time fluctuations, such as lamp drift and fiber bending loss
cannot be compensated by these approaches. They also require
precious time for lamp warm-up and calibration measurements in
a clinical setting.
Finally, most DRS systems use thermal light sources, grating
spectrographs, and cooled CCD cameras. Thermal light sources
have large footprint, short life-time, low power efficiency, and low
coupling efficiency to optical fibers. Spectrometers using grating
spectrographs and cooled CCD cameras have extremely high wave-
length resolution and sensitivity, but are very bulky and expensive
and consume a large amount of electrical power. In addition, a
stable power supply is very often required to operate a thermal
lamp and a CCD camera. The system complexity also makes it
necessary for the operator to have extensive knowledge in optical
spectroscopy and professional training on the instrument and
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 159

probes. Taken together, it is very difficult for DRS systems in their

current forms to be directly used for cancer screening in resource
limited settings.
This chapter describes a mobile VIS-DRS device that is specifi-
cally designed for detection of oral and cervical cancers in resource-­
poor settings. The device uses an innovative smart fiber-optic
probe to eliminate operator bias, battery operable components,
such as LEDs, USB spectrometers, and a laptop, to reduce size and
power consumption, and automated software to reduce the need
of operator training.

2  Materials

2.1  Optical Systems 1. Multichannel USB spectrometer with two visible channels
and Components (range: 400–635 nm, resolution: 1.8 nm), one NIR channel
(range: 750–932 nm, resolution: 0.25 nm), (AvaSpec-2048-
USB2, Avantes, Inc., Broomfield, CO, USA).
2. Cary 300 UV–VIS spectrophotometer (Agilent Technologies,
Inc., Santa Clara, CA, USA).
3. High power white LED module, 1,000 μm plastic fiber pigtail
with FC/PC connector, (Doric Lenses, Inc., Québec, Canada).
4. 850 nm LED, 50/125 μm fiber pigtail, FWHM 50 nm, output
power ~150 μW, FC/PC receptacle, (Appointech Inc., Taiwan).
5. PowerPuck 350 mA LED Drive Module, (LED Supplies, Inc.,
Randolph, VT, USA).
6. DynaOhm constant current LED driver, 25 mA, (LED
Supplies, Inc., Randolph, VT, USA).
7. Large core optical fibers, high-OH silica/silica, 200/220/
239 μm, NA = 0.22, (FVP200220240, Polymicro Technologies,
Inc., Phoenix, AZ, USA).
8. Multimode fibers, 50/125-μm, NA = 0.22 (Corning, Inc.,
Corning, NY, USA).
9. 2 × 2 multimode fiber coupler, 850 nm, Corning multimode
fiber 50/125 μm, coupling ratio 50/50, FC/PC connectors,
(AC Photonics, Inc., Santa Clara, CA, USA).
10. Spectralon® Diffuse Reflectance target, reflectance factor 99 %,
wavelength range 250–2,500 nm, (SRS-99, Labsphere, Inc.,
North Sutton, NH, USA).

2.2  Computer 1. Dell laptop with USB ports, Windows 7, (Dell, Inc., Round
and Software Rock, TX, USA).
2. LabVIEW 2011, (National Instruments Corp., Austin, TX, USA).
3. Matlab 2011b with Curve Fitting Toolbox and Optimization
Toolbox, (Mathworks, Inc., Natick, MA, USA).
160 Bing Yu et al.

2.3  Working 1. Ferrous stabilized human hemoglobin powders, (H7379,

Materials Sigma-Aldrich Co. LLC, St. Louis, MO, USA).
and Solutions 2. 1-μm polystyrene microspheres (07310-15, Polysciences Inc.,
Warrington, PA, USA).
3. Chlorhexidine digluconate solution 20 % in H2O, (Sigma-­
Aldrich, Corp., St. Louis, MO, USA).
4. Biocompatible epoxy EPO-TEK® 301 (Epoxy Technologies,
Inc., Billerica, MA, USA).
5. Epoxy MS-907 Adhesive System, (Miller-Stephenson Chemical
Company, Inc., Danbury, CT, USA).
6. Compressed nitrogen gas tank with a pressure regulator and
digital gauge (precision 0.1 psi).

3  Methods

3.1  Mobile Fiber-­ A smart fiber-optic sensor system has been developed for perform-
Optic Sensor System ing DRS measurements in developing countries. A schematic of
the main components of the smart fiber-optic probe and the instru-
ment to which it is coupled are shown in Fig. 2a. The instrument
consists of an 850 nm LED, a high power white LED, a three-
channel fiber-optic USB spectrometer, and a laptop with custom
LabVIEW and Matlab software. All these components are low
power consuming and can be powered by batteries. The smart
probe integrates a VIS-DRS channel, a self-calibration (SC)
­channel [38, 39], and a diaphragm-based Fabry–Perot interfero-
metric (DFPI) pressure sensor [40] into a single fiber-optic probe.
The SC channel records a calibration spectrum that can be used for
correction of instrument drifts and probe bending loss in real-time,
while the pressure sensor provides feedback on the probe pressure
so that the operator can manually adjust the force applied on it.
The DRS and calibration channels share the white LED as the light
source, while the pressure sensor uses the 850 nm LED as its
light source. The two visible channels of the USB spectrometer
(Spec A and B) are used for detection of the tissue DRS and SC
spectra, respectively. The NIR channel of the USB spectrometer
(Spec C) is used for detection of the interferograms from the DFPI
pressure sensor.
At the distal end, the DRS channel uses a single fiber (the blue
fiber) for DRS detection (connected to Spec A) and six source
fibers (the red fibers), forming a ring around the detection fiber for
tissue illumination. For epithelial mucosal tissue, a source–detector
separation (the radius of the illumination ring) of 640 μm was
used, which provides a simulated sensing depth of 0.5–2 mm. The
SC source fiber (one of the seven red fibers from the white LED)
is looped back, by the 99 % Spectralon diffuse reflectance coating
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 161

Fig. 2 A smart optic sensor system: (a) schematic of the probe and instrument; (b) schematic of the DFPI pressure
sensor head; and (c) photograph of the portable instrument and smart probe. LED light emitting diode, Spec
spectrometer, R1 and R2 reflection, P external pressure, D diameter of the glass diaphragm [41]. (Reprinted
with permission. Copyright the Optical Society of America 2014)

inside the rigid part of the probe housing, into a SC detection fiber
(the pink fiber) that is connected to the Spec B. All fibers for the
tissue and SC channels use the same type of 200/220-μm fiber. Six
inches of the fiber-optic cable immediately after the rigid probe tip
was mounted inside a gooseneck tube so that the probe can be easily
bent to different angles inside an oral cavity.
The DFPI sensor head, as shown in Fig. 2b, is basically a low-­
coherence Fabry–Perot interferometer formed by the cleaved end
face of the lead in/out fiber and the inner surface of a glass dia-
phragm [40]. The broadband NIR light from the 850 nm LED is
launched into a DFPI sensor through a 2 × 2 fiber-optic coupler
whose other input leg (the green fiber) is connected to Spec C
(Fig. 2a). The tip of the unused output leg is immersed in index
matching gel to minimize unnecessary back reflection. All fibers
used in the NIR channel are 50/125-μm multimode fibers. The
lead in/out fiber, fused silica ferrule and tube, as well as the dia-
phragm are bonded together using high temperature epoxy (Epoxy
MS-907). The reflected lights from the two air–glass interfaces (R1
and R2) propagate back to Spec C, generating interferogram.
External pressure P applied on the outer surface of the diaphragm
deflects it towards the fiber tip, thus reducing the cavity length of
the DFPI and causing a shift in the peak positions and a reduction
162 Bing Yu et al.

in the peak density of the interferogram on Spec C. The deflection

of the diaphragm at the center y0 (nm) which represents change in
the cavity length of the DFPI sensor under a pressure P (psi) is:

y0 = (1.74 ´ 10-5 a 4 / h3 ) P ( nm ) (1)

where a (μm) and h (μm) are the effective radius and thickness of
the diaphragm, respectively. By analyzing the interferogram using
a simple fringe peak tracking algorithm [42], the cavity length
(L + y0), and thus the applied pressure can be determined.

3.2  Software A LabVIEW program was used to automate the data collection
and analysis processes so that minimum training is required to
operate the instrument. The LabVIEW program includes the fol-
lowing function modules: (1) initializing the spectrometers,
selecting the probe and target type (phantom or tissue subject),
and loading saved probe configuration information; (2) acquiring
phantom/tissue, calibration and pressure spectra; (3) calling the
interferogram analysis algorithm in Matlab to calculate the probe
pressure P from the NIR spectrum; (4) performing self-calibra-
tion; (5) calling a Monte Carlo inversion model in Matlab to ana-
lyze the tissue spectrum if the pressure is within the preset range;
and (6) displaying the raw spectra as well as calculated probe
pressure and extracted tissue parameters, such as the hemoglobin
concentrations (HbO2, Hb, and THb), oxygen saturation (SO2),
and wavelength averaged reduced scattering coefficient. The time
required to measure and analyze the spectra from a tissue sample
is approximately 1–2 s. If and only if the measured probe pressure
falls within the preset range, e.g., 1–2 psi, the collected DRS and
SC spectra will be stored and analyzed. Otherwise, another scan
will be taken.

3.3  Data Analysis To correct for instrument drifts and fiber bending loss, the DRS
data is analyzed according to the flowchart in Fig. 3. On Day 1, a
single measurement is taken from a Spectralon® diffuse reflectance
target and is used to generate a correction factor:

Fcorr ( l ) = RSp / RSC _ Sp (2)

where RSp is the DRS spectrum measured from the Spectralon®
target and RSC_Sp is the SC spectrum collected concurrently.
Fcorr(λ) accounts for the difference in throughput between the
DRS and SC channels. Another single measurement from a ref-
erence phantom with known μa and μs’ is also required to cali-
brate the measured phantom diffuse reflectance to the Monte
Carlo simulated diffuse reflectance [28]. The Spectralon and
phantom measurements need to be retaken only if the probe or
instrument changes.
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 163

Fig. 3 Data analysis procedures with SC. RPhantom(λ)—DRS spectrum measured

from a reference phantom; RSC_Ph(λ)—SC spectrum collected concurrently with
RPhantom(λ); RTissue(λ)—DRS spectrum measured from the tissue; RSC_TS(λ)—SC
spectrum collected concurrently with RTissue(λ); and Fcorr(λ) is a correction factor
that accounts for the difference in throughput between the two channels [39].
(Reprinted with permission. Copyright SPIE 2011)

On the day for tissue study (Day 2), DRS and SC spectra will
be collected from the tissue only. No separate calibration mea-
surement is necessary. The self-calibrated phantom and tissue
spectra, both corrected by Fcorr(λ), are input into an inverse Monte
Carlo model of reflectance, which extracts the tissue μa(λ) and
μs’(λ) [28]. From μa(λ), the tissue absorber concentrations, such
as [HbO2] and [Hb], can be accurately calculated using the Beer–
Lambert Law. The total hemoglobin concentration THb
(= [HbO2] + [Hb]) and blood oxygenation SO2 (= [HbO2]/THb)
can also be determined.

3.4  Pressure The pressure response of the smart sensor was measured in a pres-
Sensor Test sure test tube. The probe tip was mounted and sealed into the test
tube through a fitting. The gas pressure in the test tube was pro-
vided by compressed nitrogen through a gas pipe, tuned using a
pressure regulator, and monitored by a digital pressure gauge with
a resolution of 0.1 psi within 0–30 psi. Figure 4a shows the mea-
sured air cavity length (L) of a DFPI sensor vs. the applied nitrogen
pressure (P). A pressure sensitivity of 49 nm/psi was calculated
from the L–P curve.
To find out how well the probe pressure on an in vivo tissue
sample can be controlled with the smart sensor, the probe was
brought in contact with a volunteer’s finger. The preset probe
164 Bing Yu et al.

a 112.8 b 5
112.7 L = - 0.049193*P + 112.66
50 out of 170
within 2-3 psi
Sensor Cavity Length (µm)

112.6 4

Measured Probe Pressure (psi)

112.5
3 Max
112.4

112.3
2 Min
112.2

112.1 1

112
0
111.9

111.8 -1
0 5 10 15 0 20 40 60 80 100 120 140 160

Nitrogen Pressure (psi) Number of Measures (<10 ms / scan)

Fig. 4 Pressure test results of a smart fiber-optic sensor: (a) measured DFPI sensor cavity length as a function
of the nitrogen pressure and (b) measured probe pressure during 170 continuous scans with best effort to
maintain a preset range of 2–3 psi. Negative pressure indicates that the probe was not in good contact with
the tissue

pressure was set to 2–3 psi and best effort was taken to maintain
the pressure within this range. A total of 170 pressure measure-
ments have been taken within ~200 s. The measured probe pres-
sures are plotted in Fig. 4b. Among the 170 measurements, 50 fell
within the preset range. The other measurements were either
above or below the preset pressure range which were mainly caused
by the shaking of the hand holding the probe during the scans.
Therefore, it requires an average of approximately four scans to
catch one valid measurement (within the preset pressure range),
which takes less than 40 ms.

3.5  Tissue To evaluate the performance of the constructed smart sensor

Phantom Test system in measuring tissue optical properties, 12 tissue-mimick-
ing phantoms were tested. The phantoms contained variable
concentrations of hemoglobin (7.73–37.92 μM, hemoglobin
powders H7379 dissolved in DI water) as the absorber obtained
by titrating stock hemoglobin solution and fixing the number of
1-μm polystyrene microspheres as the scatterer. The μa(λ) of the
phantoms was independently determined from a spectropho-
tometer (Lambda 35) measurement of a diluted hemoglobin
stock solution, while μs’(λ) was calculated using the Mie theory
for known size, density and refractive index of the scatterers. A
full DRS spectrum (420–630 nm) was measured with a SC spec-
trum concurrently from each phantom using the smart sensor
system. The phantom spectra were analyzed following the proce-
dures in Fig. 3 in which the target phantoms were treated as
tissue spectra. Figure 5 shows the extracted (measured) vs.
expected (theoretical) <μa(λ)> and <μs’(λ)> (wavelength averaged
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 165

Extracted µa (cm-1)
5

1
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5

Expected µa (cm-1)

12
Extracted µs' (cm-1)

10

5 6 7 8 9 10 11 12

Expected µs' (cm-1)

Fig. 5 Extracted vs. expected <μa(λ)> and <μs’(λ)> (wavelength averaged over 420–630 nm) of the 12 phan-
toms. The error bars were resulted from different phantom for reference

over 420–630 nm) of the 12 phantoms. Excellent agreements

between the measured and expected optical properties were
observed.

3.6  Measurement To evaluate the response of oral mucosa to increasing probe pres-
of Oral Mucosa sure, the probe was cleaned using 2 % chlorhexidine digluconate
solution and was then placed in contact with the mucosal tissue
from the inner cheek of a healthy volunteer. During the optical
measurements the volunteer was asked to close the mouth to main-
tain a relatively constant probe temperature through the study.
Room lights were dimmed during the measurements to minimize
the background light. Reflectance measurements were taken at
multiple pressure levels from 0.5 ± 0.5 psi up to 9.5 ± 0.5 psi. Prior
to the reflectance measurements a test run was conducted the vol-
unteer was trained on how to hold the probe, read the pressure
readings on the laptop screen, and maintain the probe pressure
within the preset range. At each pressure level ten pairs of DRS and
SC spectra were acquired sequentially at a 1–2-s interval. The first
measurement started immediately after the desired probe pressure
was reached and best effort was made to maintain the pressure at
the desired value during the ten scans. The probe pressure at each
valid scan was recorded. The DRS spectra measured from the
mucosal tissues analyzed following the procedures in Fig. 3.
166 Bing Yu et al.

30 15
data
10
HbO2 (µM)

20 4th degree fit

Hb (µM)
5
10
0 data
cubic fit
0 −5
0 2 4 6 8 10 0 2 4 6 8 10
Pressure (PSI) Pressure (PSI)

30 120
data data
100
5th degree fit
THb (µM)

20 6th degree fit

SO2 (%)
80

60
10
40

0 20
0 2 4 6 8 10 0 2 4 6 8 10
Pressure (PSI) Pressure (PSI)
16
data
14 6th degree fit
µs (cm−1)

12

10

8
0 2 4 6 8 10
Pressure (PSI)

Fig. 6 The tissue HbO2, Hb, THb, SO2, and <μs’(λ)> measured from the left cheek of a healthy volunteer under
increased probe pressure (ten pressure levels from 0.5 ± 0.5 psi to 9.5 ± 0.5 psi with ten measurements under
each pressure level)

Figure 6 shows the tissue HbO2, Hb, THb, SO2, and <μs’(λ)>
measured from the left cheek of the volunteer with a probe pres-
sure from 0.5 ± 0.5 to 9.5 ± 0.5 psi. A trend of decrease in HbO2,
THb, SO2, and μs’ and increase in Hb with probe pressure has
been identified. The sharpest changes in all tissue parameters
occur within 0–3 psi. These changes may be attributed to the
compression of the blood vessels which causes reduced blood
flow and alterations in the metabolism of the tissue as well as a
change in the density of the scatterers. THb also showed a
rebound above 4 psi which was likely due to the return of blood
to the compressed area.
DRS spectra have also been collected continuously from the
right cheek of the volunteer under a constant probe pressure of
1–2 psi to investigate how the tissue parameters of the oral mucosa
respond to constant probe pressure. Figure 7 shows the typical tis-
sue parameters extracted from 100 repeated scans from the volun-
teer. Significant fluctuations in the THb and SO2 have been
observed in the first 30 s (corresponding to first 20–25 data points).
The scattering coefficients were relatively constant through the
measurements. This suggested that DRS measurements should not
start until the tissue physiology stabilizes.
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 167

HbO2 (µM) 15 10

Hb (µM)
10
5
5

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Measurement # Measurement #
30 100
THb (µM)

SO2 (%)
20
50
10

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Measurement # Measurement #

14
µs (cm−1)

12

10

8
0 20 40 60 80 100
Measurement #

Fig. 7 Tissue parameters measured from the right cheek of a volunteer under a constant probe pressure
of 1–2 psi

4  Notes

1. The human hemoglobin powder used for the phantom test is

dominantly methemoglobin, which has lower solubility than
regular human hemoglobin, such as H0267 from Sigma-­
Aldrich. Better accuracy has been achieved using H0267 with
a similar smart sensor DRS system [41].
2. Temperature fluctuations, which occurs when the probe is
brought in contact with live tissue or the probe–tissue cou-
pling changes due to handshaking, can cause up to a measure-
ment drift up to ±0.5 psi. This measurement uncertainty
reduces the accuracy in probe pressure control.
3. Even though the probe pressure can be accurately measured, it
takes some training for an operator to reduce the number of
invalid scans (pressure out of desired range). Although invalid
scans can be discarded, the perturbation to tissue physiology
takes time to settle down and may affect the following valid
scans.
4. The need for a three-channel spectrometer increases the cost
of the smart sensor system. Even though the total cost of the
smart sensor system is still under $10k and can be further
reduced with mass production. More importantly, the addition
of a pressure sensor and a self-calibration channel makes the
168 Bing Yu et al.

otherwise unreliable and defective DRS measurements diag-

nostically ­useful. From this point of view, the benefits outweigh
the costs of the system.
5. Other emerging technologies, such as smartphone based spec-
trometers, can also be used to further reduce the cost and
improve the mobility of the smart fiber-optic sensor DRS sys-
tem. Smartphones represent one of the most exciting con-
sumer electronics devices that have significantly changed the
way people communicate during the past decade. Smartphones
today, offer enormous computation power on a very compact
platform. Moreover, they are integrated with some key features
such as wireless communication, camera, global positioning
systems (GPS), gyros, and accelerometers, which improve its
functionality. As a direct result of the versatility of smartphones,
they are playing a growing role in optical imaging for medical
and biological applications, such as for spectroscopy [43–46],
microscopy [45, 47, 48], counting molecules [49], and fluo-
rescence imaging of single nanoparticles and viruses [50].
Among these applications, smartphone spectrometers are par-
ticularly interesting because they can provide quantitative diagno-
sis which requires lest medical experience from the operator. In
one study, Gallegos et al. [46] demonstrated the use of a smart-
phone as the detection instrument for a label-free photonic crystal
biosensor and has achieved 0.009 nm accuracy in measuring shifts
in the resonant wavelength of the sensor. Public Lab introduced a
smartphone spectrometer kit at the price of $70 [51].
The cost of wireless technology has also decreased over the
years, thereby making smartphones a more affordable device.
The number of smartphones has increased tremendously through-
out the world in the last 5 years. Based on an UN report, there
are currently 6.8 billion smartphones subscribers all over the
world. More importantly, the report claims that the rate of cell
phone diffusion into developing countries is at 89 % as compared
to 96 % globally. This suggests cell-phone-based devices could
potentially reduce healthcare costs, provide access to advanced
laboratories through wireless communication to remote parts of
the world, and be a field-deployable means for diagnosis, etc.
Most smartphone-based spectrometers are realized through the
attachment of an external module to smartphones. The main
objective for this external module would be to decompose, using
a grating or tunable filter, the incoming broadband light into its
spectral components [44, 45].
However, current smartphone-based device accessories lack
the integration of a fiber-optic attachment which would provide
ease of access to cervical and oral cancer sites. Although Wang
et al. [44] mentioned a fiber-optic external module which could
collect the diffuse reflectance, very limited information has been
 Mobile Fiber-Optic Sensor for Detection of Oral and Cervical Cancer… 169

reported in the literatures about an external fiber-optic module.

Advancements in both smartphones and external m ­ odules/attach-
ments could provide a more robust device for diagnosis.

References
1. Ferlay J et al (2010) Estimates of worldwide comparison to clinical measurements.
burden of cancer in 2008: GLOBOCAN J Biomed Opt 11(6):064027
2008. Int J Cancer 127(12):2893–2917 16. Cardenas-Turanzas M et al (2007) The clinical
2. Vizcaino AP et al (2000) International trends effectiveness of optical spectroscopy for the
in incidence of cervical cancer: II. Squamous-­ in vivo diagnosis of cervical intraepithelial neo-
cell carcinoma. Int J Cancer 86(3):429–435 plasia: where are we? Gynecol Oncol 107
3. Westra WH (2009) The changing face of head (1 Suppl 1):S138–S146
and neck cancer in the 21st century: the impact 17. Freeberg JA et al (2007) The performance of
of HPV on the epidemiology and pathology of fluorescence and reflectance spectroscopy for
oral cancer. Head Neck Pathol 3(1):78–81 the in vivo diagnosis of cervical neoplasia;
4. Parkin DM et al (2005) Global cancer statis- point probe versus multispectral approaches.
tics, 2002. CA Cancer J Clin 55(2):74–108 Gynecol Oncol 107(1 Suppl 1):S248–S255
5. Licitra L et al (2002) Cancer of the orophar- 18. Wang A, Nammalavar V, Drezek R (2007)
ynx. Crit Rev Oncol Hematol 41(1):107–122 Targeting spectral signatures of progressively
6. Dahlstrand HM, Dalianis T (2005) Presence dysplastic stratified epithelia using angularly
and influence of human papillomaviruses variable fiber geometry in reflectance Monte
(HPV) in Tonsillar cancer. Adv Cancer Res Carlo simulations. J Biomed Opt 12(4):
93:59–89 044012
7. Chang VT et al (2009) Quantitative physiology 19. Arifler D et al (2005) Reflectance spectroscopy
of the precancerous cervix in vivo through opti- for diagnosis of epithelial precancer: model-­
cal spectroscopy. Neoplasia 11(4):325–332 based analysis of fiber-optic probe designs to
resolve spectral information from epithelium
8. Muller MG et al (2003) Spectroscopic detec- and stroma. Appl Opt 44(20):4291–4305
tion and evaluation of morphologic and bio-
chemical changes in early human oral 20. Lane PM et al (2006) Simple device for the
carcinoma. Cancer 97(7):1681–1692 direct visualization of oral-cavity tissue fluores-
cence. J Biomed Opt 11(2):024006
9. de Veld DC et al (2005) Autofluorescence and
diffuse reflectance spectroscopy for oral oncol- 21. Pavlova I et al (2009) Fluorescence spectros-
ogy. Lasers Surg Med 36(5):356–364 copy of oral tissue: Monte Carlo modeling
with site-specific tissue properties. J Biomed
10. Sharwani A et al (2006) Assessment of oral Opt 14(1):014009
premalignancy using elastic scattering spec-
troscopy. Oral Oncol 42(4):343–349 22. Hasina R, Lingen MW (2001) Angiogenesis in
oral cancer. J Dent Educ 65(11):1282–1290
11. Amelink A et al (2008) Non-invasive measure-
ment of the morphology and physiology of 23. Georgakoudi I et al (2002) Trimodal spectros-
oral mucosa by use of optical spectroscopy. copy for the detection and characterization of
Oral Oncol 44(1):65–71 cervical precancers in vivo. Am J Obstet
Gynecol 186(3):374–382
12. Rahman M et al (2008) Low-cost, multimodal,
portable screening system for early detection of 24. Liu Q, Ramanujam N (2006) Sequential esti-
oral cancer. J Biomed Opt 13(3):030502 mation of optical properties of a two-layered
epithelial tissue model from depth-resolved
13. Schwarz RA et al (2008) Autofluorescence and ultraviolet-visible diffuse reflectance spectra.
diffuse reflectance spectroscopy of oral epithe- Appl Opt 45(19):4776–4790
lial tissue using a depth-sensitive fiber-optic
probe. Appl Opt 47(6):825–834 25. Zonios G et al (1999) Diffuse reflectance spec-
troscopy of human adenomatous colon polyps
14. Mirabal YN et al (2002) Reflectance spectros- in vivo. Appl Opt 38(31):6628–6637
copy for in vivo detection of cervical precancer.
J Biomed Opt 7(4):587–594 26. Farrell TJ, Patterson MS, Wilson B (1992) A
diffusion theory model of spatially resolved,
15. Arifler D et al (2006) Spatially resolved reflec- steady-state diffuse reflectance for the nonin-
tance spectroscopy for diagnosis of cervical vasive determination of tissue optical proper-
precancer: Monte Carlo modeling and ties in vivo. Med Phys 19(4):879–888
170 Bing Yu et al.

27. Volynskaya Z et al (2008) Diagnosing breast 39. Yu B, Fu HL, Ramanujam N (2011)
cancer using diffuse reflectance spectroscopy Instrument independent diffuse reflectance
and intrinsic fluorescence spectroscopy. spectroscopy. J Biomed Opt 16(1):011010
J Biomed Opt 13(2):024012 40. Yu B et al (2003) Fiber Fabry-Perot sensors for
28. Palmer GM, Ramanujam N (2006) Monte detection of partial discharges in power trans-
Carlo-based inverse model for calculating tis- formers. Appl Opt 42(16):3241–3250
sue optical properties. Part I: Theory and vali- 41. Yu B et al (2014) Diffuse reflectance spectros-
dation on synthetic phantoms. Appl Opt copy of epithelial tissue with a smart fiber-optic
45(5):1062–1071 probe. Biomed Opt Express 5(3):675–689
29. Thueler P et al (2003) In vivo endoscopic tis- 42. Qi B et al (2003) Novel data processing tech-
sue diagnostics based on spectroscopic niques for dispersive white light interferome-
absorption, scattering, and phase function
­ ter. Opt Eng 42:3165–3171
properties. J Biomed Opt 8(3):495–503 43. Zhang J (2008) Spectrometers miniaturized
30. Wang Q et al (2008) Measurement of internal for working with cellular phones and other
tissue optical properties at ultraviolet and visi- portable electronic devices. Int. Patent
ble wavelengths: development and implemen- Application No. PCT/US2008/083613
tation of a fiberoptic-based system. Opt 44. Wang S, Zhou X (2009) Spectroscopic sensor
Express 16(12):8685–8703 on mobile phone. US Patent 7,420,663
31. Reif R, A’Amar O, Bigio IJ (2007) Analytical 45. Chu K et al (2011) Microscopy and spectros-
model of light reflectance for extraction of the copy on a cell phone. In: Bio-optics: design
optical properties in small volumes of turbid and application, Monterey, California United
media. Appl Opt 46(29):7317–7328 States. p BWA3
32. Utzinger U, Richards-Kortum RR (2003) 46. Gallegos D et al (2013) Label-free biodetec-
Fiber optic probes for biomedical optical spec- tion using a smartphone. Lab Chip 13(11):
troscopy. J Biomed Opt 8(1):121–147 2124–2132
33. Reif R et al (2008) Analysis of changes in 47. Sumriddetchkajorn S, Somboonkaew A,
reflectance measurements on biological tissues Chanhorm S (2012) Mobile device-based digi-
subjected to different probe pressures. J tal microscopy for education, healthcare, and
Biomed Opt 13(1):010502 agriculture. In: 9th International conference
34. Ti Y, Lin WC (2008) Effects of probe contact on electrical engineering/electronics, com-
pressure on in vivo optical spectroscopy. Opt puter, telecommunications and information
Express 16(6):4250–4262 technology (ECTI-CON), Phetchaburi
35. Chan EK et al (1996) Effects of compression 48. Tseng D et al (2010) Lensfree microscopy on
on soft tissue optical properties. IEEE J Sel a cellphone. Lab Chip 10(14):1787–1792
Top Quantum Electron 2(4):943–950 49. Ayas S et al (2013) Counting molecules with a
36. Nichols MG, Hull EL, Foster TH (1997) mobile phone camera using plasmonic
Design and testing of a white-light, steady-­ enhancement. ACS Photonics. doi:10.1021/
state diffuse reflectance spectrometer for deter- ph400108p
mination of optical properties of highly 50. Wei Q et al (2013) Fluorescent imaging of
scattering systems. Appl Opt 36(1):93–104 single nanoparticles and viruses on a smart
37. Marin NM et al (2006) Calibration standards phone. ACS Nano 7(10):9147–9155
for multicenter clinical trials of fluorescence 51. Colaner S (2013) Public labs open source smart-
spectroscopy for in vivo diagnosis. J Biomed phone spectrometer let’s you find your wave-
Opt 11(1):014010 length. HotHardware, 18 Nov 2013. http://
38. Yu B et al (2008) Diffuse reflectance spectroscopy hothardware.com/News/Public-Labs-Open-
with a self-calibrating fiber optic probe. Opt Source-Smartphone-Spectrometer-Lets-You-
Lett 33(16):1783–1785 Find-Your-Wavelength/
 Chapter 12

Opto-Fluidics Based Microscopy and Flow Cytometry

on a Cell Phone for Blood Analysis
Hongying Zhu and Aydogan Ozcan

Abstract
Blood analysis is one of the most important clinical tests for medical diagnosis. Flow cytometry and optical
microscopy are widely used techniques to perform blood analysis and therefore cost-effective translation
of these technologies to resource limited settings is critical for various global health as well as telemedicine
applications. In this chapter, we review our recent progress on the integration of imaging flow cytometry
and fluorescent microscopy on a cell phone using compact, light-weight and cost-effective opto-fluidic
attachments integrated onto the camera module of a smartphone. In our cell-phone based opto-fluidic
imaging cytometry design, fluorescently labeled cells are delivered into the imaging area using a disposable
micro-fluidic chip that is positioned above the existing camera unit of the cell phone. Battery powered
light-emitting diodes (LEDs) are butt-coupled to the sides of this micro-fluidic chip without any lenses,
which effectively acts as a multimode slab waveguide, where the excitation light is guided to excite the
fluorescent targets within the micro-fluidic chip. Since the excitation light propagates perpendicular to the
detection path, an inexpensive plastic absorption filter is able to reject most of the scattered light and create
a decent dark-field background for fluorescent imaging. With this excitation geometry, the cell-phone
camera can record fluorescent movies of the particles/cells as they are flowing through the microchannel.
The digital frames of these fluorescent movies are then rapidly processed to quantify the count and the
density of the labeled particles/cells within the solution under test. With a similar opto-fluidic design, we
have recently demonstrated imaging and automated counting of stationary blood cells (e.g., labeled white
blood cells or unlabeled red blood cells) loaded within a disposable cell counting chamber. We tested the
performance of this cell-phone based imaging cytometry and blood analysis platform by measuring the
density of red and white blood cells as well as hemoglobin concentration in human blood samples, which
showed a good match to our measurement results obtained using a commercially available hematology
analyzer. Such a cell-phone enabled opto-fluidics microscopy, flow cytometry, and blood analysis platform
could be especially useful for various telemedicine applications in remote and resource-limited settings.

Key words Flow cytometry, Fluorescent microscopy, Imaging, Cell-phone based diagnostics, Blood
analysis, Telemedicine, Global health

1 Introduction

Blood analysis, such as blood cell count or hemoglobin concen-

tration measurement, is one of the most important clinical tests
for disease diagnosis. For example, complete blood cell count is a

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_12, © Springer Science+Business Media New York 2015

171
172 Hongying Zhu and Aydogan Ozcan

frequently ordered clinical test for evaluating the overall health

condition of patients [1, 2]. A typical blood test requires at least
milliliters of blood sample and the cells can be manually counted
using a hemocytometer with microscopy or automatically counted
with a hematology analyzer or a flow cytometer. Manual count-
ing is the most accessible method but it is tedious and subject to
error and bias. On the other hand, automated counting methods
using for example a blood analyzer or flow cytometry are easy to
perform and highly accurate but require relatively expensive and
bulky instruments, which limit their applications mainly to use in
centralized laboratories with advanced infrastructure. Therefore,
cost-effective translation of these blood analysis technologies into
compact and portable point-of-care devices is highly desirable.
Cell-phone based telemedicine devices form an emerging field
of interest in recent years [3–18]. Nowadays, cell phones are widely
used all over the world, their number reaching almost seven billion
worldwide by the end of 2013 [19]. In addition to this massive
volume, with the rapid growth of the consumer electronics market,
almost all the cell phones are equipped with high-performance
image sensors and digital processors. Thus, cell phones provide
ubiquitous opportunities for cost-effective translation of existing
biomedical technology platforms into mobile health (“mHealth”)
based medical diagnostic devices, which might offer new opportu-
nities for the medical practice in developed as well as developing
countries.
Towards this end, we have demonstrated the integration of
flow cytometry [8] and optical microscopy [17] on a cell phone
using compact, light-weight, and cost-effective opto-mechanical
attachments to perform rapid blood analysis. These opto-
mechanical attachments mainly include simple lenses, plastic color
absorption filters, light-emitting diodes (LEDs) and batteries.
In the flow-cytometry design, the liquid sample is flowing through
a micro-fluidic chip and is continuously transported to the imaging
field of view using a syringe pump. The micro-fluidic chip is posi-
tioned above an inexpensive lens that is in contact with the existing
lens of the cell-phone camera unit. LEDs illuminate the micro-
fluidic chip from two sides through direct butt coupling approach
as shown in Fig. 1. The micro-fluidic chip acts as a multilayered
opto-fluidic waveguide in which the poly(dimethylsiloxane)–
liquid–glass interfaces are surrounded by air. In this opto-fluidic
pumping scheme [20, 21], the guided excitation light propagates
perpendicular to the detection path and as a result of this an
inexpensive plastic absorption filter is able to reject the scattered
excitation and provide the dark-field background needed for fluo-
rescent imaging. While the fluorescently labeled cells are continu-
ously flowing through the micro-fluidic chip, a real-time fluorescent
microscopic movie of the flowing cells is recorded by the cell-
phone camera. The digital frames of these movies are processed to
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 173

a b

c
Glass (or PDMS)
(n1=1.45 (or 1.41))
Liquid (n2=1.33)
Glass (n1=1.45)

Battery Battery
powered Fluorescence emission powered
LED LED
Absorption filter

External lens

CMOS Cellphone camera unit

Fig. 1 (a) Schematic diagram of the opto-mechanical attachment for fluorescent imaging cytometry on a cell
phone. (b) The picture of our opto-fluidic fluorescent imaging cytometer on a cell phone. (c) Schematic illustra-
tion for opto-fluidic pumping/excitation scheme and the cell-phone based optical imaging system. (Reprinted
with permission from Zhu, H., Mavandadi, S., Coskun, A. F., Yaglidere, O., and Ozcan, A., Opto-fluidic Fluorescent
Imaging Cytometry on a Cellphone, Anal. Chem., 83, 6641–6647, 2011. Copyright 2011 American Chemical
Society)

determine the total count of the labeled cells flowing through the
imaging field of view.
Using the same opto-fluidic pumping/excitation scheme, we
are also able to obtain fluorescent microscopic images of stationary
cells loaded into a cell counting chamber or a disposable micro-
fluidic device. These microscopic images can be directly processed
on the smartphone using a custom-written application to deter-
mine the number of cells within a given region of interest.
In this chapter, we introduce the basic operation principles and
the design of this cell-phone based flow-cytometry and optical
174 Hongying Zhu and Aydogan Ozcan

microscopy toolset. Sample handling protocols for the application

of this cell-phone based imaging toolset for blood analysis, includ-
ing white and red blood cell counting and hemoglobin measure-
ments are also presented in detail.

2 Materials

2.1 Optical 1. Sony Ericsson U10I Aino™ (Amazon, USA).

Components 2. Samsung Galaxy SII (Amazon, USA).
for Cell Phone
3. Aspherical lens (f = 4.5 mm, #C230TME-A, Thorlab, NJ, USA).
4. Yellow Kodak Wratten Color Filter (#NT54-467, Edmund
Optics, NJ, USA).
5. Aspherical lens (f = 8 mm, #A240, Thorlab, NJ, USA).
6. Plano-convex lens (f = 15 mm, #45-302, Edmund Optics, NJ, USA).
7. Light emitting diodes (λ = 470 nm, λ = 430 nm, Digikey, Thief
River Fall, USA).

2.2 Chemical 1. SYTO 16 green fluorescent nucleic acid stain (#S7578,

Reagents and Other Invitrogen, Carlsbad, USA).
Components 2. 1× phosphate-buffered silane (BP2438-4, Sigma-Aldrich, St.
Louis, USA).
3. Red blood cell lysing buffer solution (Hydri-Max TM, Sigma-
Aldrich, St. Louis, USA).
4. Poly(dimethylsiloxane) (PDMS) (Dow Corning, USA).
5. Harris Uni-Core hole puncher (Ted Pella, CA, USA).
6. High-frequency plasma generator (model BD-10 AS, Electro-
Technic Products Inc., IL, USA).
7. Tygon tubing (i.d., 0.01 in./o.d., 0.03 in.) (#06418-01, Cole-
Parmer, IL, USA).
8. Syringe pump (Chemxy, TX, USA).

3 Methods

3.1 Cell-Phone First, warm the SYTO16 solution to room temperature and then
Based Flow Cytometry briefly centrifuge it in a microcentrifuge tube to bring the dimethyl
sulfoxide (DMSO) solution to the bottom of the vial. Gently rotate
3.1.1 Fluorescent
the whole blood sample and mix it. After this step, add 20 μL of
Labeling of White Blood
SYTO16 to 200 μL of whole blood. Incubate the whole blood and
Cells in Whole Blood
dye solution mixture in dark for approximately 30 min. To avoid
sedimentation, the sample is gently rotated during incubation.
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 175

Note that no red blood cell lysing step is performed. In addition, since
the intrinsic fluorescence quantum yield of SYTO16 is very low,
the unbound SYTO16 is not separated from the whole blood.
After incubation, dilute the labeled whole blood sample by tenfold
using PBS buffer which is then delivered into our micro-fluidic
chip using a syringe pump with a typical flow rate of 1 μL per min.

3.1.2 Optical Design The cell-phone based flow-cytometry device (see Fig. 1) consists of
for Cell-Phone Based Flow a camera phone and an opto-mechanical attachment, which is
Cytometry designed by autoCAD and printed with a 3D printer using
ABSplus™ modeling thermoplastic material. Sony Ericsson U10I
Aino™, which has an 8 Mpixel color RGB sensor installed on it, is
chosen as the base cell-phone device to build our first-generation
opto-fluidic fluorescent flow cytometry unit (see Note 1). A single
plano-convex lens with a focal length of f2 is placed in front of the
existing lens (f ~ 4.65 mm) of cell-phone camera as shown in Fig. 1
to create an opto-fluidic fluorescent flow cytometry attachment
unit for the cell phone (see Note 2). In the initial experiments, an
aspherical lens with f2 = 4.5 mm is used to get 1× magnification,
which provides us a sufficiently large field-of-view and a good spa-
tial resolution that are needed for white blood cell counting. The
micro-fluidic chip is placed 4.5 mm above the external lens (see
Note 3). During the measurement process, the liquid sample of
interest flows through a PDMS based micro-fluidic chip
(see Fig. 1a, b) using a syringe pump (see Note 4). As shown in
Fig. 1c, blue LEDs are butt-coupled to the two sides of the micro-
channel. In addition to the sample delivery function, this micro-
fluidic chip also works as an opto-fluidic multimode slab waveguide
to guide the excitation light. This opto-fluidic waveguide consists
of three different refractive index layers, i.e., PDMS–liquid–glass.
Since air–glass and air–PDMS interfaces have stronger refractive
index contrast than the glass–liquid or PDMS–liquid interfaces,
the coupled LED excitation light can be tightly confined within
this multimode opto-fluidic waveguide structure to uniformly
excite the labeled cells within the micro-fluidic channel. The fluo-
rescence emission from the labeled cells is then collected perpen-
dicularly to the excitation light path. To further reject the scattered
excitation photons and create a dark-field background for the fluo-
rescent image, an inexpensive plastic filter is placed between the
microchannel and the external lens (see Note 5).

3.1.3 Digital Analysis The Near High Definition (nHD) mode of the Sony Ericsson cam-
of Fluorescent Flow Videos era is chosen to record the fluorescent videos of the flowing blood
cells. Each fluorescent movie that is recorded with the nHD mode
has a resolution of 640 × 352 pixels per frame at a frame rate of
~7–10 frames per second (fps). To count the number of the parti-
cles/cells flowing through our image field of view, these recorded
video frames are post-processed through a PC using a repeating
176 Hongying Zhu and Aydogan Ozcan

sequence of particle detection and tracking algorithms. Briefly, the

tracking algorithm is started in the first frame of the video to detect
all the visible particles using the contour detection algorithm out-
lined in ref. 22. After the initialization, the process of detection
and matching is repeated to find the coordinates of the particles as
they move through the sample chamber. In order to improve the
accuracy of our detection, the number of detected particles and
their velocities over multiple frames (i.e., longer time periods) are
averaged over predefined distances in the chamber. In this case, a
cascade of virtual counters (in the form of lines) is set up at specific
cross-sections of the chamber and each virtual counter monitors
the particles that go through it independent of the others. Tracking
the particles passing through multiple cascade lines allows us to
more accurately count the number of cells passing through the
microchannel (see Note 6).

3.1.4 White Blood Cell (a) WBCs in fresh whole blood samples were first labeled with
(WBC) Density SYTO®16 fluorescent nucleic acid dyes and then diluted the
Measurement Using labeled blood in PBS buffer using the protocol described in
Cell-Phone Based Subheading 3.1.1.
Opto-Fluidic Fluorescent (b) The diluted and labeled blood samples were then flowed
Flow Cytometry through a disposable micro-fluidic chip using a syringe pump
at a flow rate of ~1 μL/min.
(c) The cell-phone camera continuously recorded a video of fluo-
rescent emission from the flowing WBCs. To improve the
counting accuracy, we defined a cascade of five counters with a
distance separation of ~0.3 mm from each other along the
microchannel; and these digital counters dynamically monitored
the number of WBCs that flow through the microchannel.
(d) To further improve our counting accuracy, the program
counted the number of WBCs that passed through each coun-
ter line over a period of 210 frames and the number of the
counted WBCs from these five independent line counters was
averaged to get the final number of WBCs.
(e) Based on the volume flow rate and the average number of the
counted WBCs, the WBC density (#/μL) in the blood sample
can be estimated. The same process was repeated for 5–6 min
with continuous blood flow and the WBC density for each
sample was plotted as a function of time as shown in Fig. 2a, b.
(f) The average WBC density for each blood sample was calcu-
lated by further averaging each one of these dynamic curves
over 5–6 min. Our WBC density measurements were com-
pared against the standard test results obtained from a com-
mercially available hematology analyzer (i.e., Sysmex KX-21N).
As shown in Fig. 2a, b, cell-phone based flow-cytometry WBC
density measurement results match very well with the standard
test results within <5 % error.
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 177

a b
White blood cell density (# of cells/µL)

White blood cell density (# of cells/µL)

7000 10000

6000 9000

5000 8000

4000 7000
Cellphone cytometry WBC counting results Cellphone cytometry WBC counting results
Cellphone cytometry WBC average counting results (4947/µL) Cellphone cytometry WBC average counting results (7354/µL)
3000 Sysmex KX-21N WBC counting results (5000/µL) 6000
Sysmex KX-21N WBC counting results (7800/µL)

2000 5000
0 50 100 150 200 250 300 0 50 100 150 200 250
Time (s) Time (s)

Fig. 2 WBC counting results obtained using our cell-phone based imaging flow-cytometer. (a) For a lower WBC
density sample (5,000 cells/μL) and (b) for a higher WBC density sample (7,800 cells/μL). (Reprinted with per-
mission from Zhu, H., Mavandadi, S., Coskun, A. F., Yaglidere, O., and Ozcan, A., Opto-fluidic Fluorescent Imaging
Cytometry on a Cellphone, Anal. Chem., 83, 6641–6647, 2011. Copyright 2011 American Chemical Society)

(g) Blood samples from 12 different patients were imaged using

the same method to further evaluate the accuracy of this cell-
phone based imaging cytometer. Each sample was monitored
for 5–6 min and the average WBC density of the whole blood
sample was estimated by processing its corresponding fluores-
cent video as detailed above. As shown in Fig. 3, the cell-phone
based imaging cytometer showed comparable results to the
results obtained with a Sysmex KX-21N hematology analyzer.
A correlation coefficient of ~0.93 between the two devices was
obtained for these 12 patients’ blood samples. Bland–Altman
analysis was also performed on these results. As shown in
Fig. 3b, a bias of −339 cells/μL is reported and 95 % limits of
agreement were found as −1,026 cells/μL and 47 cells/μL for
a wide range of WBC concentrations. The negative bias indi-
cates that this cell-phone cytometry platform is slightly under-
estimating the WBC density, which may be due to for example
cell loss that occurs within the micro-fluidic channel inlet/out-
let and tubing (see Note 7).

3.2 Static Cell-phone based flow cytometry, as detailed in the previous

Microscopic Imaging sections, could especially be useful for the detection of rare cell
Based Blood Analysis populations in the blood, which requires the screening of relatively
large volumes of blood. On the other hand, for some other tests
3.2.1 Optical Design
small sample volume per test (e.g., ~10 μL of whole blood per test
for the Cell-Phone Based
from venous or finger prick) is highly desirable. Towards this end,
Blood Analyzer
we also have developed an optical microscopy based automated
blood analyzer running on a cell phone, which can measure white
178 Hongying Zhu and Aydogan Ozcan

a b

Difference of (cell-phone cytometry-

8000 800

hematology analyzer) (# of cells/µL)

counting results (# of cells/µL)

7500
Cellphone cytometry WBC

400 95% limit

7000

6500 0
6000 Bias= −339
−400
5500

5000 −800
4500
−1200 95% limit
4000
4500 5000 5500 6000 6500 7000 7500 8000 4500 5000 5500 6000 6500 7000 7500 8000
Sysmex KX-21N WBC counting Average of WBC counting by
results (# of cells /µL) both methods (# of cells/µL)

Fig. 3 (a) Comparison of WBC density measurement results obtained with our cell-phone based imaging
cytometer against the results of a commercially available hematology analyzer for 12 patients. The error bar
for each data point is generated from blood density measurements taken over 200 s. (b) The Bland–Altman
analysis results, evaluating the accuracy of the cell-phone based cytometry for WBC density measurements
versus a standard hematology analyzer. (Reprinted with permission from Zhu, H., Mavandadi, S., Coskun, A. F.,
Yaglidere, O., and Ozcan, A., Opto-fluidic Fluorescent Imaging Cytometry on a Cellphone, Anal. Chem., 83,
6641–6647, 2011. Copyright 2011 American Chemical Society)

blood cell (WBC) and red blood cell (RBC) density as well as
hemoglobin concentration. The same opto-fluidic excitation
scheme described in the previous sections is also used in the design
of this automated blood analyzer. An Android OS based cell phone
(Samsung Galaxy SII) is chosen as the base for this automated
blood analyzer prototype. The camera of this phone has an 8
MPixel color (RGB) CMOS sensor installed on it and its built-in
lens has a focal length of f ~ 4 mm. The opto-mechanical attach-
ment for this blood analysis platform consists of two parts
(see Fig. 4): (a) A base attachment fixed on the top of the cell-
phone camera unit, which includes two AA batteries and a univer-
sal port for three different add-on components; and (b) Three
different attachments, each designed for WBC and RBC imaging/
counting and hemoglobin concentration measurement. Each add-
on component has an inexpensive plano-convex lens for forming
the imaging system along with the existing lens of the cell-phone
camera in addition to LEDs for sample illumination. Once the spe-
cific add-on component is clicked into the universal port of the
fixed base, the lens on it gets aligned with the fixed lens on the
cell-phone camera similar to the opto-fluidic flow-cytometry sys-
tem described in the previous section. The existing LEDs on the
attachment are then powered up by the batteries on the base to
illuminate the sample of interest.
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 179

Fig. 4 (A-1) and (A-2) Illustration and picture of the blood analysis platform installed on a cell phone. (B-1) and
(B-2) Illustration and picture of the white blood cell counting and density measurement device. (C-1) and (C-2)
Illustration and picture of the red blood cell counting and density measurement device. (D-1) and (D-2)
Illustration and picture of hemoglobin density measurement device. (Reprinted with permission from Zhu, H.,
Sencan, I., Wong, J., Dimitrov, S., Tseng, O., Nagashima, K., and Ozcan, A., Cost-effective and Rapid Blood
Analysis on a Cell-phone, Lab Chip, 13, 1282–1288, 2013. Copyright Royal Society of Chemistry)

The WBC imaging/counting attachment (see Fig. 4b-1, b-2)

is essentially a wide-field fluorescent microscope. First fluorescently
labeled WBCs in diluted whole blood samples are loaded into a
standard non-grid cell counting chamber, which has a channel
depth of ~100 μm to provide a well-defined sample volume
(~10 μL). Eight blue LEDs (~470 nm) directly coupled to the
sides of the counting chamber illuminate the sample from two
sides symmetrically. The cell counting chamber filled with diluted
blood sample acts as a multimode opto-fluidic waveguide to guide
the excitation light for the fluorescently labeled cells inside the
counting chamber. The fluorescent emission from these labeled cells
is captured by the cell-phone camera image system and the unwanted
scattered excitation light is removed by the absorption filter.
180 Hongying Zhu and Aydogan Ozcan

As we discussed earlier, we can adjust the image field-of-view and

magnification by changing the focal length of the external lens in
our design. The lens used for WBC add-on component has a
focal length of f2 = 15 mm, which provides an overall system
demagnification factor of f2/f = 3.75. Therefore, this optical
imaging geometry has a modest spatial resolution but a rather
wide field-of-view [7]. For example it enables us to count the
labeled WBCs over a field-of-view (FOV) of ~0.2–1 cm2. As for
the cell overlap probability, for 2,000 WBCs (assuming a mean
diameter of for example ~12 μm for each cell), within an FOV of
>0.2 cm2, the fraction of cells that do not overlap with others on
the sample plane is >95 % [23].
The RBC imaging/counting attachment (see Fig. 4c-1, c-2) is
designed to image unlabeled cells in diluted whole blood samples
using bright field illumination/imaging mode. The lens used in
RBC counting component has a focal length of f2 = 4 mm, which
provides unit magnification and an FOV of ~14 mm2. A single
white light LED is placed ~4 cm away from the counting chamber,
which is able to uniformly illuminate the sample and generate
bright-field images of unlabeled RBCs. Due to spatial aberrations
towards the edges of the image FOV, we only count the cells within
the central region of the image (e.g., ~1.2–2 mm2). As for the RBC
overlap probability, for 1,000 RBCs (assuming a mean diameter of
~7 μm for each cell) within an FOV that is ≥1.2 mm2, the fraction
of cells that do not overlap with others on the sample plane is
≥88 % [23].
The final specific add-on component is for the hemoglobin
concentration measurements (see Fig. 4d-1, d-2). The hemoglobin
concentration of blood samples is measured using absorbance,
which is proportional to the hemoglobin concentration based on
the Beer–Lambert law. To measure the hemoglobin absorbance,
first the transmission light intensity of water sample, used as a ref-
erence is measured, followed by the blood sample of interest. Based
on the hemoglobin absorption peak (between 400 and 450 nm), a
single blue LED emitting at approximately 430 nm is used as our
light source. For each test, first a disposable plastic cuvette filled
with the liquid sample is inserted into the plastic sample holder.
The central region of the cuvette is aligned with a 1 mm diameter
pinhole, which is placed on the sample holder. The LED is in direct
contact with this pinhole so that it illuminates the central region of
the cuvette. Then the sample holder is slid into this attachment,
which has an external lens with a focal length of 8 mm located at
the bottom of the device. Finally the transmission light intensity of
the sample is recorded by the cell-phone camera; based on the
transmission light intensity values from water sample versus blood
sample, the absorbance of the blood sample is calculated, as further
detailed in the next subsection.
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 181

3.2.2 Smart Application Based on an Android phone, we are able to process the captured
for Blood Cell Counting images on the cell phone using a custom designed smart application.
and Hemoglobin This “blood analysis” smart application runs as follows (see Fig. 5):
Concentration
(a) The “blood analysis” application is first initiated by the user by
Measurements
logging into the application and starting a new test.
(b) Once a new test is selected, the user can choose one of the
three options: white blood cell, red blood cell, and hemoglo-
bin measurements. After choosing the type of test, the user will
put the corresponding add-on component onto the cell-phone
base attachment as illustrated in Fig. 4.
(c) After the specific opto-mechanical add-on is inserted to the cell
phone, the user can start to take a picture of the sample of
interest with the cell-phone camera, where the screen of the
phone shows the raw image.
(d) The user can then select the parameters required to process the
raw image. Some of these parameters include: (1) Region of inter-
est (ROI) for analysis. (2) Image pixel size, which is dependent

Fig. 5 Work flow for the smart application installed on the Android cell phone for rapid blood analysis. (Reprinted
with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S., Tseng, O., Nagashima, K., and Ozcan, A., Cost-
effective and Rapid Blood Analysis on a Cell-phone, Lab Chip, 13, 1282–1288, 2013. Copyright Royal Society
of Chemistry)
182 Hongying Zhu and Aydogan Ozcan

on the magnification of the imaging system (determined by the

focal length of the external lens); (3) Blood sample dilution fac-
tor; and (4) Cell counting chamber height (determining the
volume of the sample). All these parameters are input to
the application manually by the users, which can be adjusted
based on various cuvette types and dilution factors, etc. Finally,
the user clicks on the “Analyze” button to start the image anal-
ysis process on the phone.
(e) The test results calculated by the smartphone are then dis-
played on the phone screen. The cell count results are reported
as the “number of cells per μL” and the hemoglobin concen-
tration measurement is reported as “gram (g) per dL”. All
these test results can be stored in the cell-phone memory card,
uploaded to a central database/server or be directly shared
with healthcare professionals.

3.2.3 Imaging Process (a) For the WBC counting process (see Fig. 6), the ROI is cropped
Methods for the Blood Cell from the whole field-of-view of the raw fluorescent image.
Counting and Hemoglobin (b) The cropped region is converted from RGB (red, green, blue)
Concentration channel into HSV (hue, saturation, value) space.
Measurements
(c) The saturation channel is extracted, which has the maximum
intensity contrast between the fluorescent labeled WBCs and
the dark-field background.

Fig. 6 Block diagram of the digital processing steps that are implemented on the cell phone for WBC concen-
tration measurements. (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S., Tseng, O.,
Nagashima, K., and Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip, 13, 1282–
1288, 2013. Copyright Royal Society of Chemistry)
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 183

Fig. 7 Block diagram of the digital processing steps that are implemented on the cell phone for RBC concentra-
tion measurements. (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S., Tseng, O.,
Nagashima, K., and Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip, 13, 1282–
1288, 2013. Copyright Royal Society of Chemistry)

(d) The binary image is generated by applying an intensity threshold

to the raw image. Furthermore, based on the size of the cells
and their connectivity, we can locate and count the fluores-
cently labeled cells.
(e) Using the number of pixels, the pixel size and the counting
chamber height, the image volume of the selected ROI can be
calculated. Finally the cell density of the sample is computed as
C = (N × F)/V, where C is the cell concentration, N is cell
count per ROI, F is the sample dilution factor, and V is the
sample volume within the ROI.
(f) The same image processing algorithm running on the smart-
phone is also applied to calculate the RBC concentration of
blood samples. As detailed in Fig. 7, instead of the saturation
channel, this time the “value” channel is extracted for the RBC
images since it provides a better image contrast for bright-field
microscopic images. The other elements of the RBC counting
algorithm are the same as in WBC concentration measure-
ments described above.
(g) In the hemoglobin measurement process (see Fig. 8), the trans-
mission light intensity through a water sample is recorded by the
cell-phone camera (to be used as a reference) and then the light
intensity through the lysed blood sample of interest is recorded.
184 Hongying Zhu and Aydogan Ozcan

Fig. 8 Block diagram of the digital processing steps that are implemented on the cell phone for hemoglobin
concentration measurements. (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S.,
Tseng, O., Nagashima, K., and Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip,
13, 1282–1288, 2013. Copyright Royal Society of Chemistry)

The same ROI of each captured image is cropped and is

converted from RGB to grayscale. The mean values of all the
pixels within this selected ROI are extracted from these gray-
scale images. Finally the hemoglobin absorbance value (A) can
be calculated as A = log(IB/IW), where IW is the water trans-
mission value, and IB is the blood sample transmission value.
The hemoglobin concentration of an unknown blood sample
can be estimated using a calibration curve calculated for this
system, which is discussed in Subheading 3.2.6.

3.2.4 White Blood Cell 1. Whole blood samples from anonymous donors with WBC
Density Measurement concentrations ranging from ~3,000/μL up to ~12,000/μL
Based on Static were tested with our cell-phone based imager.
Microscopic Imaging 2. For each test, 10 μL of whole blood sample was mixed with
85 μL PBS buffer and 5 μL of nucleic acid staining (SYTO16)
at room temperature for ~20 min in dark.
3. Then 10 μL of this diluted and labeled whole blood sample
was loaded into the standard cell counting chamber that has
channel height of 0.1 mm. The counting chamber was placed
flat for approximately 2 min for the cells to sediment, after
which the cell phone took the microscopic image of the
labeled WBCs using the opto-mechanical attachment shown
in Fig. 4b-1. The fluorescent images of static WBCs were digi-
tally processed with our cell-phone application (running on
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 185

a b

Difference of White Blood Cell Counting Results Between

13000 1200

Sysmex KN21 and Cellphone (# of cells per mL)

Cellphone White Blood Cell Counting Results

Mean+1.96SD
12000 Linear fit 1000
11000 Y=0.97x-14, R=0.98
800
y=x
10000 ⫾7% 600
(# of cells per mL)

9000 400
Mean=230
8000 200
7000
0
6000
−200
5000
−400 Mean-1.96SD
4000
−600
3000
−800
4000 6000 8000 10000 12000 4000 6000 8000 10000 12000
Sysmex KN21 White Blood Cell Counting Results Average of White Blood Cell Counting Results
(# of cells per mL) Between Cellphone and Sysmex KN21 (# of cells per mL)

Fig. 9 Automated WBC density measurement results of our cell-phone based blood analyzer. (a) Comparison
of cell-phone based WBC density measurement results against the standard test results obtained using
Sysmex KN21 for 30 different blood samples. Each data point was measured three times. (b) The Bland–
Altman analysis results for evaluation of the accuracy of the cell-phone blood analyzer against a standard
hematology analyzer (Sysmex KN21). (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov,
S., Tseng, O., Nagashima, K., and Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip,
13, 1282–1288, 2013. Copyright Royal Society of Chemistry)

Android OS) using the algorithm detailed in Fig. 6. Around

600–2,500 WBCs within an FOV of ~21 mm2 for each cell-
phone image were counted.
4. Cell-phone based counting results were compared with the
standard test results obtained with a Sysmex KN21 hematol-
ogy analyzer. A correlation coefficient of ~0.98 was obtained
between the two methods for 30 different blood samples.
As shown in Fig. 9a, the absolute error of our cell-phone blood
test results was within 7 % of the standard results obtained with
Sysmex KN21. Similarly, Bland–Altman analysis was also per-
formed on these results (see Fig. 9b), which shows a bias of 230
cells/μL, with 95 % limits of agreement of 955 and −495 cells/
μL for a wide range of WBC concentrations.

3.2.5 Red Blood Cell 1. 1 μL of whole blood was mixed with 999 μL PBS buffer to
Density Measurement prepare the diluted blood sample.
Based on Static 2. 10 μL of this diluted whole blood sample was loaded into the
Microscopic Imaging disposable cell counting chamber, which is the same cuvette as
the one used for the WBC counting.
3. Then the specific RBC opto-mechanical attachment shown in
Fig. 4c-1 was used to take bright-field images of the RBCs
after cell sedimentation. These bright field images of the
RBCs within the sample cuvette were then further processed
186 Hongying Zhu and Aydogan Ozcan

a b

Difference of Red Blood Cell Counting Results Between

4x105

Sysmex KN21 and CellPhone (# of cells per mL)

Linear fit
Cellphone Red Blood Cell Counting Results

6.0x106 5
Y=0.96x+2.5x10 , R=0.98
y=x
3x105 Mean+1.96SD
5.5x106
⫾5% 2x105
(# of cells per mL)

5.0x106 1x105
4
4.5x106 0 Mean= -2.9x10

−1x105
4.0x106
−2x105
3.5x106
−3x105 Mean-1.96SD
3.0x106
−4x105

3.0x106 3.5x106 4.0x106 4.5x106 5.0x106 5.5x106 6.0x106 3.0x106 3.5x106 4.0x106 4.5x106 5.0x106 5.5x106 6.0x106
Sysmex KN21 Red Blood Cell Counting Results Average of Red Blood Cell Counting Results
(# of cells per mL) Between Cellphone and Sysmex KN21 (# of cells per mL)

Fig. 10 Same as Fig. 9, except for RBC density measurements. Each data point in (a) was measured three
times. (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S., Tseng, O., Nagashima, K., and
Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip, 13, 1282–1288, 2013. Copyright
Royal Society of Chemistry)

as detailed in Fig. 7. To ensure a good counting accuracy,

only the RBCs within the central field-of-view of ~1.2 mm2
were counted, which typically had ~400 to 700 RBCs.
4. Cell-phone red blood cell counting results (N = 12) were com-
pared with the standard test results obtained using a Sysmex
KN21 blood analyzer. The two devices showed a decent correla-
tion coefficient of ~0.98 as shown in Fig. 10a. We also performed
Bland–Altman analysis on these results as detailed in Fig. 10b
which revealed a bias of −2.9 × 104 cells/μL with 95 % limits of
agreement of ~2.5 × 105 cells/μL and −3.2 × 105 cells/μL.

3.2.6 Hemoglobin 1. The test cuvette was filled with DI water and measured its
Concentration transmission light intensity over 9 mm2 (1000 × 1000 pixels)
Measurement Based yielding I0.
on Static Microscopic 2. Lysed blood samples were prepared by mixing 10 μL of whole
Imaging blood sample with 90 μL RBC lysing buffer solution to lyse
the RBCs. After the RBCs were lysed completely, another
900 μL of PBS buffer was added to further dilute the lysed
sample.
3. Then this diluted sample was loaded into a new cuvette and
the transmission light intensity was measured over the same
area, this time yielding I.
4. The absorbance value (A) for each blood sample was calcu-
lated by using I0 and I values. An internal linear calibration
curve/equation was generated by measuring 60 different
human blood samples with known hemoglobin concentrations
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 187

that were all measured by Sysmex KN21 hematology analyzer

ranging from 11 to 16 g/dL, which resulted in:
A = − 0.28H + 0.056
where A is the absorbance value and H is the hemoglobin
density of the sample in g/dL. To measure the hemoglobin
concentration in an unknown human blood sample, the smart
application on the Android phone computes the absorbance
value (i.e., A) based on which the hemoglobin density of the
blood sample is calculated using this calibration equation
shown above.
5. Blood tests for 37 human blood samples were performed to
test the efficacy of this approach. The hemoglobin concen-
trations obtained from the cell-phone blood analyzer were
compared with the standard test results measured by a
Sysmex KN21. Figure 11a shows the comparison of these
two methods, achieving a correlation coefficient of 0.92.
As desired, the absolute error of our cell-phone measurements
is <5 %. Figure 11b shows the Bland–Altman analysis,
revealing a bias of 0.036 g/dL with 95 % limits of agreement
of 0.63 and −0.54 g/dL.
6. In this chapter, we review opto-fluidic flow cytometry and
automated blood analyzer designs implemented on smart-
phones, which might enable cost-effective and rapid blood
analysis even in resource limited and remote locations. We
evaluate the performance of these cell-phone based blood ana-
lyzer platforms by measuring the densities of WBCs and RBCs

a b
Difference of Hemoglobin Concentration Results Measured
Cellphone Measured Hemoglobin Concentration (g/dL)

18 1.0
Linear fit
by Sysmex KN21 and Cellphone (g/dL)

Mean+1.96SD
Y=1.05x-0.97, R=0.92
y=x 0.5
16
⫾5%

Mean=0.036
0.0
14

−0.5 Mean-1.96SD
12

−1.0
12 14 16 18 12 13 14 15 16 17
Sysmex KN21 Measured Hemoglobin Average of Hemoglobin Concentration Results Measured
Concentration (g/dL) by Sysmex KN21 and Cellphone (g/dL)

Fig. 11 Same as Fig. 9, except for hemoglobin density measurements. Each data point in (a) was measured
three times. (Reprinted with permission from Zhu, H., Sencan, I., Wong, J., Dimitrov, S., Tseng, O., Nagashima,
K., and Ozcan, A., Cost-effective and Rapid Blood Analysis on a Cell-phone, Lab Chip, 13, 1282–1288, 2013.
Copyright Royal Society of Chemistry)
188 Hongying Zhu and Aydogan Ozcan

as well as hemoglobin concentrations of anonymous human

blood samples, yielding comparable results to standard tests
obtained using a commercially available FDA-approved hema-
tology analyzer. The results obtained with these cell-phone
based blood analysis systems can be stored at the memory card
of the phone or be sent to a remote clinic or hospital providing
various tele-diagnosis opportunities. In addition to blood anal-
ysis, this cell-phone based imaging and cytometry toolset
might have many other potential applications. For example
they could be useful for water quality monitoring in field set-
tings by screening pathogens or be used to image and quantify
other lab-on-chip sensing and diagnostic platforms developed
for various point-of-care applications.

4 Notes

1. In general the same design can be implemented on any type of

recent generation camera phones.
2. The ratio of f/f2 determines the magnification of this imaging
geometry. Therefore, based on the application requirements,
one can adjust the image magnification, resolution, and field-
of-view by changing f2 using different lenses.
3. In this cell-phone attachment, the micro-fluidic chip of inter-
est is placed on the sample tray and it is positioned 4–5 mm
above the external lens. There is no extra focusing or mechani-
cal alignment required in this system. In addition, our cell-
phone based flow cytometer is a low NA imaging system and it
has large depth-of-field (i.e., >50 μm), as a result of which all
the blood cells within the micro-fluidic channel appear in focus
as they are flowing through the channel volume.
4. The PDMS based micro-fluidic chip was fabricated through
standard soft-lithographic fabrication techniques. Briefly, the
PDMS is peeled off from the mold and holes are punched to
form the inlet and outlet. The PDMS and the glass substrate
are then cleaned using high-frequency plasma generator.
Immediately after that, they are bonded and the Tygon tub-
ings are inserted into the micro-fluidic chip inlet/outlet and
finally sealed with epoxy. The dimensions of the micro-fluidic
chip are 44 μm × 3 mm × 15 mm (height × width × length).
5. In the opto-fluidic design, the illumination light source (for
example LEDs) and the plastic absorption filter can be easily
changed to different wavelengths and consequently the system
is compatible with different fluorophores.
6. The algorithm used for particle tracking and detection here is
developed in OpenCV, which is available for the iOS platform
 Opto-Fluidics Based Microscopy and Flow Cytometry on a Cell Phone… 189

as well as the Android OS. Therefore, the same algorithm can

also be implemented on various smartphones and tablets.
7. Another important parameter for an imaging cytometer is the
measurement throughput, which is mainly determined by the
frame rate of the cell-phone’s camera. In our current imple-
mentation, the camera of Sony Ericsson has a relatively slow
frame rate of ~7–10 fps. To further increase our cytometry
throughput, a cell-phone camera with a higher frame rate can
be used, such as LG Dare VX9700, which can achieve a frame
rate of for example 120 fps. This could potentially help us
improve the flow rate and thus the counting throughput by for
example >12-fold, also shortening the imaging time for a
blood sample to less than 20 s per test.

References
1. Beck N (2009) Diagnostic hematology. device development and education applications.
Springer, Berlin PLoS One 6:e17150
2. Mellors JW, Munoz A, Giorgi JV, Margolick 10. Martinez AW, Philips ST, Carrilho E, Thomas
JB, Tassoni CJ (1997) Plasma viral load and SW, Sindi H, Whitesides GW (2008) Simple
CD4+ lymphocytes as prognostic markers of telemedicine for developing regions: camera
HIV-1 infection. Ann Intern Med 126: phones and paper-based microfluidic device for
946–954 real-time, off-site diagnosis. Anal Chem 80:
3. Wurm EM, Wellenhof RH, Wurm R, Soyer HP 3699–3707
(2008) Telemedicine and teledermatology: 11. Zhu H, Sikora U, Ozcan A (2012) Quantum
past, present and future. J Dtsch Dermatol Ges dot enabled detection of Escherichia coli using
6:106–112 a cell-phone. Analyst 137:2541–2544
4. Pamplona VF, Mohan A, Oliveira MM, Raskar 12. Mudanyali O, Dimitrov S, Sikora U,
R (2010) NETRA: interactive display for self- Padmanabhan S, Navruz I, Ozcan A (2012)
evaluation of an eye for visual accommodation Integrated rapid-diagnostic-test reader plat-
and focal range. Proc. of Frontiers in Optics, form on a cellphone. Lab Chip 12:2678–2686
2010, Paper FTuB4. 13. Preechaburana P, Gonzalez MC, Suska A,
5. Breslauer DN, Maamari RN, Switz NA, Lam Flippini D (2012) Surface plasmon resonance
WA, Fletcher DA (2009) Mobile phone based chemical sensing on cell phones. Angew Chem
clinical microscopy for global health applica- Int Ed 12:11585–11588
tions. PLoS One 4:e6320 14. Coskun A, Wong J, Khodadah D, Nagi R, Tey
6. Tseng T, Mudanyali O, Oztoprak C, Isikman A, Ozcan A (2012) A personalized food aller-
SO, Sencan I, Yaglidere O, Ozcan A (2010) gen testing platform on a cellphone. Lab Chip
Lensfree microscopy on a cell-phone. Lab Chip 13:636–640
10:1787–1792 15. You DJ, Park TS, Yoon J (2013) Cell-phone-
7. Zhu H, Yaglidere O, Su T, Tseng D, Ozcan A based measurement of TSH using Mie scatter
(2010) Cost-effective and compact wide-field optimized lateral flow assays. Biosens Bio-
fluorescent imaging on a cellphone. Lab Chip electron 40:180–185
11:315–322 16. Navruz I, Coskun AF, Wong J, Mohammad S,
8. Zhu H, Mavandadi S, Coskun AF, Yaglidere Tseng D, Nagi R, Phillips S, Ozcan A (2013)
O, Ozcan A (2011) Opto-fluidic fluorescent Smart-phone based computational microscopy
imaging cytometry on a cellphone. Anal Chem using multi-frame contact imaging on a fiber-
83:6641–6647 optic array. Lab Chip 13:4015–4023
9. Smith ZJ, Chu K, Espenson AF, Rahimzadeh 17. Zhu H, Sencan I, Wong J, Dimitrov S, Tseng
M, Gryshuk A, Molinaro M, Dwyre DM, Lane O, Nagashima K, Ozcan A (2013) Cost-
S, Mattews D, Wachsmann-Hogiu S (2011) effective and rapid blood analysis on a cell-
Cell-phone-based platform for biomedical phone. Lab Chip 13:1282–1288
190 Hongying Zhu and Aydogan Ozcan

18. Balsam J, Rasooly R, Bruck HA, Rasooly A 21. Kiesel P, Beck M, Johnson N (2011)
(2014) Thousand-fold fluorescent signal Monitoring CD4 in whole blood with an opto-
amplification for mHealth diagnostics. Biosens fluidic detector based on spatially modulated
Bioelectron 51:1–7 fluorescence emission. Cytometry A 79A:
19. International Telecommunication Union, 317–324
Market information and statistics (2013). 22. Suzuki S, Abe K (1985) Topological structural
http://www.itu.int/en/ITU-D/Statistics/ analysis of digital binary image by border fol-
Documents/facts/ICTFactsFigures2013.pdf lowing. Vis Graph Image Process 30:32–46
20. Schmidt O, Bassler M, Kiesel P, Knollenberg 23. Ozcan A, Demirci U (2008) Ultra wide-field
C, Johnson N (2007) Fluorescent spectrometer- lens-free monitoring of cells on-chip. Lab Chip
on-a-fluidic-chip. Lab Chip 7:626–629 8:98–106
 Chapter 13

Optofluidic Device for Label-Free Cell Classification

from Whole Blood
Tsung-Feng Wu and Yu-Hwa Lo

Abstract
A unique optofluidic lab-on-a-chip device that can detect optically encoded forward scattering signals is
demonstrated. With a unique design of a spatial mask that patterns the intensity distribution of the illumi-
nating light, the position and velocity of each travelling cell in the flow can be measured with submicrom-
eter resolution, which enables the generation of a cell distribution plot over the cross section of the
channel. The distribution of cells is highly sensitive to its size and stiffness, both being important biomark-
ers for cell classification without cell labelling. The optical-coding technique offers an easy route to classify
cells based on their size and stiffness. Because the stiffness and size of neutrophils are distinct from other
types of white blood cells, the number of neutrophils can be detected from other white blood cells and red
blood cells. Above all, the enumeration of neutrophil concentration can be obtained from only 5 μL of
human blood with a simple blood preparation process saving the usual steps of anticoagulation, centrifuga-
tion, antibody labelling, or filtering. The optofluidic system is compact, inexpensive, and simple to fabri-
cate and operate. The system uses a commodity laser diode and a Si PIN photoreceiver and digital signal
processing to extract vital information about cells and suppress the noise from the encoded optical scatter-
ing signals. The optofluidic device holds promise to be a point-of-care and home care device to measure
neutrophil concentration, which is the key indicator of the immune functions for cancer patients undergo-
ing chemotherapy.

Key words Optical-coding technique, Neutrophils, White blood cells, Microfluidics, Optofluidic,
Point-of-care, Labsel-free detection

1 Introduction

Counting and classification of human blood cells are clinically

important as the first steps of diagnosis of immunodeficiency, infec-
tion, and inflammatory responses. The impedance-based flow
cytometers, aka Coulter counter hematology analyzers, have been
the main tools for such purposes. To achieve high sensitivity and
detection accuracy, many Coulter hematology analyzers are
equipped to detect the fluorescent signals from the blood cells.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_13, © Springer Science+Business Media New York 2015

191
192 Tsung-Feng Wu and Yu-Hwa Lo

However, to enable point-of-care blood analysis, both the devices

and the operation procedures for blood analysis have to be
simplified significantly.
Microfluidic devices have been attractive to point-of-care
applications because they help miniaturize the device, provide a
platform for integration of different functions, and are suitable for
low cost, high volume production. The current designs of micro-
fluidic Coulter counter for blood enumeration essentially follow
the same principals as the bench-top machines except for a much
reduced form factor [1]. However, the electrode design in such
microfluidic devices has produced more nonuniform electric-field
distribution, higher noise, and compromised reliability compared
to the bench top systems. On the other hand, the microfluidic
device based on optical scattering of cells has drawn significant
attention because of its high precision and promising performance.
Both being label free techniques, the optical scattering signals con-
tain richer and cleaner cell features than cell impedance signals and
impose no restrictions on the channel geometry. In contrast, good
quality cell impedance signals can be obtained only when the cell
volume occupies an appreciable portion of the total detection vol-
ume, thus limiting the channel dimension and raising concerns on
throughput and clogging.
Label-free cell detection methods are stressed for point-of-care
applications because label-free detection reduces not only the
operation and system costs but also the complexity of the sample
preparation process. Most existing methods that require cell label-
ling require long and complicated sample preparation procedures
such as anticoagulation, staining, antibody labelling, centrifuga-
tion, or filtering. These preparation steps need additional equip-
ment and chemicals, add possibilities for errors, increase the test
time, and are often too complicated to be performed by people
without medical trainings.
As an effective label-free optical detection method on a micro-
fluidic platform, the optical-coding technique is invented to per-
form blood cell counting and classification. The device is able to
enumerate and classify cells with minimal invasiveness and simple
sample preparation procedure. Because of the hydrodynamic iner-
tial effect of cells travelling in the microfluidic channel, the equilib-
rium position of cells inside the microchannel is determined by the
size and stiffness of cells, which can be used as effective biomarkers
to classify cell types. Due to the parabolic velocity profile of a lami-
nar flow in the microchannel, the cell position inside the channel is
directly related to its travelling speed under a given flow rate. The
optical coding method, realized by a specially designed mask that
patterns the light illumination of the travelling cells, produces both
the velocity and position of each cell inside the microchannel. Such
information can then be converted to cell size and cell stiffness.
Without the needs for sheath flows, one can increase the sample
 Optofluidic Device for Label-Free Cell Classification from Whole Blood 193

flow rate significantly to achieve high detection throughput. As a

result, the test can be conducted within 10 min with a small amount
(e.g., 5 μL) of whole blood that is heavily (200–1,000 times)
diluted to avoid clogging and coagulation problems without
reagents used in typical clinical blood process.
To demonstrate the feasibility of point-of-care applications,
this work has specifically been focused on the enumeration of neu-
trophils from the whole blood because the number of neutrophils
provides important and direct information of patient’s immune
functions. For example, a healthy person has a white blood cell
concentration in the range of 4,000 and 11,000 per microliter of
blood, and 50–70 % of white blood cells are neutrophils. Clinically,
neutropenia is diagnosed if a patient has less than 1,500 neutro-
phils per microliter of blood. Severe neutropenia increases the pos-
sibility of infections which can be life threatening especially for
chemotherapy patients [2]. Because neutrophils are known to be
softer and larger than most other white blood cell groups such as
lymphocytes and monocytes, hydrodynamically neutrophils take
their equilibrium position closer to the center of the microfluidic
channel and travel faster than other white blood cells.
Figure 1 shows a schematic illustration of the equilibrium
positions in the microchannel for larger and softer cells relative to
smaller and stiffer cells. To summarize, the equilibrium positions

Fig. 1 (a) Equilibrium position of cells affected by the inertial effect, which mainly depends on the size of cells.
The larger cells experience a greater lift force that pushes them toward the center of the microchannel. (b) and
(c) show the effect of drag force affected by the velocity gradient (UA–UB) normal to the cell travel direction. The
higher deformability of softer cells makes them nearer the center of the microchannel, thus travelling at a
higher speed, than those stiffer cells
194 Tsung-Feng Wu and Yu-Hwa Lo

of cells are determined by the balance of two forces, lift force due
to the inertial effect and fluidic dynamic drag force. Larger cells
experience a greater inertial force which pushes the cells farther
away from the channel wall (Fig. 1a). On the other hand, the
greater deformability of softer cells enables a better alignment of
the cell orientation with the streamline, so softer cells experience
less drag force than stiffer cells and tend to be closer to the center
of the channel (Fig. 1b, c). As a result, the combined effects of
larger size and greater deformability separate neutrophils from
the other WBCs by their travel speed and their location in the
microchannel.
The optically encoded microfluidic device demonstrated in this
work consists of a straight microfluidic channel with spatial pat-
terns that modulate the incident light illuminating each cell pass-
ing the patterns. The spatial pattern converts the forward scattering
signal of each cell into an intensity-modulated temporal signal
according to its velocity and position [3]. As shown in Fig. 2, cells
within the microfluidic channel produce forward scattering signals
of unique features in their waveforms. For example, if the trajec-
tory of cells passing through the sensing area is close to the side
wall of the microchannel (e.g., trajectory 1 or 3 in Fig. 2), the wave
form of the forward scattering signal will be different from the
forward scattering signal for cells travelling through the center of
the microchannel (e.g., trajectory 2 in Fig. 2). Therefore, from the
ratio of W2 to W1 or W3 to W4 in the forward scattering signal, we
can find the trajectory of each cell. Here we use two pairs of
patterned slits, (W1,W2) and (W3,W4), as opposed to a single pair to

Fig. 2 Illustrations of different scattering signal waveforms from cells taking different travel paths in the micro-
fluidic channel. The white areas represent the transparent trapezoidal slits. The bases of each trapezoidal slit
are 50 μm and 100 μm, as indicated in the figure. The solid (1), semi-dashed (2) and dashed (3) red lines are
exemplary cell paths over the sensing area. The waveforms of the forward scattering signal of the cells taking
the paths 1, 2, 3 are illustrated on the right, with the characteristic widths W1, W2, W3, W4 in the signals. Such
information, as explained in the text, allows us to find the travel speed and position for each single cell within
the microchannel (Color figure online)
 Optofluidic Device for Label-Free Cell Classification from Whole Blood 195

attain more reliable results from the redundancy. Although the

spatial modulation technique has been explored by other research
groups [4], the previous methods detect fluorescent signals from
labelled cells using expensive and sophisticated optics and photo-
detectors such as photomultiplier tubes (PMTs). In this study, the
optically encoded forward scattering signal can be readily detected
by an off-the-shelf silicon PIN photodetector. Besides signal inten-
sity, we are particularly interested in the signal waveform from
which we can obtain cell speed and position, which, as explained
previously, give rise to information of cell size and stiffness for cell
classification. Since the scattering signals have been encoded, digi-
tal signal processing algorithms can be applied not only to extract
the cell features but also enhance the signal-to-noise ratio [5].
In the following we introduce the label-free optical-coding
technique for detection and enumeration of neutrophils in detail
[6]. The device and the entire detection system to implement the
optical-coding technique can be highly compact and compatible
with the platforms of mobile devices, thus holding the promise for
point-of-care and personalized health care.

2 Materials

2.1 System 1. Laser beam (λ = 488 nm (or 650 nm) diode laser, 40 mW,
for Detection Spectra-physics).
of Space-Time 2. Silicon photoreceiver (PDA36A, Thorlabs, NJ, USA).
Coded Optical
3. Monochrome CCD camera, C-mount (XC-ST30, Sony, Japan).
Scattering Signal
4. XY translation stage with standard micrometers (460A-XY,
Newport, CA, USA).
5. Optical table.
6. LabView Express (2010, National Instruments, TX, USA).
7. DAQ Broad (BNC-2110, National Instruments, TX, USA).
8. Syringe pump (11 Elite, Harvard Apparatus, MA, USA).

2.2 Sample Solution 1. 10× red blood cell lysing buffer for multi-species (00-4300-54,
eBioscience, CA, USA) is diluted with deionized water to form
1× lysing buffer before RBC lysing.
2. 10 mM ethylenediaminetetraacetic acid (EDTA), 1 wt% bovine
serum albumin (BSA), and 1× phosphate buffered saline (PBS)
are mixed as the cell suspension solution.
3. Whole blood is collected from healthy donors by using a finger
prick (see Note 1).

2.3 Device Material 1. Polydimethylsiloxane kit (Sylgard 184, Dow Corning) was
purchased directly and mixed prior to use.
196 Tsung-Feng Wu and Yu-Hwa Lo

3 Methods

3.1 Setup A 488 nm (or 650 nm) laser light is introduced from the bottom
of the Microfluidic of the microfluidic device. The CCD camera is used to assure align-
Device ment between the laser beam and the spatial patterns on the glass
slide to let the laser beam illuminate cells inside the microchannel.
The schematic and photo image of the system are shown in Fig. 3a, b,
respectively. The laser light and CCD camera have been aligned
with each other. In the actual operation, the microfluidic device is
placed on the XY translational stage for device alignment. An
attenuator is inserted in front of the laser source to avoid saturating
the CCD camera. By adjusting the XY translation stage, the entire
pattern is uniformly illuminated by the laser beam as shown in
Fig. 3c. Figure 3c also shows the exact features of the spatial mask
pattern, where four trapezoidal slits are aligned with the 100 μm

Fig. 3 (a) The schematic of the optical-coding system, consisting of a PIN photoreceiver, a diode laser, a micro-
fluidic device, and a computation system for data acquisition and analysis. (b) The actual image of the optical-
coding system shows the detail of the optics setup. (c) The CCD image of the spatial patterns on the microfluidic
device under uniform laser illumination. The red arrows show the channel width. (d) A time-domain encoded
forward scattering signal from a cell collected by a silicon PIN photodetector (Color figure online)
 Optofluidic Device for Label-Free Cell Classification from Whole Blood 197

wide channel, labelled in the image. A movable Si PIN photore-

ceiver is placed between the microfluidic device and the CCD cam-
era at an angle of 2° ~ 5° from the incident beam to collect the
forward scattering signal from cells. The transimpedance gain of
the photoreceiver is set at 40 dB. The sample exits the device
through the outlet at the end of the channel.
Because the incident light is encoded by the patterned slits, the
waveform of the scattering light from each cell contains informa-
tion of the cell position and speed as shown in Fig. 2. All forward
scattering signals are collected by a photoreceiver and recorded by
LabView. The sampling rate for data recording is at 10,000 sam-
ples per second. A higher sampling rate did not produce better
results but increase the data processing time. Figure 3d shows an
example of the forward scattering signal of a single cell. Custom
MATLAB algorithms are applied to extract the forward scattering
intensity, position along the x-axis direction, travelling speed across
the sensing area, and position along the y-axis of each individual
cell. All such information is used to identify and enumerate the
neutrophils from human whole blood.

3.2 Fabrication 1. Fabrication of SU-8 Mold for PDMS replica.

of the Microfluidic (a) PDMS replica design: A microfluidic channel is designed
Device to be a 5 cm long and 100 μm wide with one inlet for
sample introduction and one outlet for the exit. The device
layout is drawn by AutoCAD and printed out on a trans-
parency mask (CAD/Art Services, Inc., Oregon, USA).
(b) Negative photoresist, SU8-2050, is spun on a 4-in.
mechanical silicon wafer at 3,500 rpm for 40 s.
(c) Soft bake: As-spun wafer is placed on a hotplate at 65 °C
for 3 min and then heated to 95 °C for 6 min.
(d) Exposure: After cooling down to 65 °C from 95 °C, the
wafer is UV exposed at a dosage of 150 mJ/cm2. The
designed pattern is photolithographically transferred from
the mask to the photoresist.
(e) Post-exposure bake (PEB): The wafer is placed on a hot-
plate at 65 °C for 1 min and then heated to 95 °C for
another 6 min, followed by cooling down to 65 °C.
(f) Development: SU-8 developer is used to remove the unex-
posed photoresist, followed by rinsing with isopropyl alco-
hol (IPA) and deionized water (see Note 2).
(g) Hard bake: the as-developed wafer is placed in the oven
and heated to 150 °C for 30 min.
(h) Characterization: The thickness of final SU-8 mold is
measured to be 45 μm.
198 Tsung-Feng Wu and Yu-Hwa Lo

2. PDMS Replica.
(a) PDMS (Sylgard 184, Dow Corning) is mixed with the cur-
ing agent at the weight ratio of 15:1.
(b) The pre-polymer mixture is poured onto the SU-8 mold
on silicon wafer and baked in the oven at 60 °C for 4 h.
(c) The PDMS replica can be easily cut off with the blazer for
further use.
3. Mask Pattern.
(a) Mask silts design: Four transparent trapezoidal slits are
designed as shown in Fig. 2. Each individual trapezoidal
slit has its base lengths of 50 μm and 100 μm and is sepa-
rated from the neighboring silts by 50 μm. The total pat-
tern length for the sensing area is 450 μm.
(b) Negative photoresist, NR9-1500, is spun on a 1 × 2 in.
glass substrate at 4,000 rpm for 30 s.
(c) After the soft bake at 150 °C for 1 min, the glass slide is
exposed under I-line for 90 s.
(d) As-exposed glass slide is placed on a hotplate at 100 °C for
1 min, followed by immersion of the slide in the developer
solution for 6 s. Rinse with deionized water after
development.
(e) A spatial mask is formed by sputtering a Ti/Au
(100 nm/200 nm) metal film on the patterned glass, fol-
lowed by the lift-off process to form the four trapezoidal
slit patterns.
4. Assembly of microfluidic devices.
(a) The PDMS replica from Subheading 3.2, step 2 and the
patterned glass slide from Subheading 3.2, step 3 are
treated with ozone plasma.
(b) After the ozone plasma treatment, the PDMS replica is
bonded onto the patterned glass slide with careful align-
ment to form the microfluidic device.

3.3 Determination 1. Use a finger prick to take 5 μL of whole blood from healthy
of the Number donors. The amount of blood is calibrated with a micropipette.
of Neutrophils 2. The 5 μL whole blood is first diluted with 1 mL of buffer solu-
tion described in Subheading 2.2, item 2.
3. Without any centrifuge, 500 μL of RBC lysing buffer is added
to 150 μL of diluted whole blood sample to dilute the white
blood cell concentration by 870 times.
 Optofluidic Device for Label-Free Cell Classification from Whole Blood 199

a 140 b 20
19 30
120 18

Position along
100 V>30cm/s 17

y-axis (mm)
16 20
Cell Count

80 15
60 14
13 10
40
12 Neutrophils
20 11
10 0
0
0 20 40 60 80 100
15 20 25 30 35 40
Velocity (cm/s) Position along x-axis (mm)
c400000
Neutrophils
300000
SSC-A

200000

Monocytes
100000 Lymphocytes

0
0 300000 600000 900000 1200000 1500000

FSC-A

Fig. 4 (a) Travelling speed histogram of white blood cells from 5 μL RBC-lysed whole blood sample. The histo-
gram indicates a distinguishable population of neutrophils due to their high travel speed resulted from the
higher deformability. The event, which speed is less than 30 cm/s, is the signal from other WBC types plus RBC
residues since the sample did not completely filter out by the centrifuge. (b) The spatial distribution of white
blood sample over the cross section of the channel. A separate band for neutrophils (as the gating shows)
separates them from the other cells. Due to the property of symmetry, only the upper half of the channel cross
section is shown. (c) Scatter plot of 5 μL diluted whole blood sample measured with a commercial flow cytom-
eter (Accuri C6)

4. After 10 min in the ambient, the Tygon tubings (ID: 0.02 in.
OD: 0.03 in.) are inserted into the inlet and outlet of
microfluidic devices. The diluted white blood cell sample is
directly introduced into the microfluidic device by using a
syringe pump (Pump Elite 11, Harvard Apparatus, MA,
USA) at the flow rate of 75 μL/min.
5. Record signals from the white blood cell sample for 90 s, which
means a sample volume of 112.5 μL is interrogated.
6. Digital signal processing is applied to obtain the speed of each
cell, as shown in Fig. 4a. The neutrophils have faster speed
than other WBCs.
7. In combination with the cell position information, the velocity
of cell can be further converted into cell distribution over the
cross section of the channel, as shown in Fig. 4b.
200 Tsung-Feng Wu and Yu-Hwa Lo

Table 1
Summary of test results from 870× diluted RBC-lysed blood samples
using our device and a commercial flow cytometer (Accuri C6)

Trial I II III IV
Our device (neutrophil 5.80 7.74 5.72 6.20
counts/μL)
Accuri C6 (neutrophil 4.38 7.09 6.39 6.77
counts/μL)

8. Repeat the above procedures for four samples from the healthy
donors and measure the samples with a commercial flow
cytometer (Accuri C6) as shown in Fig. 4c.
9. The comparison between the optofluidic device and the com-
mercial flow cytometer is shown in Table 1.

4 Notes

1. When doing the finger prick, keep fingers as warm as possible

in order to extract the sufficient volume of blood.
2. If a white film appears when rinsing with IPA, the unexposed
photoresist is underdeveloped. Re-immersing the wafer in
SU-8 developer can remove the white film and complete the
development process.

References
1. Cheung K, Gawad S, Renaud P (2005) Impedance 4. Keisel P, Bassler M, Beck M, Johnson N (2009)
spectroscopy flow cytometry: on-chip label-free Spatially modulated fluorescence emission from
cell differentiation. Cytometry A 65A:124–132 moving particles. Appl Phys Lett 94:041107
2. Sipsas NV, Bodey GP, Kontoyiannis DP (2005) 5. Mei Z, Wu TF, Pion-Tonachini L, Qiao W, Zhao
Perspectives for the management of febrile neu- C, Liu ZW, Lo YH (2011) Applying an optical
tropenic patients with cancer in the 21st century. space-time coding method to enhance light scat-
Cancer 103:1103–1113 tering signals in microfluidic devices.
3. Wu TF, Mei Z, Pion-Tonachini L, Zhao C, Qiao Biomicrofluidics 5:034116
W, Arianpour A, Lo YH (2011) An optical- 6. Wu TF, Mei Z, Lo YH (2012) Optofluidic
coding method to measure particle distribution device for label-free cell classification from whole
in microfluidic devices. AIP Adv 1:022155 blood. Lab Chip 12:3791–3797
 Chapter 14

A Wearable Sensing System for Assessment of Exposures

to Environmental Volatile Organic Compounds
Cheng Chen, Francis Tsow, Xiaojun Xian, Erica Forzani,
Nongjian Tao, and Raymond Tsui

Abstract
A portable chemical sensing system that integrates sample preconcentration, separation and detection as
well as wireless communication functionalities into a compact, wearable format can provide continuous
and real-time monitoring of volatile organic compounds in the environment. The sensing modality relies
on tuning forks coated with molecularly imprinted polymers that, in conjunction with sample preconcen-
tration, offer selective detection down to parts-per-billion levels. The use of capillary columns allows indi-
vidual components of complex mixtures to be detected at these highly sensitive levels even in the presence
of interferents. The wireless capability facilitates the utilization of a paired smartphone as the user interface
as well as a vehicle for additional processing and storage of the measured data. This integrated approach
offers a cost-effective and reliable platform for personal exposure assessment.

Key words Environmental monitoring, Chemical sensors, Volatile organic compounds, Wearable
detection system, Wireless sensing, Personal exposure assessment

1 Introduction

Exposure to airborne toxicants can result in both short-term safety

risks and long-term health issues. The ability to quantitatively detect
and identify trace chemicals in air is therefore a key requirement in
the field of environmental monitoring. However, the combination
of target analytes at difficult-to-measure concentration levels with a
multitude of interfering chemicals in ambient air and other environ-
mental factors make the monitoring a challenging task.
Volatile organic compounds (VOCs) such as alkyl, aromatic
and chlorinated hydrocarbons (HCs) are known to be harmful to
humans [1, 2]. Traditionally, monitoring of VOCs in the environ-
ment has relied on collecting air samples in the field and then per-
forming a laboratory analysis via gas chromatography–mass
spectrometry (GC-MS) [3]. While this method is able to separate
and identify each component in a gas mixture, the equipment used

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_14, © Springer Science+Business Media New York 2015

201
202 Cheng Chen et al.

is typically expensive and bulky, its operation labor intensive, and

incapable of supplying data in real-time. Alternate methods have
been developed to provide portability and real-time detection
capability. One approach uses colorimetric tubes containing sens-
ing materials to which specific analytes bind and result in a color
change [4]. The binding reaction, however, is irreversible, so this
method is for single use only and unsuitable for continuous moni-
toring. Another approach is to use a device that relies on photo
ionization detection (PID), in which a ultra-violet lamp ionizes gas
molecules of the analyte to provide a current that can be measured
[5]. Stand-alone PID devices, however, can only provide broad-
band detection. For selective sensing, a PID device must be used in
conjunction with filters and separation tubes. Such constraints
limit the practicality of this method.
In recent years, chemical sensors and wearable sensing sys-
tems based on inexpensive, commercially available tuning forks
(TFs) have been developed for the detection of various chemical
vapors including VOCs [6–8]. The TF used is a quartz crystal
resonator with an intrinsic resonance frequency f0 of 32.768 kHz.
This frequency is related to the effective spring constant k of the
TF by the expression, f0 = (1/2π) × √(k/m), where m is the effec-
tive mass. Upon the adsorption of analyte molecules on the
exposed prongs of a TF sensor, the increase in m causes a change
in f0. When this frequency change is accurately measured and the
appropriate calibration has been applied, the concentration of the
detected analyte can be determined. For the detection of VOCs,
the sensitivity, selectivity, and response time of these compact TF
sensors can be improved significantly by applying a thin coating
of a molecularly imprinted polymer (MIP) to the prongs of the
TFs [8, 9]. These MIPs are synthesized to provide sites that bind
to the target analytes primarily via π-π and van der Waals interac-
tions, thus resulting in selective but reversible binding [10].
These MIP-coated TFs can achieve a mass detection capability of
a few pg per mm2 of sensor surface area, yielding a detection limit
for VOCs at the ppb level.
To enable usage of these modified TF sensors in a real-time
and portable fashion in the field, a multiple-TF module can be
integrated with control and communications circuitry in a compact
and lightweight device to wirelessly interface with a smartphone
[8]. The phone provides a graphical user interface, capabilities to
store and further process the collected detection data, as well as
geolocation and additional telecommunications functionalities.
Further enhancements are accomplished by adding to the portable
device a miniaturized chromatographic column to provide separa-
tion of complex environmental samples [9] and a preconcentration
module to increase sensitivity for difficult to detect samples [11].
 A Wearable Sensing System for Assessment of Exposures… 203

Particle Preconcentration Separation Detection

filter module module module

Air Interference
inlets V1 filter V2 V3

Zero
Precon-
Electronics
filter V4
centrator circuitry
Pump GC
and
column
batteries

Fig. 1 A schematic representation of an integrated chemical detection system, consisting of modules for: flu-
idic and thermal control (with a pump. filters and valves V1–V4), sample preconcentration, sample separation
(by a gas chromatographic (GC) column), sample detection (by tuning fork sensors), and electronics

These can all be accomplished while maintaining the lightweight

(~0.5 kg) and compact (13 × 10 × 5 cm3) attributes of this approach,
making it a good candidate for use as a wearable unit in occupa-
tional and environmental health monitoring as well as epidemio-
logical studies. The schematic representation of such a highly
integrated device is shown in Fig. 1.
Here, we describe the methods of integrating sample precon-
centration, separation, and detection into a system that is wirelessly
paired with a smartphone for the portable monitoring of VOCs in
various environments. The analytes of interest in this case are
vapors of benzene, toluene, ethylbenzene, and xylenes (BTEX), an
important group of aromatic VOCs that have harmful effects on
human health. Examples of field testing of the system in an indus-
try setting for exposure to petrochemical products and in geo-
graphical mapping of BTEX exposure to local activities are also
described. These examples serve to illustrate the compelling advan-
tages of such a mobile unit over conventional bench-top systems.
The compact size and low weight of the mobile system allow the
user to walk through industry facilities (or contaminated accident
sites) and conduct accurate exposure analysis with high spatial and
time resolution in a very cost-effective manner. Furthermore,
usage of the near-ubiquitous smartphone as a system component
provides a pathway to use each mobile unit as a collection node
whose data can then be transmitted wirelessly over a wide-ranging
cellular communication system. The data can be aggregated and
analyzed, then re-provisioned to supply exposure information that
covers a large and evolving geographical area. This networking
capability may prove invaluable in situations where human health
or safety concerns are at stake.
204 Cheng Chen et al.

2 Materials

2.1 Sample 1. Stainless steel tubing (outer diameter: 0.0625 in., inner
Preconcentration diameter: 0.0335 in.).
Module 2. Graphitized carbon black (Sigma-Aldrich Co. LLC, St. Louis,
MO).
3. Thermally conductive compound (Resbond 920, Cotronics
Corp., Brooklyn, NY).
4. Heating wire (Kanthal Nikrothal 80 Plus, 35 Ω/m; Sandvik
Materials Technology, Clarks Summit, PA).

2.2 Sample 1. Carbowax-coated stainless steel capillary column (Quadrex

Separation Module Corp., Bethany, CT).
2. Cyanopropyl phenyl silicone-coated stainless steel capillary
column (Quadrex Corp., Bethany, CT).

2.3 Sample 1. Quartz tuning forks (resonance frequency: 32.768 kHz;

Detection Module Citizen Finetech Miyota Co., Ltd., order via Newark Corp.,
Chicago, IL).
2. Cartridge for holding multiple tuning forks (custom-made).

2.4 Fluidic 1. Micro vane pump (SP 135 FZ-LC; Schwarzer Precision USA,
and Thermal Control Madison, CT).
Module 2. Miniature three-way valves, total of 4 (LHLA0531211H; The
Lee Co., Essex, CT).
3. Particle filter (custom-made with commercial glass fiber).
4. Zeroing filter (custom-made from activated carbon).
5. Interference filter (custom-made using polyester fiber).
6. PTFE tubing (EW-06605-27, outer diameter: 0.125 in., inner
diameter: 0.0625 in.; Cole-Parmer, Vernon Hills, IL).
7. Connectors (L410-J1A, elbow tube fitting; Value Plastics,
Inc., Fort Collins, CO).

2.5 Electronics 1. Microcontroller (MSP430F2013, Texas Instruments Inc.,

Module Dallas, TX).
2. Custom-made high-resolution frequency counter, with noise
level <0.2 mHz at 0.2-Hz data resolution (see Note 1).
3. Lithium-ion polymer battery (Tenergy V2; Tenergy Corp.,
Fremont, CA).
4. Bluetooth module (RN41-I/RM; Microchip Technology Inc.,
Tempe, AZ).

2.6 User Interface 1. Smartphone (Q9h; Motorola Mobility LLC, Chicago, IL).
2. Application software (custom-made using Microsoft Visual
Studio).
 A Wearable Sensing System for Assessment of Exposures… 205

3 Methods

3.1 Monitoring The preconcentrator is built using a 1-in. section of stainless steel
System Setup tubing with an outer diameter of 0.0625 in. The tubing is packed
with about 3 mg of the graphitized carbon black to provide a high
3.1.1 Preconcentration
density of binding sites for adsorbing VOCs. The outside of the
Module
tubing was coated with a thermal conductive material, and a heat-
ing wire was wrapped around the tubing. Then another layer of the
same thermal conductive material was coated on the heating wire
along with a thermistor. The resistance value of the thermistor is
monitored and converted into a temperature reading by the elec-
tronics module, which is used to turn on and off the heating wire
to control the 300 °C temperature needed to facilitate desorption
of the VOCs that have been collected.

3.1.2 Separation Module The type and configuration of the stainless steel capillary columns
used in the sample separation module are dependent on the
intended air-monitoring application. For fast separation and detec-
tion of VOCs, a Carbowax-coated column and a cyanopropyl phe-
nyl silicone-coated column are connected in series.

3.1.3 Detection Module The MIP used to coat the prongs of the TF sensors is designed to
provide selective detection of mono-aromatic and alkyl hydrocar-
bon VOCs. For the detection of BTEX, a polystyrene-based MIP
is synthesized by polymerizing divinylbenzene. A mixture of ethyl-
benzene and xylene is used as the porogenic solvent.
The MIP coating on the TFs is formed using the following
procedure:
1. The TFs are immersed in toluene containing 10 % (v/v) phen-
yltrimethoxysilane for 24 h to form a hydrophobic surface layer.
2. TFs are then rinsed with toluene and dried with clean air.
3. Next, the TFs are dipped in 20 mM dodecanethiol in isopro-
panol for 60 min to make the quartz surfaces and the elec-
trodes of the TFs hydrophobic.
4. The TFs are dip-coated with the MIP (see Note 2).
5. The TFs are mounted onto the sensor cartridge.

3.1.4 Fluidic The various components in this module are connected to those in
and Thermal Control the previously described modules using flexible PTFE tubing and
Module tube fittings (see Note 3). Air samples are drawn into the monitor-
ing system via a small vane pump (see Note 4). A three-way valve
(V1, Fig. 1) controls whether the air is drawn through a particle
filter (during detection mode) or a zero filter (during calibration
mode). The interference filter absorbs acid vapors from the sampled
206 Cheng Chen et al.

air to avoid interfering with the tuning fork sensors. The pump and
valves, as well as the heater for the preconcentrator, are powered
and controlled through the electronics module described below.

3.1.5 Electronics Module Custom designed and built circuitry is used to control all the mod-
ules within the integrated device. Figure 2 shows a block diagram
for the operation of the detection and wireless communication
modules. The circuitry includes a precision digital counter to mea-
sure the responses of the TFs in the sensor array, and a Bluetooth
chip for the wireless communication with the smartphone that is
utilized as the user interface.
1. Each TF in the sensor array is driven with an oscillator driver
chip and then multiplexed sequentially to the clock input of a
microcontroller (MCU1, Fig. 2). Microcontroller 1 uses the
clock signal to generate a pulse with duration of 32,768 clock
cycles. This pulse is sent to the inputs of the analog-to-digital
converter (ADC) of a second microcontroller (MCU2).
2. The input signal is compared against the reference clock cycle
with a much higher frequency (in the several-MHz range) that
is generated by MCU2. This allows the shift in the resonant
frequency of a TF sensor to be measured with very high resolu-
tion (a few mHz).

Functional Block Diagram

MCU1
ADC/
GPIO

MUX

Computation UART BT
RX TX
TX RX
Sensor
Array
MCU2

Fig. 2 A block diagram that illustrates the operational scheme for the sample detection and wireless commu-
nication modules. Microcontrollers (MCUs) handle conversion of the sensor data from analog into digital for-
mat, as well as control of components in the fluidic and thermal control module. The Bluetooth (BT) chip
communicates with its counterpart within the smartphone used as the user interface to the detection system
 A Wearable Sensing System for Assessment of Exposures… 207

Fig. 3 (a) Photograph of the enclosure holding all the components of the integrated detection system, with the
smartphone utilized as the user interface on top. (b) Photograph of the smartphone screen showing a separa-
tion chromatogram of a BTEX mixture detected by the system

3. The TF data is converted into a serialized digital data stream

and transmitted wirelessly to the smartphone by the comput-
ing function and a Universal Asynchronous Receiver/
Transmitter (UART) built into MCU2.
4. The electronics module also controls the operation of the
pump and the valves in the Fluidic and Thermal Control mod-
ule and the heating of the Preconcentrator module of the inte-
grated device.
5. Three lithium-ion polymer batteries are used to power all the com-
ponents for up to 8 h of continuous environmental monitoring.

3.1.6 User Interface A smartphone running the Windows operating system (Motorola
Q9h) is used as the interface to the integrated device (Fig. 3a). A
custom application written using Java programming language is
loaded onto the phone to control the electronics in the integrated
device. After receiving the sensor data, the smartphone also pro-
vides the functions of additional processing, displaying, and stor-
age of the data.

3.2 Device Operation Operation of the device, as described below, is fully automated.
Initiation of a measuring cycle as described below is accomplished
by pushing a button on the smartphone.

3.2.1 Preconcentration 1. The pump turns on and the valves open to the configuration to
draw an air sample through the particle filter into the precon-
centrator. The duration of this step is adjusted to account for
the anticipated VOC concentration. The use of a longer pre-
concentration period usually produces larger signal peaks for a
given analyte concentration in the sampled air, due to the lon-
ger time for accumulation and interaction of the analyte with
the carbon black (see Note 5).
208 Cheng Chen et al.

2. Valves are then switched to draw air cleaned by the zero filter
to flow through the separation column and over the TF sen-
sors for 1 min. This allows the sensors to establish a baseline
reading.

3.2.2 Desorption 1. The temperature of the preconcentrator is raised to 300 °C in

1 min by applying power to the heating wire. This releases the
adsorbed VOCs back into a gaseous state.

3.2.3 Sample Injection 1. The valves are switched to flow filtered clean air through the
hot preconcentrator to transport the VOC vapors to the sepa-
ration column.
2. This injection stage has a 15-s duration.

3.2.4 Data Collection 1. As the components of the air sample are separated in the capil-
and Analysis lary column, they are transported to the TF sensors for
measurement.
2. Sensor responses are wirelessly sent to the smartphone where
the application software analyzes the data and displays the
measured results (Fig. 3b).

3.2.5 System Cleaning 1. Upon completion of the previous stage, the preconcentrator is
heated again.
2. Filtered clean air is flowed through the system to flush it clean
and to prepare it for the next measuring cycle.

3.3 Field Testing The wearable detection system was tested at the Shell Oil refinery
in Martinez, California. In this case, the system was configured
3.3.1 Exposure
with a long capillary column (19 m) to counter the high concen-
to Petrochemical Products
trations of hydrocarbon interferents present in the atmosphere in
in an Industry Setting
addition to the BTEX vapors targeted for measurement. The sys-
tem was operated as described in Subheading 3.2. As can be seen
in Fig. 4a, low levels of benzene vapor (5 ppb) can be clearly
resolved from n-hexane with a concentration 1,000 times higher
also present in the sampled air. The separation (elution) time here
was ~8 min. To detect all the individual components of BTEX,
longer elution times are needed (Fig. 4b). Note that all three iso-
mers of xylene can be resolved. An analyte with low vapor pressure
or high interaction with the coating of the capillary column will
result in a longer elution time. For a given analyte, slower elution
times are also encountered when longer capillary columns are used
to obtain higher separation resolution of complex samples with
multiple components, as in the case here for gasoline vapors.

3.3.2 Mapping BTEX The environment at a number of locations with varying amounts
Exposure to Local Activities of traffic and different activities in the metropolitan area of Phoenix,
and Environmental Arizona were tested for BTEX using the wearable detection
Conditions system. The system was operated as described in Subheading 3.2.
 A Wearable Sensing System for Assessment of Exposures… 209

a b
9 benzene 35

30 benzene

Frequency Change (Hz)

Frequency Change (Hz)

toluene m-xylene
25
6
20
ethyl-
n-hexane 15 benzene p-xylene
3 o-xylene
10

5
0 0
5 6 7 8 10 20 30 40 50 60
Time (min) Time (min)

Fig. 4 Separation and detection of air samples containing BTEX vapors and related interferents. (a) Detection
of benzene at a concentration of 5 ppb, after the separation of n-hexane interferents with a concentration of
5 ppm. (b) The separation chromatogram for all the other components of the BTEX vapors sampled. Note that
all three isomers of xylene are resolved

The preconcentration time used was 10 min. Among the locations

tested are schools in residential areas, a highway with busy traffic,
and the Phoenix Airport (Fig. 5). Using the capabilities of the
Global Positioning System (GPS) built into the smartphone, the
measured data can be served up using graphical interfaces such as
Google Earth or Google Map to provide location-stamped data
values as shown in Fig. 5. As expected, air samples at the schools
were found to have zero or low levels of BTEX components
(1.5 ppb or less). The benzene level at a gas station was measured
to be 5 ppb, while 10 ppb of toluene was present inside a gambling
casino. These results show the effectiveness of a highly portable
integrated detection system for monitoring how personal exposure
varies with activity and location.

4 Notes

1. Equipment with similar capability is available commercially but

is in the form of desktop instrumentation. The significantly
larger size precludes the portable/wearable format afforded by
the current system. The higher cost and non-user friendly
interface also make the desktop-type of equipment unsuitable
for use by end consumers or technical personnel with only
minimum training.
210 Cheng Chen et al.

1- School

> 5 ppb
B = 0 ppb E = 0 ppb
T = 0 ppb X = 0 ppb 7- Casino
2- Mall - Parking

B = 0 ppb E = 0 ppb
T = 10 ppb X = 0 ppb
B = 2.4 ppb E= 0 ppb
T = 0 ppb X= 0 ppb
6- Gas Station
3- Airport

B = 5 ppb E = 0 ppb
T = 0 ppb X = 0 ppb
B = 1.5 ppb E = 0 ppb
T = 0 ppb X = 0 ppb 4- School 5- Highway Traffic

B = 1.5 ppb E = 0 ppb B = 2.4 ppb E = 1.8 ppb

T = 0 ppb X = 0 ppb T = 0 ppb X = 0 ppb

Fig. 5 Correlation of BTEX exposures to sampling locations (within the metropolitan area of Phoenix, Arizona)
on Google Map, making use of GPS-based locationing capability in the smartphone. High levels of benzene and
toluene are found at a gas station and inside a gambling casino, respectively

2. The mass of the MIP coatings ranged from 0.5 to 3 μg, resulting
in a variation of within 20 % for the sensitivities of the TFs.
3. It is important to minimize dead volumes associated with the
connections between all the components involved in gas
handling.
4. An air flow rate of 8 mL/min is found to be optimal for the
current system configuration.
5. For a BTEX mixture with a concentration of 20 ppb for each
component, a preconcentration time of 20 min is sufficient. In
the case of a component concentration of 10 ppm, there is no
need for preconcentration.

Acknowledgments

This research was supported by the National Institute

of Environmental Health Sciences of the National Institutes of
Health, through the Genes, Environment and Health Initiative
Program (#5U01ES016064).
 A Wearable Sensing System for Assessment of Exposures… 211

References
1. Gauderman WJ, Avol E, Lurmann F, Kuenzli 7. Ren M, Forzani ES, Tao N (2005) Chemical
N, Gilliland F, Peters J, McConnell R (2008) sensor based on microfabricated wristwatch
Childhood asthma and exposure to traffic and tuning forks. Anal Chem 77:2700–2707
nitrogen dioxide. Epidemiology 16:737–743 8. Tsow F, Forzani E, Rai A, Wang R, Tsui R,
2. Carrieri M, Tranfo G, Pigini D, Paci E, Mastroianni S, Knobbe C, Gandolfi AJ, Tao
Salamon F, Scapellato ML, Fracasso ME, NJ (2009) A wearable and wireless sensor sys-
Manno M, Bartolucci B (2010) Correlation tem for real-time monitoring of toxic environ-
between environmental and biological moni- mental volatile organic compounds. IEEE
toring of exposure to benzene in petrochemi- Sensors J 9:1734–1740
cal industry operators. Toxicol Lett 192: 9. Iglesias RA, Tsow F, Wang R, Forzani ES, Tao
17–21 N (2009) Hybrid separation and detection
3. Schweigkofler M, Niessner R (1999) device for analysis of benzene, toluene, ethyl-
Determination of siloxanes and VOC in land- benzene, and xylenes in complex samples. Anal
fill gas and sewage gas by canister sampling and Chem 81:8930–8935
GC-MS/AES analysis. Environ Sci Technol 10. Lieberzeit PA, Gazda-Miarecka S, Halikias K,
33:3680–3685 Schirk C, Kauling J, Dickert FL (2005)
4. RAE Systems (http://www.raesystems.com/ Imprinting as a versatile platform for sensitive
products/colorimetric-gas-detection-tubes) materials – nano-patterning of the polymer
5. Freedman AN (1980) The photoionization bulk and surfaces. Sens Actuators B
detector: theory, performance and application 111–112:259–2632
as a low-level monitor of oil vapour. J 11. Chen C, Tsow F, Driggs Campbell K, Iglesias
Chromatogr A 190:263–273 R, Forzani E, Tao N (2013) A wireless hybrid
6. Boussaad S, Tao NJ (2003) Polymer wire chemical sensor for detection of environmental
chemical sensor using a microfabricated tuning volatile organic compounds. IEEE Sensors J
fork. Nano Lett 3:1173–1176 13:1748–1755
 Chapter 15

Quantitative Point-of-Care (POC) Assays Using

Measurements of Time as the Readout:
A New Type of Readout for mHealth
Gregory G. Lewis and Scott T. Phillips

Abstract
A paper-based microfluidic device was used to quantitatively detect active enzyme analytes in samples at
mid to low femtomolar levels. The device uses a hydrophobic oligomer that controls the wetting proper-
ties of the paper within the device. When the target analyte is present within the sample, the oligomer
depolymerizes, thus switching the paper to hydrophilic, allowing for the sample to wick through the
device. Measuring the time for the sample to wick to a control region relative to an assay region within the
device results in sensitive, quantitative measurements of the target enzyme (e.g., alkaline phosphatase or
β-d-galactosidase). This device requires the use of only a timer for quantifying a target analyte, and thus
the platform may be appropriate for use in resource-limited environments, where access to expensive diag-
nostic equipment is limited. A smartphone with integrated application software (the software has yet to be
developed) could be used for timing the assay and for relating the time measurement to the quantitative
readout for the assay. In future versions of this assay, it should be possible to configure the smartphone to
start and stop the time-based measurement to further simplify the assay for the user.

Key words Microfluidics, Paper-based microfluidics, Point-of-care diagnostics, Quantitative assays,

Depolymerization, Developing world

1 Introduction

Point-of-care (POC) diagnostics are designed for detecting and

often quantifying analytes in settings that lack laboratory infra-
structure. Quantitative POC assays are more difficult to perform
than qualitative assays, yet they provide information that often is
necessary to make an accurate diagnosis. Performing quantitative
POC assays in resource-limited environments is challenging, as
many of the components typically required (e.g., power supplies,
electronic readers, clean water, temperature control, etc.) are
unavailable, or available in limited supply. Due to these restric-
tions, the World Health Organization (WHO) has outlined seven

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_15, © Springer Science+Business Media New York 2015

213
214 Gregory G. Lewis and Scott T. Phillips

guidelines for developing POC assays for use in resource-limited

environments, including assays that are equipment free [1–4].
As a step towards this goal, we recently described a POC assay
platform (a paper-based microfluidic device) that is capable of
quantifying active enzyme analytes by measuring the time required
for one region of a paper-based device to turn color relative to
when another region turns color (Fig. 1a) [5]. The assay and the
microfluidic device operate as follows. The device has an entry
point for addition of the sample, as well as hydrophilic channels of
paper that split the sample into two equal directions (i.e., layer 3,
Fig. 1b). Layer 3 also includes (1) buffer salts that are re-dissolved
by the sample to control the pH of the fluid as it distributes through
the device, as well as (2) the cofactor MgCl2, which is needed by
certain enzyme analytes (other cofactors could be added for differ-
ent enzyme targets).
In the left-hand channel in Fig. 1b, the sample re-dissolves a
substrate for a target enzyme analyte (Fig. 1d), beginning in layer
4. If the target enzyme is in the sample, it reacts with this substrate
and causes release of one molecule of glucose per enzymatic reac-
tion. Once the sample continues through layer 4 into layer 5, it
encounters bead-bound glucose oxidase (dark blue in Fig. 1b, d),
which remains immobilized in the fibers of the paper. The glucose
oxidase oxidizes the released glucose and generates hydrogen per-
oxide as the sample travels laterally in layer 5 of the device. Once
the sample reaches the vertical conduit on the far left-hand side of
the device in Fig. 1b, it encounters an oligomer (Fig. 1c) that is
hydrophobic and thus alters the wetting properties of the paper
from hydrophilic to hydrophobic [6]. In the absence of hydrogen
peroxide, the sample travels slowly through this hydrophobic
region, but in the presence of hydrogen peroxide, the oligomer
converts into hydrophilic products through a cascade depolymer-
ization reaction, thus switching the wetting properties of the paper
from hydrophobic back to hydrophilic [7]. This switching reaction
amplifies the effects of hydrogen peroxide on the flow rate through
layer 4 by converting a single large hydrophobic oligomer into
many small hydrophilic products. This switching reaction also
allows the sample to pass through the layers containing the oligo-
mer with a rate that depends on the concentration of hydrogen
peroxide in the sample, which ultimately reflects the concentration
of the target enzyme analyte. Once the sample passes through the
layer containing the oligomer, it continues to travel in the vertical
direction until it re-dissolves dried green food coloring and carries
the highly colored solution to the top layer where the bright green
color becomes visible.
The right-hand channel in the cross-section in Fig. 1b contains
the same reagents in the same order as the left region, with the
exception of bead-bound glucose oxidase. In this right region, the
enzyme analyte (if present) will react with the substrate deposited
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 215

Fig. 1 Depiction of a paper-based microfluidic device used for quantitatively detecting active enzyme analytes,
as well as the reagents used in the assays. (a) Picture of the paper-based microfluidic device. The device is
made from stacked layers of patterned paper held together with spray adhesive and then laminated. The
dimensions of the paper portion of the device are 20 mm × 10 mm × 1.8 mm, the black regions are hydropho-
bic wax and the white regions are hydrophilic paper. (b) Cross-section of the device shown in (a), along the
white dotted line. The left channel in (b) is the assay channel and the right channel is the reference channel.
(c) A hydrophobic oligomer is incorporated into the device (layer 4 in (b)) and provides an amplified signal for
detection events. The oligomer depolymerizes from head-to-tail in response to hydrogen peroxide that is gen-
erated from the detection event (d). (d) Reagents are incorporated into the device to provide selective detection
of the target enzyme analyte. The enzyme reacts with the reagent and produces glucose, which is converted
to hydrogen peroxide within the device
216 Gregory G. Lewis and Scott T. Phillips

into the channel and generate glucose, but hydrogen peroxide will
not be generated, therefore hydrogen peroxide will not be present
to react with the oligomer. Hence, the time required for the sam-
ple to pass through this control region (and carry the green color
to the top of the device) depends on the temperature and humidity
under which the assay is conducted, as well as on the viscosity of
the sample. These factors will affect sample distribution rates in the
left region as well, and therefore this control region normalizes the
output of the assay for the effects of these variables on sample dis-
tribution. This normalization is implemented by measuring the
time required for the right region to turn green relative to when
the left region turns green.
The assay is capable of detecting low to mid femtomolar con-
centrations of active enzymes in buffered solutions (Fig. 2) and low
picomolar concentrations in complex fluids, such as serum [5]. The
platform should be compatible with other classes of analytes as well.

Fig. 2 Calibration curve for alkaline phosphatase, which is a model enzyme that
we used to demonstrate typical results for a quantitative POC assay. The data
points are an average of three measurements and the error bars reflect the stan-
dard deviations of these averages. The inset shows an expanded view of the
bracketed region. The data for alkaline phosphatase are provided in black,
whereas results for samples containing catalase (instead of alkaline phospha-
tase) are provided in green. Similar results for β-d-galactosidase are provided in
blue. Catalase and β-d-galactosidase were used as control enzymes because
they belong to different classes of enzymes and do not react with the substrate
for alkaline phosphatase (i.e., glucose-6-phosphate) that is included in layers 4
and 5 of the device. Limits of detection are dependent on the length of the hydro-
phobic oligomer, but range from 128 fM to low nM
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 217

This chapter describes our current method for fabricating the

paper-based microfluidic devices that support this type of
quantitative assay for enzymes. Extensions of this assay into the
realm of mHealth should be straightforward either by (1) using
the timing function on a smartphone to time the assay (with inte-
grated software for interpreting the measurement), or (2) using an
attachment for the smartphone that enables the phone to control
the timing and readout without intervention by the user. Both of
these capabilities are currently being developed. The methods
described in this chapter provide the foundation upon which these
mHealth capabilities are based.

2 Materials

2.1 Patterning 1. CleWin Layout Editor (freely accessible on the internet

Microfluidic Channels through WieWeb Software) (see Note 1) and Adobe®
into Paper Illustrator® (Adobe Systems).
2. Whatman Chromatography Paper No. 1 or Whatman Filter
Paper No. 1 (20 × 20 cm sheets) (see Note 2).
3. Wax printer (e.g., Xerox Phaser 8560).
4. Oven that can be held at constant temperature (150 °C) with
shelves. The flat shelves must be larger than 20 cm × 20 cm (l × w).
5. Black wax (e.g., Xerox Phaser 8560 Solid Ink (108R00726)).

2.2 Depositing Assay 1. Hydrophobic oligomer (Fig. 1c) [6, 8] dissolved in organic
Reagents into Paper solvent (e.g., ethyl acetate or tetrahydrofuran) at fixed concen-
Microfluidic Devices trations (see Note 3). Concentrations of oligomer ranged from
0.1 to 100 mM depending on the oligomer used. These solu-
tions are made fresh prior to use (see Note 4). Oligomers that
vary in length can be selected depending on the desired sensi-
tivity for an assay: longer oligomers provide increased sensitiv-
ity while shorter oligomers allow tuning of the assay for a
desired clinical range for an analyte. The oligomers are not
currently commercially available, but they can be prepared
using the procedures described in ref. 6.
2. Capillary micropipettes (e.g., Drummond 0.25 μL disposable
micropipettes) for depositing 0.25 μL of oligomer solution
(see Note 5).
3. Synthetic food dyes (Assorted Food Colors & Egg Dye; Wal-
Mart brand) were used to give colorimetric responses and to
track the distribution of fluids within a device. The dyes were
used as 1:5 mixtures of dye to distilled water. The synthetic
food dyes contain the following components: RED 40 (diso-
dium salt of 6-hydroxy-5-[(2-methoxy-5-methyl-4-sulfophenyl)
azo]-2-naphthalenesulfonic acid.), BLUE 1 (disodium salt
218 Gregory G. Lewis and Scott T. Phillips

of ethyl [4-[p-[ethyl(m-sulfobenzyl)amino]-a-(o-sulfophenyl)
benzylidene]-2,5-cyclohexadien-1-ylidene](m-sulfobenzyl)
ammonium hydroxide plus p-sulfobenzyl and o-sulfobenzyl
salts), YELLOW 5 (trisodium salt of 4,5-dihydro-5-oxo-1-(4-
sulfophenyl)-4-[4-sulfophenylazo]-1H-pyrazole-3-carboxylic
acid), and GREEN (a 1:1 mixture of YELLOW 5 and BLUE 1).
4. Assay substrate [5, 9] (e.g., glucose 6-phosphate) in aqueous
solution at a constant concentration (20 mM) (see Note 6).
The assay substrate reacts with the target enzyme to generate
1 unit of glucose for each enzymatic reaction.
5. Buffer solutions (e.g., 40 mM HEPES buffer, pH 8.0) for
depositing assay substrates (see Note 7). The hydrophobic
oligomer responds to hydrogen peroxide more effectively at
pH 8 than in more acidic media. The buffer salts dried in the
device ensure that the sample is at this pH value.
6. Cofactor (e.g., magnesium chloride) for the target enzyme to
increase the activity of the target enzyme and improve the sen-
sitivity of the assay (see Note 8).
7. Immobilized enzymes (i.e., glucose oxidase and catalase) for
removal of glucose and hydrogen peroxide from the sample
and/or generation of hydrogen peroxide from glucose that is
generated in response to the target enzyme (see Note 9).

2.3 Fabricating 1. Flat sheet of glass or non-vinyl plastic (rectangular in shape,

Three-Dimensional not larger than 25 cm × 30 cm; the thickness should be between
Paper-Based 1 mm and 10 mm).
Microfluidic Devices 2. Laser engraver (e.g., Epilog Mini 45 W CO2 laser cutter;
Epilog Laser).
3. Spray Adhesive (e.g., 3M™ Super 77™ Multipurpose Adhesive).
4. Cold laminator (e.g., Drytac® JetMounter™ JM26 laminator).
5. Cold laminate (e.g., Protac™ Ultra UV, 8.0 mil).

3 Methods

3.1 Patterning The hydrophilic paper is patterned with hydrophobic wax to form
Microfluidic Channels defined hydrophilic regions that control sample distribution within
into Paper the paper.
1. Patterns of hydrophobic regions for each layer of paper in the
3D microfluidic device are designed using CleWin Layout
Editor (see Note 10) [10], with different layers of the device
designed on different layers in CleWin so that the features in
each layer are aligned perfectly (see Note 11). The design is
saved as a post-script (.ps) file (see Note 12).
Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 219

2. The post-script file is opened in Adobe® Illustrator® (see Note 13).

The size of the page within Adobe® Illustrator® is set to
20.0 cm × 20.0 cm, which is the size of the paper onto which
the design will be printed.
3. The patterns of hydrophobic regions designed for each layer of
the device are arranged in Adobe® Illustrator® and are placed
in separate Adobe® Illustrator® files. Multiple copies of each
pattern can be arranged in the same Adobe® Illustrator® file.
White space (~9 mm) is maintained on all edges of the Adobe®
Illustrator® page. The Adobe® Illustrator® file is saved as a por-
table document format (.pdf).
4. The .pdf file is opened in Adobe® Acrobat® (see Note 14). The
page is printed using black wax and a Xerox Phaser 8560
printer (see Note 15) [11]. The paper is fed through tray 1,
which is set to 200 mm × 200 mm card stock (see Note 16).
5. The printed paper is placed flat into a 150 °C oven (printed
side up) for 1 min 45 s (see Notes 17 and 18). The paper is
removed from the oven and allowed to cool to room tempera-
ture before further use [11]. The paper is ready for further
manipulation after 10 s (Fig. 3a).

Fig. 3 Photographs showing how paper is prepared for making the devices
shown in Fig. 1. (a) Sheets of patterned paper. Individual layers are numbered
according to Fig. 1b. The individual sheets of patterned paper are outlined by the
white dotted lines. (b) Photograph depicting the method for depositing reagents
into the defined regions onto the paper. The example is green dye being depos-
ited on layer 2 using a pipette. (c) Photograph showing the procedure for deposit-
ing the hydrophobic oligomer (Fig. 1c) using a capillary micropipette
220 Gregory G. Lewis and Scott T. Phillips

Table 1 Reagents used, and their function within each layer of the device shown in Figs. 1 and 4

Layer # Reagents Function

1 None Sample addition and visualization of assay results
2 (a) Dye (a) Allows visualization of assay
(b) Preproccesing reagents (b) Removes chemical impurities (e.g., glucose) from the sample
3 Enzyme cofactors Improves activity of the target enzyme analytes
4 (a) Hydrophobic oligomer (a) Changes wetting properties from hydrophobic
to hydrophilic in response to hydrogen peroxide.
(b) Enzyme substrate (b) Reacts with the target enzyme to produce glucose
5 (a) Enzyme substrate (a) Reacts with the target enzyme to produce glucose
(b) Bead-bound glucose (b) Generates hydrogen peroxide by consuming glucose
oxidase that is produced during the detection event

3.2 Depositing Assay reagents are incorporated into the layers of patterned paper
Reagents for Assays before assembling the microfluidic devices. The choice of reagents
and their purpose are summarized in Table 1. An expanded view of
each layer of the microfluidic device is depicted in Fig. 4.
1. Before the layers of the device are assembled, assay reagents
(e.g., glucose 6-phosphate, which is the substrate for alkaline
phosphatase) are deposited (3 μL) as aqueous solutions into
appropriate hydrophilic regions of paper within the device
(e.g., layers 4 and 5 in the example device shown in Fig. 1b)
(see Note 19). These solutions are allowed to dry under ambi-
ent conditions for 30 min before the device is assembled. This
layer provides selectivity for the target analyte and generates
hydrogen peroxide when the analyte is present.
2. The immobilized enzymes (e.g., glucose oxidase) are depos-
ited (3–8 μL) as aqueous solutions (in 40 mM HEPES, pH
8.0) in the appropriate hydrophilic regions of paper within the
device (e.g., layer 2 and 5 in the example device shown in
Fig. 1b) (see Note 20). These solutions are allowed to dry
under ambient conditions for 30 min before the device is
assembled. The enzymes in layer 2 remove chemical impurities
(e.g., glucose and hydrogen peroxide) from the sample prior
to the sample reaches the assay reagents.
3. The food coloring (1.0 μL) is deposited into the center of a
2.4-mm diameter hydrophilic region of patterned paper
(Fig. 3b) that will be designated as the signaling component
within the 3D microfluidic device (e.g., layer 2 in the exam-
ple device shown in Fig. 1b). The layer is allowed to dry
under ambient conditions for 30 min before it is assembled
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 221

Fig. 4 Expanded view of the device in Fig. 1, showing each layer separately. The
regions of hydrophobic wax are colored gray and the hydrophilic regions of paper
are white. The device shown is 10 mm wide × 20 mm long × 1.8 mm thick

into the device [11]. This layer enables visualization of the

sample in the top layer of the device. The sample becomes
brightly colored so that the user easily can see when to start
and stop timing the assay.
4. Hydrophobic oligomer solutions in organic solvent (THF or
ethyl acetate) (see Note 21) are deposited (using a 0.25 μL
micropipette) into the center of 2.4-mm diameter circular
hydrophilic regions of patterned paper (Fig. 3c) in layer 4 in
the example device shown in Fig. 1b [6, 7]. This layer provides
a change from hydrophobic to hydrophilic in the presence of
hydrogen peroxide, which enables the quantitative assay.

3.3 Assembling Three-dimensional devices allow for sample distribution to multi-

Three-Dimensional ple regions of the device simultaneously, without significantly
Paper-Based increasing the footprint of the device.
Microfluidic Devices 1. The sheet of patterned paper containing the bottom layer of
the device (layer 5 in Fig. 1b) is sprayed lightly with spray
adhesive (see Note 22) (Fig. 5a) [12]. Avoid using excessive
amounts of spray adhesive, as the wicking properties of the
paper can be affected (see Note 22). The next sheet of paper is
222 Gregory G. Lewis and Scott T. Phillips

placed on top of the first layer by aligning the edges of both

sheets so that the patterned features of both layers are in con-
tact (Fig. 5b). This process is repeated with each layer until all
layers of the device are aligned and are in contact.
2. In order to obtain individual devices from the assembled sheet
of devices, they must be cut from the sheet. Patterns that outline
the desired 3D microfluidic device are designed using CleWin
Layout Editor (see Note 10) [10]. The design is saved as a post-
script (.ps) file (see Note 12). This file contains the pattern for
cutting out individual devices from an assembled sheet.
3. The printing specifications must be adjusted for using a laser
printer to cut out individual devices. The post-script file is
opened in Adobe® Illustrator®. The design is highlighted using
the “select all” function and the line widths are set to 0.01
point. The patterns to outline the devices are aligned to match
all of the devices in the sheet. The design is centered onto an
Adobe® Illustrator® page (see Note 14). The file is saved as a
portable document format (.pdf). The file prepared in steps 2
and 3 is used as a pattern by the laser cutter for separating
devices from each other after assembling all devices as a sheet.
4. The assembled sheets of devices are taped onto the upper left
corner of the sheet of glass (or plastic). This device/glass unit
is placed within the Epilog Engraver laser cutter with the upper
left corner of the glass placed against the top/left corner of the
perpendicular rulers (Fig. 5d).
5. The .pdf file is opened in Adobe® Acrobat®, and the file is
printed using the Epilog Engraver laser cutter using the fol-
lowing settings (see Notes 23 and 24): the job type is set to
Vector, the speed is set to 75 %, the power is 65 %, and the
frequency 1,053 Hz. Vector sorting should be selected, the
pull-down menu should be set to “optimize,” and the auto
focus box should be selected. The printing options should be
set to “no page scaling,” and “auto-rotate and center” should
be unselected [10, 11].
6. When the laser cutter has finished cutting the devices, the glass
sheet is removed from the laser cutter and the devices are
removed from the glass sheet using tweezers (Fig. 5e).
7. Two sheets of plastic laminate are cut to completely cover the
sheets of assembled devices (22 cm × 22 cm).
8. Patterns in the laminate cover of the device are cut using a laser
cutter. They must first be designed using CleWin Layout
Editor (see Note 1). Patterns that expose the sample addition
region for the 3D microfluidic device are designed using
CleWin Layout Editor (see Note 10) [5, 11]. The design is
saved as a post-script (.ps) file (see Note 12).
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 223

Fig. 5 Photographs showing the steps involved in assembling the device. (a) Application of spray adhesive to
layer 5. (b) Assembling sheets of patterned paper with applied spray adhesive. Layer 3 is placed on top of layer
4 by aligning the edges of the sheets of paper. (c) The laminate sheet is taped onto a glass sheet prior to cutting
the laminate using the laser cutter. The sheet of laminate is then aligned with the top left corner of the laser
cutter. (d) The assembled sheets of devices are taped onto a glass sheet prior to cutting using the laser cutter.
(e) Photographs of the cut laminate and sheet of devices. The features of the laminate and the devices are
aligned in order to seal the top of the devices. A region of backing (~5 cm) is peeled back to allow for easy
alignment of features without the sheet of laminate sticking to the devices. After alignment, the un-backed
laminate is sealed onto the paper and the remainder of the backing is then removed. (f) The assembled device
is laminated to secure the layers together. Layer 1 of the device is already sealed with laminate, the laminate
sheet for the bottom of the devices has some of the backing removed (~5 cm) and is then aligned with the top
sheet of laminate. (g) Individual devices are then cut out of a sheet of laminated devices
224 Gregory G. Lewis and Scott T. Phillips

9. The printing specifications must be adjusted for using a laser

printer to cut out patterns in the laminate sheet. The post-
script file is opened in Adobe® Illustrator®. The design is high-
lighted using the “select all” function and the line widths are
set to 0.01 point. The patterns to outline the devices are
aligned to match all of the devices in the sheet. The design is
centered onto Adobe® Illustrator® page (see Note 14). The file
is saved as a portable document format (.pdf). The file pre-
pared in steps 8 and 9 is used as a pattern by the laser cutter
for cutting the top laminate sheet.
10. One sheet of plastic laminate is taped onto the upper left
corner of the sheet of glass (or plastic). The plastic laminate
is aligned against the top left corner of the glass sheet. This
laminate/glass unit is placed within the Epilog Engraver
laser cutter with the upper left corner of the glass placed
against the top/left corner of the perpendicular rulers
(Fig. 5c).
11. The .pdf file is opened in Adobe® Acrobat® and the file is
printed using the Epilog Engraver laser cutter using the fol-
lowing settings (see Notes 23 and 24): the job type is set to
Vector, the speed is set to 100 %, the power is 30 %, and the
frequency 1,053 Hz. Vector sorting should be selected, the
pull-down menu should be set to “optimize,” and the auto
focus box should be selected. The printing options should be
set to “no page scaling,” and “auto-rotate and center” should
be unselected [10, 11].
12. When the laser cutter has finished cutting the plastic laminate,
the glass sheet is removed from the laser cutter and the plastic
laminate is removed from the glass sheet using tweezers
(Fig. 5e).
13. The plastic backing on the laminate sheet (that has been pat-
terned) is removed and placed on the top of the assembled
devices, adhesive side down (see Note 25). The features of the
assembled devices are aligned with the laminate sheet so that
the cut sample addition regions match with the sample addi-
tion regions of the devices.
14. The plastic backing on the laminate sheet (that has not been
patterned) is removed and placed on the opposite side of the
assembled devices, adhesive side down (Fig. 5e).
15. The assembled and laminated devices are pressed using a lami-
nator to ensure proper adhesion of the laminate (see Note 26)
(Fig. 5f).
16. Individual devices are then cut out of the sheet using scissors
(or a paper cutter) (Fig. 5g).
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 225

3.4 Conducting Performing the assay requires only that the user add the sample to
Quantitative Assays the paper-based microfluidic device and then record the time for
Using Paper-Based the right hydrophilic circle in Fig. 1a to turn green after the left
Devices hydrophilic circle in Fig. 1a turns green.
1. The sample is added to the central circular region on the top of
the device (i.e., the sample addition region in Fig. 1) (see Note 27).
A minimum sample volume of 30 μL is required for the assay
to function properly, although any sample volume above the
minimum will allow for proper quantification of the target ana-
lyte (see Note 28).
2. Once the left-hand region of the device changes color, time is
measured until the right-hand region of the device changes color.
3. The quantity of target enzyme is determined by comparing the
measured time to a calibration curve (see Note 29).
4. The calculated quantity of target enzyme is adjusted for
temperature using the correction factors described in ref. 5
(see Note 30).

4 Notes

1. Other vector graphics editor programs may be used instead of

CleWin Layout Editor. The program must allow the user to
draw and arrange lines (and simple geometric features) with
micron-level control over line widths.
2. The type of paper used for these experiments was Whatman
Chromatography Paper No. 1, but other types of papers can be
used depending on the desired properties (e.g., faster flow rate,
better sample removal, more uniform reagent distribution).
3. Optimum concentrations of oligomers were determined exper-
imentally [6, 7] and can be varied to adjust the sensitivity of
the assay as needed. Optimum concentrations were determined
by varying the concentration of oligomers and calculating the
resulting limit of detection. The limit of detection values were
then plotted against the corresponding concentrations of
oligomer and fitted with a parabolic curve. The local minimum
of the curve was used to determine the optimum concentra-
tion of oligomer for the assay. Procedures for synthesizing the
oligomers are provided in refs. 6, 8.
4. Solutions of oligomer dissolved in ethyl actetate (or THF)
tend to change concentration after storing at room tempera-
ture for several hours due to slow evaporation of solvent (even
when sealed). Fresh solutions provide the most reproducible
assay results.
226 Gregory G. Lewis and Scott T. Phillips

5. Glass capillary tubes (1 μL) can be substituted for micropipettes,

but we have found that glass capillary tubes give slightly less
accurate results than micropipettes.
6. Other assay substrates can be used instead of glucose
6-phosphate (e.g., lactose). The assay substrate must be cho-
sen so that one enzymatic reaction with the target active
enzyme will generate one (or more) units of glucose. Examples
of appropriate reagents are provided in refs. 5, 9.
7. The hydrophobic oligomer responds most effectively at basic
pH values. A pH of 8.0 was chosen to balance enzymatic activ-
ity for the target enzyme(s) and the rate of response of the
hydrophobic detection reagent.
8. Magnesium chloride is a cofactor/promoter of both alkaline
phosphatase and β-d-galactosidase (the two active enzyme tar-
gets tested). Other cofactors could be selected and added to
the device to promote enzymatic activity. Cofactors used for
target enzymes may act as inhibitors for glucose oxidase, or
inversely cofactors for glucose oxidase may inhibit the target
enzyme, so care must be taken in selecting which cofactors to
include in the device.
9. Biotin functionalized glucose oxidase (or catalase) was pur-
chased (or synthesized) for use in the assays [13, 14]. The pro-
cedure for immobilization can be found in ref. 5. The glucose
oxidase removes background glucose from the sample and
generates hydrogen peroxide. The hydrogen peroxide is con-
verted to oxygen and water by the immobilized catalase.
10. The hydrophobic regions in patterned paper are the solid lines
in the graphics program, and the hydrophilic regions are the
regions without shading.
11. The features and layout of the devices can be altered in order
to raise or lower the limit of detection of the assay for target
active enzymes [5, 6]. Examples of such alterations include
changing the length of the lateral flow channel in layer 5 in
Fig. 1b, or increasing the number of repetitions of layer 4
within the device.
12. When saving as a .ps file, only one layer of the program can be
saved at a time, so all of the layers for the device need to be
moved onto the same layer within the program, and arranged
laterally so that there is no overlap between different layers.
13. At this point, each of the layers should contain patterns for
regions that will become hydrophobic; the lines in the pro-
gram mark the locations where the wax will print.
14. Make sure to check that the size of the page remains
20 cm × 20 cm square when opened in Adobe® Acrobat®.
Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 227

15. When printing from the .pdf file, make sure that the page
scaling menu is set to “none” and “auto-rotate and center” is
not checked.
16. From the main menu on the printer: move the arrow to
“Paper Tray Setup” and press OK. This action brings up the
“Paper Tray Setup” menu. Move the arrow to “Tray 1 Paper”
and press OK. This action will bring up the “Tray 1 Paper” menu
that states the current setup for tray 1. Move the arrow to
“Change Setup…” and press OK. This action brings up the
“Tray 1 Paper Size” menu. Move the arrow to “New Custom
Size” and press OK. This action brings up the “Short Edge”
menu. Move the arrow to “Change…” and press OK. Move
the arrow up or down until you reach 200 mm (the units are
set to mm and cannot be changed). Press OK. This action
will bring up the “Long Edge” menu. Move the arrow to
“Change…” and press OK. Move the arrow up or down
until 200 mm is reached, and press OK; the “Tray 1 Paper
Type” menu will appear. Move the arrow to “Card Stock”
and press OK. This action will return you to the “Paper Tray
Setup” menu. Move the arrow to “Exit” and press OK. This
action returns to the main menu.
17. A piece of aluminum foil is used to cover the shelf in the oven
while heating to provide a clean surface. The printed chroma-
tography paper is placed, wax side up, on the aluminum foil.
18. The paper needs to stay in the oven for between 1 min 30 s and
1 min 45 s. If the paper is heated longer than 2 min, the wax
will bleed laterally, and the sample will not wick through the
layer. If the paper is heated for less than 1 min 30 s, then the
wax will not fully penetrate the paper and separated hydro-
philic regions of paper may become connected.
19. The concentration of the assay reagents was 20 mM. The con-
centration can be adjusted as needed to improve the sensitivity
of the assay. The immobilized enzymes are stabilized on paper
and can be stored overnight at room temperature before
assembly.
20. The immobilized glucose oxidase and immobilized catalase are
spotted individually and dried [15, 16].
21. Hydrophobic oligomers where n ≥ 5 (Fig. 1c) are not soluble
in ethyl acetate (the solvent used for oligomers n < 5); instead
they are dissolved in THF [5, 6]. The symbol n refers to the
number of repeating units in the oligomer. The concentration
of the hydrophobic oligomer solution directly affects the sen-
sitivity of the assay. The oligomers can be prepared as outlined
in refs. 6, 8.
228 Gregory G. Lewis and Scott T. Phillips

22. Spray adhesive is used to assemble the devices in this procedure;

however, other methods could be used (such as double-sided
tape) if desired [11]. The minimum amount of spray adhesive
was used to hold the devices together, as the adhesive would
alter the wicking properties of the paper [12].
23. Changing the properties of the laser cutter: Within Acrobat,
select “File” and “Print.” Select Epilog Engraver from the drop
down menu, and click on “Properties” next to the printer selec-
tion. Change the selection to the desired values and click “OK.”
24. Choosing to print on the computer: the “Go” button on the
printer must be pressed. If after pressing this button, the laser
does not print, check to make sure that all lines in the graphics
file are set to 0.01 point. If any of the lines are larger than this
value, the laser cutter will not be able to use vector cutting. If
the lines are smaller than this value, they may be too thin for
effective cutting of the tape.
25. Remove the upper 5 cm of the plastic backing and fold the back-
ing onto itself. The cut features of the laminate can then be
aligned with the sheet of devices before the region with the
removed backing is pressed onto the sheet. Smooth the laminate
from the center outwards and then slowly remove the backing as
the laminate sheet is pressed against the sheet of devices.
26. A Drytac® JetMounter™ JM26 laminator was used with
Protac™ Ultra UV (8.0 mil) laminate. Other laminators and
laminates could be used as long as both are designed for cold
lamination. The laminate serves to seal the devices (minimiz-
ing evaporation) and provide a more rigid support. Lamination
instructions were followed as indicated in the laminator opera-
tion manual. The use of a laminator serves to compress the
devices while they are laminated, this ensures that all of the
device features are in contact with each other.
27. These devices are compatible with buffer solutions, along with
complex fluids (e.g., serum) that may contain the target active
enzyme [6].
28. The amount of sample added to the device is dictated by the
volume of liquid that is absorbed by the entire device [17].
29. A calibration curve at 19 °C and 20 % relative humidity is avail-
able in ref. 6
30. For temperatures below 15 °C, 1 min is added to the measured
assay time. For temperatures below 33 °C, 4 min is subtracted
from the measured assay time. From 15 to 33 °C, the mea-
sured time is decreased by 0.3079 × ΔT (where ΔT is the dif-
ference between the temperature at which the assay was
conducted and the temperature at which the calibration curve
was generated) [6].
 Quantitative Point-of-Care (POC) Assays Using Measurements of Time… 229

References
1. Yager P, Domingo GJ, Gerdes J (2008) Point- 9. Mohapatra H, Phillips ST (2013) Reagents and
of-care diagnostics for global health. Annu Rev assay strategies for quantifying active enzyme
Biomed Eng 10:107–144 analytes using a personal glucose meter. Chem
2. Yager P, Edwards T, Fu E, Helton K, Nelson K, Commun 49:6134–6136
Tam MR, Weigl BH (2006) Microfluidic diag- 10. Carrilho E, Martinez AW, Whitesides GM
nostic technologies for global public health. (2009) Understanding wax printing: a simple
Nature 442:412–418 micropatterning process for paper-based micro-
3. Urdea M, Penny LA, Olmsted SS, Giovanni fluidics. Anal Chem 81:7091–7095
MY, Kaspar P, Shepherd A, Wilson P, Dahl CA, 11. Noh H, Phillips ST (2010) Fluidic “timers” for
Buchsbaum S, Moeller G, Hay Burgess DC paper-based microfluidic devices. Anal Chem
(2006) Requirements for high impact diagnos- 82:8071–8078
tics in the developing world. Nature 444:73–79 12. Lewis GG, DiTucci MJ, Baker MS, Phillips ST
4. Peeling RW, Holmes KK, Mabey D, Ronald A (2012) High throughput method for prototyp-
(2006) Rapid tests for sexually transmitted ing three-dimensional, paper-based microflu-
infections (STIs): the way forward. Sex Transm idic devices. Lab Chip 12:2630–2633
Infect 82:v1–v6 13. Liu H, Xiang Y, Lu Y, Crooks RM (2012)
5. Lewis GG, Robbins JS, Phillips ST (2013) Aptamer-based origami paper analytical device
Point-of-care assay platform for quantifying for electrochemical detection of adenosine.
active enzymes to femtomolar levels using mea- Angew Chem Int Ed 51:6925–6928
surements of time as the readout. Anal Chem 14. Xiang Y, Lu Y (2011) Using personal glucose
85:10432–10439 meters and functional DNA sensors to quan-
6. Lewis GG, Robbins JS, Phillips ST (2013) Phase- tify a variety of analytical targets. Nat Chem 3:
switching depolymerizable poly(carbamate) 697–703
oligomers for signal amplification in quantita- 15. Mateo C, Palomo JM, Fernandez-Lorente G,
tive time-based assays. Macromolecules 46: Guisan JM, Fernandez-Lafuente R (2007)
5177–5183 Improvement of enzyme activity, stability and
7. Lewis GG, DiTucci MJ, Phillips ST (2012) selectivity via immobilization techniques.
Quantifying analytes in paper-based microfluidic Enzyme Microb Technol 40:1451–1463
devices without using external electronic read- 16. Khan MS, Li X, Shen W, Garnier G (2010)
ers. Angew Chem Int Ed 51:12707–12710 Thermal stability of bioactive enzymatic papers.
8. Robbins JS, Schmid KM, Phillips ST (2013) Coll Surf B 75:239–246
Effect of aromaticity on the rate of azaquinone 17. Phillips ST, Lewis GG (2013) Advances in
methide-mediated release of benzylic phenols. materials that enable quantitative point-of-care
J Org Chem 78:3159–3169 assays. MRS Bull 38:315–319
 Chapter 16

Smartphone-Based Fluorescence Detector for mHealth

Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly

Abstract
We describe here a compact smartphone-based fluorescence detector for mHealth. A key element to
achieving high sensitivity using low sensitivity phone cameras is a capillary array, which increases sensitivity
by 100×. The capillary array was combined with a white LED illumination system to enable wide spectra
fluorescent excitation in the range of 450–740 nm. The detector utilizes an orthographic projection sys-
tem to form parallel light projection images from the capillaries at a close distance via an object-space tel-
ecentric lens configuration that reduces the total lens-to-object distance while maintaining uniformity in
measurement between capillaries. To further increase the limit of detection (LOD), a computational image
processing approach was employed to decrease the level of noise. This enables an additional 5–10× decrease
in LOD. This smartphone-based detector was used to measure serial dilutions of fluorescein with a LOD
of 1 nM with image stacking and 10 nM without image stacking, similar to the LOD obtained with a com-
mercial plate reader. Moreover, the capillary array required a sample volume of less than 10 μl, which is an
order of magnitude less than the 100 μl required for the plate reader.
As fluorescence detection is widely used in sensitive biomedical assays, the approach described here
has the potential to increase mHealth clinical utility, especially for telemedicine and for resource-poor set-
tings in global health applications.

Key words mHealth, Smartphone, Capillary, Fluorescence detection, Orthographic projection,

Mobile camera, Camera

1 Introduction

mHealth (mobile computing, medical sensors, and communica-

tions technologies for health care) [1] has the potential to address
needs for medical diagnostics with clinical utility in telemedicine.
In recent years, many potential mHealth technologies have been
developed, including a reader for lateral flow immuno-
chromatographic assays [2], fluorescence detector [3–5], wide-
field fluorescent microscopy [6], capillary array for immunodetection
for Escherichia coli [7], lens-free microscopy [8], fluorescent imag-
ing cytometry [9], microchip ELISA-based detection of ovarian
cancer HE4 biomarker in urine [10], detection systems for

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_16, © Springer Science+Business Media New York 2015

231
232 Joshua Balsam et al.

melanoma or skin lesions [11–13], loop-mediated isothermal

amplification (LAMP) genetic testing device [14], acoustic wave
enhanced immunoassay [15], colorimetric reader [16], phone-
assisted microarray reader for mutation detection [17], and mobile
phone camera for DNA detection [18]. Another application for
mHealth is medical diagnostics in low and middle-income coun-
tries where resources are more limited. Most of the population in
these countries lacks access to suitable medical diagnostics. The
challenge and need for these countries is to develop simple, low-
cost diagnostics for use in resource-poor settings with minimal
medical infrastructure [19–22].
Many mHealth technologies are based on optical detection
utilizing the CMOS camera native to a mobile device, such as a
smartphone. However, the limiting factor for many of these tech-
nologies is the low sensitivity of the CMOS camera, which is too
low to be useful for many optical modalities, such as low intensity
fluorescent signal detection.
To improve the sensitivity of a smartphone camera for fluores-
cent detection, we recently developed a computational approach
using “image stacking” [23] where images are acquired in video
mode using a webcam (which utilize CMOS sensors similar to typ-
ical camera phones). This enabled many individual frames to be
captured and then combined with an image stacking algorithm to
average the values of each pixel and significantly reduce random
noise. This approach improved the LOD by 5–10×, and enabled
the detection of very weak signals that would otherwise be masked
by noise.
In addition to computational image enhancement through
image stacking, we have also used optical amplification of signals to
increase sensitivity. We developed a capillary array which enabled a
~100× increase in detection sensitivity [3, 5, 24]. These arrays are
three-dimensional (3D) detection systems, where X columns and
Y rows of capillaries are arrayed in two dimensions and light propa-
gates via the capillary walls to provide a third Z dimension for illu-
mination along the axes of the capillaries oriented normal to the
array distribution. Combining computational image enhancement
with capillary optical amplification resulted in an ~1,000× increase
in detection sensitivity [24].
For mHealth applications, it is essential that the detector be
simple and compact, which requires an optical configuration capa-
ble of imaging optical signals over short distances through the long
parallel capillaries tubes (with the sample) distributed over a large
space. To reduce the imaging distance for the 3D capillary array,
orthographic projection was applied [3]. In previous work, we
demonstrated the potential of capillary arrays to increase sensitivity
for fluorescence detection [3, 5, 24]. Here, we provide the meth-
odology for this technology.
 Smartphone-Based Fluorescence Detector for mHealth 233

2 Materials

2.1 Smartphone- 1. Samsung Galaxy SII cell phone (Samsung Electronics Co.)
Based Detector with a built-in lens with a focal ratio of f/2.6 and a 4 mm focal
length or similar phone.
(a) Alternative imager is a generic webcam.
2. Any simple lens with a diameter larger than the capillary array
to be imaged.
(a) Lenses with shorter focal ratios result in a more compact
device. In this instance, plano-convex lenses with diame-
ters of 45 or 20 mm were used (150 mm and 10 mm focal
lengths, respectively).
3. Green emission filter HQ535/50 M (Chroma Technology
Corp Rockingham, VT).
4. Blue excitation filter D486/20× (Chroma Technology Corp,
Rockingham, VT).
5. Illumination: LED illumination box containing red, green,
blue, and white LEDs was custom built by Luminous Film
(Shreveport, Louisiana, www.luminousfilm.com/led.htm).
(a) As an alternative an LED flashlight can be used (see Note 1).
(b) As an another alternative to LED illumination, low cost
lasers equipped with line generator can be used (from vari-
ous ebay vendors).
6. Fluorescein (Sigma-Aldrich, St. Louis, MO).
7. For flashlight configuration two 1 cm × 1 cm mirrors.
8. Capillaries, heparinized borosilicate glass capillaries
(Drummond Scientific, Broomall, PA).
(a) The length of the heparinized capillaries used is 32 mm
with outer diameter of 0.8 mm, the inner diameter of
0.65 mm and with glass thickness of around 50 μm.

2.2 Computer 1. Image capturing: Webcam images can be captured using

Control and Data CamApp v. 1.0.0.9 (Aveo Corp., 2008) or any other software
Analysis which allows for adjustment of the webcam’s gain, exposure
time and other settings. The stock camera application on the
smartphone can be used.
2. Image analysis software: ImageJ, free NIH software (http://
rsb.info.nih.gov/ij/download.htm).
3. Data analysis software: Excel (Microsoft, Redmond, WA) and
Sigma plot (Sigma plot, Ashburn, VA).
234 Joshua Balsam et al.

2.3 Detector 1. Black and clear 3.2 mm poly-methylmethacrylate (PMMA/

and Assay Plate acrylic) (Piedmont Plastics, Beltsville, MD) (see Note 2).
Fabrication 2. Clear 0.5 mm polycarbonate film (Piedmont Plastics, Beltsville,
MD) (see Note 2).
3. 3 M 9770 Adhesive transfer Tape (Piedmont Plastics, Beltsville, MD).
4. Epilog Legend CO2 65 W laser cutter (Epilog, Golden, CO).

3 Methods

The main two components of the system are the fluorescence

detector and assay plate array or assay capillary array. Other meth-
ods used are image capturing and image analysis.

3.1 Fabrication The detector and the assay plate were fabricated by laser machining
Methods which enable rapid fabrication and is ideal for prototyping.
1. Micromachining: The detector and the fluidics were made of
3.2 or 1 mm black acrylic, micromachined with a computer
controlled 65 W Epilog Legend CO2 laser system (see Note 3).
2. Bonding: The polycarbonate bottom was bonded to the acrylic
with double sided pressure-sensitive adhesive transfer tape (see
Note 4) which was bonded to the micromachined layer (so
there is no contact between the adhesive to the fluids).
Assembled devices were bonded together and were processed
with a hot roll laminating machine to eliminate air bubbles and
for uniform bonding.

3.2 Fluorescence The basic optical configuration of the compact smartphone-based

Detector fluorescence detector [3, 5, 24] is shown schematically in Fig. 1a,
with (1) camera phone, (2) emission filter, (3) plano-convex lens
(20 mm diameter, 10 mm focal length) focusing the image onto
the camera’s CMOS sensor, (4) alignment fixture, (5) interchange-
able capillary tube array to image the samples, (6) excitation filter,
and (7) illumination box. In the final device, the detector is
enclosed in a black acrylic box to prevent interference of ambient
light (see Notes 5–7). The smartphone can be replaced by webcam
or other video capturing device.

3.3 Illumination A multiwavelength spatial LED illuminator [25] shown in Fig. 1-

Modules VII. The multiwavelength spatial LED illuminator module com-
prises a custom built multiwavelength LED illumination box with
two different types of LEDs: (1) white, which generates broad emis-
sion spectra in the red, green, and blue spectral ranges [25], and (2)
discrete RGB LEDs which generate at the red, green, and blue
wavelengths. The LED illuminator contains four RGB LED strips
and four white LED strips, both at 18 LEDs per foot. Each LED
emission color is controlled individually with a switch in the front
 Smartphone-Based Fluorescence Detector for mHealth 235

a b
i
ii
df
iii
iv

vi

vii

c i d Smartphone
ii
iii
viii
iv
v
Flashlight
vi
vii Filter

Fig. 1 Smartphone based fluorescence detector for mHealth. (a) A schematic of the fluorescent sensor and (b) a
photograph of the device. The main components are: (i) camera phone, (ii ) emission filter, (iii ) secondary lens, (iv )
alignment fixture, (v ) capillary tube array, (vi ) two spaced excitation filters, and (vii ) multiwavelength LED light
box. In (a), df is the distance between the capillaries and the camera lens. (c) Flashlight configuration with (i )
camera phone, (ii ) emission filter, (iii ) secondary lens, (iv ) alignment fixture, (v ) capillary tube array, (vi ) excitation
filters and a diffuser (vii ) light path with two mirrors (viii ) LED flashlight, and (d) a photograph of the flashlight
based device

panel. The top of the box consists of a diffusion panel (milky white
plastic panel), which assures uniformity of the light and uniform
illumination of the sample chip [25]. The dimensions of the diffu-
sion panel are 8.5 × 5.5 cm. The LED box is powered by 12 V DC.
Alternatively LED flashlight (Fig. 1c VIII) with a diffusion
panel can be used as a low-cost illuminator (see Note 3) with two
mirrors to focus the light to the capillary array (Fig. 1 VII).

3.4 Optical System To image the three-dimensional capillary array, orthographic pro-
jection optics using a telecentric lens configuration to represent the
three-dimensional array in two dimensions was used. The main
challenge of three-dimensional imaging is uneven distribution of
the entire light source (Fig. 2a). Fluorescence measurement of a
capillary array is shown in Fig. 2b, in which the center capillaries
236 Joshua Balsam et al.

Fig. 2 The light distribution of orthographic projection optics. (a) An ImageJ 3D analysis of the planar illumina-
tion source is shown as seen with the single lens system. (b) An image of an array with 16 capillary tubes using
the single lens optics with the corresponding 3D visualization (c). (d) An ImageJ 3D analysis of imaging with
telecentric lens with the corresponding array of 16 capillary tubes (e) and the 3D visualization (f), demonstrat-
ing a significantly more uniform illumination field
 Smartphone-Based Fluorescence Detector for mHealth 237

appear brighter than the peripheral capillaries resulting in a higher

signal being measured at the center of the image (Fig. 2c). To cor-
rect this, an object-space telecentric lens configuration was used
(Fig. 1-III). The term “object-space telecentric” indicates that an
object at any distance from the lens will appear the same size in the
resultant image. The basic optical configuration of the object-space
telecentric optics is shown schematically in Fig. 1a, with the plano-
convex lens (45 mm diameter, 150 mm focal length) focusing the
image onto the camera’s CMOS sensor. At the ideal camera-to-
lens distance, the focal points of the camera lens (Fig. 1a-I) and the
secondary lens (Fig. 1a-III) are aligned. This two-lens system
forms an object-space telecentric lens. The light distribution in this
optical configuration is more homogenous (Fig. 2d showing the
entire light source) compared with light distribution without the
lens (Fig. 2a). In addition, the optical system includes excitation
and emission filters. To allow only fluorescence emission and to
block the LED excitation light which will increase the noise and
reduce the detection sensitivity, a green emission filter D535/40 m
was used. In order to transmits the only wavelengths of the LED
that efficiently excite the fluorescence dye, a blue excitation filter
was used (HQ480/20×).

3.5 Waveguide While it is possible to use plates for assay detection, the waveguide
Capillary Array capillary array [3, 5, 24] shown schematically in Fig. 1a-V (actual
photo shown in Fig. 1b) was used for fluorescence amplification. In
this array, the excitation light-wave energy propagating through the
capillary walls can interact directly with and excite the fluorophore
molecules (via evanescent waves), which in turn emit light that can
be detected at the end of the capillary by an imaging detector. To
orient all the capillary channels towards the camera image sensor
simultaneously, two four-by-four arrays of holes in 3.2 mm thick
plates of black acrylic were fabricated which hold the capillaries in a
parallel configuration (see Note 8). The length of the capillaries
used is 32 mm with inner diameter of 0.8 mm. When loading a
capillary in the array with a sample, it is important to avoid air bub-
bles that can reduce the fluorescence signal (see Note 9). For the
fluorescence detection used in this work, light is emitted by a LED
passed through the excitation filters (Fig. 1a-V), carried through
the capillaries (Fig. 1a-V), collected by a telecentric lens (Fig. 1aIII)
which forms a parallel projection of the capillary array through the
emission filter (Fig. 1-II), and is then finally captured by the camera
(Fig. 1a-I). The resulting orthographic projection optical system
allows for a more compact device (with a phone-to-capillary dis-
tance of 32 mm as shown in Fig. 1b) which enhances portability for
mHealth. In previous designs that utilized conventional lens sys-
tems (such as the stock lens that came with the camera), a larger
distance (~150 mm) between the imaging device and the capillaries
was required in order to obtain images with uniform brightness
across the capillary array (see Note 10).
238 Joshua Balsam et al.

3.6 Image Capturing Images can be captured by mobile phone camera or by webcam.
Regardless of the device, the signals from the images can be ana-
lyzed by the following procedure:
1. Mobile phone imaging: Images were captured with the expo-
sure time for the cell phone camera set to its maximum of
1/15 of a second and the camera gain set to its maximum of
800 ISO. The images obtained are in 8-bit color JPEG format,
maximum JPEG quality was used (level 100, compression ratio
2.6:1). At this level of compression, only extremely minor arti-
facts exist. Each color image is essentially three different mono-
chrome images, each one representing a color channel (red,
green, and blue) and each pixel of each image having a value
between 0 and 255. Because the signal of interest for fluores-
cein is in the green spectrum, the green channel alone is ana-
lyzed and the red and blue channels are discarded.
To reduce background noise in the final image and improve
the signal-to-noise ratio (SNR), an average dark frame was sub-
tracted from each image to be analyzed. A dark frame is an
image with identical exposure time and gain settings as the
original image, but taken with no light reaching the camera
sensor (see Note 11 for how best to block light from reaching
the sensor). When several of these dark frames are averaged
together, the resulting image represents the average back-
ground noise generated by the camera sensor. This averaging
was done in ImageJ using many (>30) dark frames with the
mean value for each pixel calculated and combined to form the
final dark frame. The green channel was extracted from this
mean dark frame image and subtracted from the green channel
of the corresponding original image containing the signal of
interest. This process has been described in greater depth in
previous work [23, 26].
2. Webcam imaging: The incoming video stream from the
Webcam is enhanced using a software-based video processing
amplifier built into the CamApp software (available in most
webcam interface software). Exposure time and gain settings
are set to their maximum unless this setting causes over-expo-
sure (in the case of very high fluorescent signal, in which case
exposure time should be reduced to remove over-exposure).
The video processing amplifier can also apply adjustments to
the video stream to change brightness and contrast. These
should be chosen such that no pixels are recorded as having a
luminance value of zero or maximum (for an 8-bit image this
corresponds to values of 0 or 255). Because all pixels are gen-
erating a signal at the hardware level (due either to photons or
to on-chip sources of noise), if any pixels in the final recorded
image have zero signal it means that signal is being lost.
The goal of the brightness and contrast adjustment is to force
the range of the recorded pixel values to be as large as possible to
 Smartphone-Based Fluorescence Detector for mHealth 239

take advantage of dynamic range afforded by the 8-bit sensor (i.e.,

to broaden the histogram). In the final recorded image, the lumi-
nance of the darkest pixels should be close to zero and that of the
brightest should be as high as possible without being saturated.
Other image processing operations, such as gamma correc-
tion and sharpness, can introduce nonlinear or highly localized
adjustments to the luminance function. Both gamma and sharp-
ness were set to zero to minimize unnecessary or biased manipu-
lation of the data stream.

3.7 Computational The images can be analyzed using any standard image processing
Image Enhancement software. For our system, we chose the freeware ImageJ developed
to Improve Detection at NIH (see Note 12). It enables image stacking, 2D linear image
Sensitivity analysis of the intensity of each spot in a rows or columns (Fig. 3a),
as well as 3D (Fig. 2b) spatial analysis (2D of the spatial position of
the spots and the third dimension the intensity of the spots), to
provide an enhanced visual representation of the image brightness,
which is shown in Fig. 3b. For spot analysis, the mean intensity
value of every assay spot was exported to an Excel spreadsheet and
to the scientific data analysis and graphing software Sigmaplot,
which was used to plot the data. Several analyses were conducted,
including subtracting baseline noise level and calculating the
Signal-to-Noise ratio (S/N). A computational image enhancement
of images captured from webcam video [23] was used to increase
the sensitivity of mHealth detection. The schematic of image stack-
ing for enhanced imaging is shown in Fig. 3-I. In video mode, the
webcam captures n individual frames each with underlying signal
of interest (marked with white circles) and noise. In image stack-
ing, the pixels of the n individual frames are averaged together
(stacked image), which reduces the standard deviation of this back-
ground noise, and its mean value can be subtracted from each
pixel, resulting in an enhanced signal to noise ratio.

3.8 Factors Although mobile device cameras (e.g., smartphones and web-
Contributing to System cams equipped with CMOS) each inherently have less sensitivity
Sensitivity compared with CCDs, photomultipliers, or avalanche photodi-
odes, the sensitivity of the mobile device camera was increased to
the level of a photomultiplier-based detector through the use of
combination of factors:
1. The use video mode combine with image stacking described
above increases sensitivity.
2. The quality of filters is very critical; using high quality narrow
band filters at the Emission/Excitation wavelengths will reduce
noise and improve detection.
3. While camera phone lenses are not interchangeable, with many
webcams it is possible to change the lens (we used an f/1.2
lens [23]) to maximize the amount of light transmitted to the
sensor (see Note 13).
240 Joshua Balsam et al.

II 1 2 3 4 5 6 IV 1 2 3 4 5 6
I
1 A VI
B
2 C 100

3 D
E
4 F
10

SNR
5 III V
1
n-1
n 0.1
0.001 1 1000
Stacked Concentration (nM)
image

Fig. 3 Computational image enhancement of images captured in video mode (using a webcam detector). A sche-
matic of image stacking for enhanced imaging is shown in (I). In video mode, the n individual frames each with
underlying signal of interest (marked with white circles), and interfering noise. In image stacking, each pixel is
averaged throughout the n frames, which reduces the standard deviation of this background noise and its mean
value can be subtracted from each pixel, resulting in an enhanced signal-to-noise ratio (SNR). To demonstrate
this, a 36 well plate was loaded with six concentrations of fluorescein (column 1–6), each in six replicas (rows
A–F). The signals of the wells were detected by the CMOS webcam operating in a still single frame mode (II) with
the corresponding ImageJ 3D analysis in (III). The plate analyzed in video mode and the image enhanced by image
stacking (IV) with the corresponding ImageJ 3D image in (V). The SNRs for both modes were plotted (VI). Triangles
are data points for video captured stacked images and circles are single image mode with no stacking. The LODs,
the mean plus three times standard deviation of a control (water), is marked as a dashed line. The fluorescein
concentrations used were (column #1–6): 10,000, 1,000, 100, 10, 1 nM, and control (water)

4. Increasing the intensity of the LED illumination (the use of

more LEDs) will increase fluorescent signal.
5. The use of low cost lasers equipped with line generator may
increase light intensity and provide narrow wavelength illumi-
nation, making emission filters unnecessary.
6. For single frame imaging, some webcams allow for long
exposure times (>1 s).

4 Notes

1. If using a flashlight for illumination, longer exposure may be

needed.
2. Controlling optical noise: Fluorescence emission and scattered
excitation light can propagate through the chip, causing cross-
talk between adjacent channels. This can become a major source
of optical noise in the system [27–29] by increasing background
noise, thus reducing the sensitivity of the measurements. To
 Smartphone-Based Fluorescence Detector for mHealth 241

limit the effect of fluorescence background, polycarbonate and

acrylic (not Mylar, which is the commonly used material for
lamination based fabrication) were used as the main fabrication
materials due to their lower fluorescence background [29] and
ease of cutting via laser. Using absorbing material (black plastic
in the case of visible wavelengths) decreases crosstalk and elimi-
nates the problem of autofluorescence.
3. For fabrication, the laser power and speed for cutting poly-
mers has to be determined empirically. It is recommended to
use the minimum laser power to reduce overheating or exces-
sive burning of the material.
4. For strong bonding, remove 1 cm of the adhesive tape cover,
align the tape with acrylic surface, and attach the exposed tape
to the acrylic surface. With a ruler, press the tape to the surface
and slowly move the ruler across the tape with little pressure in
order to prevent air bubbles. When bonding the tape, it should
be aligned with the acrylic. For assembling of the assay plate,
remove the protective cover from the other side of the double
side adhesive (taped to the acrylic) and remove the protective
cover from the polycarbonate. Then, align the two pieces and
apply pressure so that all wells are fully isolated from one
another. The use of a hot-roll laminator makes this step simple
and repeatable.
5. Make sure the lens, filters, and sample plate are vertically cen-
tered and aligned.
6. Make sure the arrows on the coated filters are facing the camera,
and that if the filters being used have angular dependence (e.g.,
interference filters) the excitation source beam is confined to
the correct range of angles. The filter holder should be leveled
and at a height which allows proper focusing of the lens.
7. The imager enclosure must be sealed to block external light
sources. A long exposure can be used to detect light leaks.
8. For the capillary array it is critical that all the capillaries be
aligned normal to the collection optics. This can be achieved
using two plates of black acrylic with holes for the capillaries
which hold the capillaries in a parallel configuration.
9. Avoiding air bubbles: Another issue with accurately measuring
signals is occasionally air bubbles will be trapped in the capil-
laries. This greatly reduces the fluorescent signal measured
from a capillary and must be avoided. For arrays of capillaries
it is possible to use a micro-pipette for sample loading. With
the capillary array inclined at an angle (e.g., 45°), align the
pipette tip with the bottom end of the capillary so that the two
are in contact, and inject the sample slowly while watching for
air bubbles. The length of the capillary should not be so great
that, when placed vertically, the fluid in the capillary forms a
242 Joshua Balsam et al.

droplet at the bottom end, as this will increase the chance of

leakage during measurement.
10. Measurement uniformity must be checked by loading each
sample channel with an identical concentration of fluorescent
sample. Nonuniformities should be corrected by modifying
the optical configuration when possible. Otherwise, post-
processing corrections can be used to correct for repeatable
nonuniformity. It must be determined whether the source of
the nonuniformity is additive or multiplicative in nature, so
that the calculated correction factor can be properly applied
(i.e., by either subtraction or division, respectively).
11. Imager Qualification: Cell-phone camera and webcam perfor-
mance varies depending both on the device and the applica-
tion used to collect the images from the camera. Prior to
preparing a sample for measurement it is important to under-
stand the performance of the camera and application being
used. First, block any light from reaching the camera sensor
by taping a layer of aluminum foil over the lens aperture. It is
not recommended to use plastic for this purpose because
many plastics are partially transparent to infrared light which
many camera sensors are sensitive to. After blocking the aper-
ture, navigate to the camera control application and increase
the exposure compensation (measured in EV, exposure value)
to its maximum. Alternatively, if exposure time can be set
directly, increase it to its maximum value. In general this will
be something around 1/15th of a second for camera phones,
though for some webcams it may be much longer. Also
increase camera gain setting (usually labeled ISO in phones)
to its maximum. Leave the resolution at its maximum setting,
and if there is an option for white balance, leave it set to auto.
While monitoring the live view from the camera, point the
blocked aperture towards a bright light source to verify that
no light is reaching the sensor. Take several images with the
aperture blocked, export them to a computer if using a cell
phone, and open them in ImageJ. Open the “Brightness and
Contrast” setting window (from the ImageJ menu bar:
Image > Adjust > Brightness/Contrast), and slide the adjust-
ment labeled “Maximum” to the left until you can clearly see
the fluctuations in the image signal (Fig. 4). The image should
resemble Fig. 4a, with a substantial number of pixels (30 % in
this case) having recorded values greater than zero, and these
pixels having a somewhat even distribution over the image
frame. If the image resembles Fig. 4b, with very few pixels
recording a value greater than 0 (only 1.5 % in this example),
and with those pixels concentrated only in the corners of the
image, the current settings in the camera capture application
or the camera itself are not suitable for sensitive measure-
 Smartphone-Based Fluorescence Detector for mHealth 243

ments. The first step in resolving this problem is to experi-

ment with different settings in the capture application,
followed by using third-party applications for camera control
(as they may treat the raw data in such a way that it is less
compressed in the final image). If an image similar to Fig. 4a
cannot be obtained, this imager device is likely not suitable
for sensitive fluorescent measurements.
12. Image Analysis: Images of fluorescent assays should be opened
in ImageJ (or other software capable of pixel value quantifica-
tion). If a color image sensor has been used (typical of cell-
phone cameras and webcams), the RGB color channels should
be split apart. With the image open in ImageJ, from the menu
click Image > Color > Split Channels. The original image will be
split into its RGB components. Knowing the emission wave-
length of the fluorophore(s) being used, discard those compo-
nents which do not contain the wavelengths of interest (i.e., for
fluorescein, a green dye, discard the red and blue components
as they contain mostly noise). The final image, now mono-
chrome, must be measured to extract the assay data. Using the
brightness and contrast control window, increase the screen
contrast until the individual samples can be distinguished from
one another. Using an appropriate selection size and shape,
define regions of interest around each sample spot (including
spots for any control samples). The ROI Management tool in

Fig. 4 Characteristics of cell-phone images. Two images captured with the same cell-phone camera with the
aperture blocked. Both images are identically stretched in ImageJ to reveal the faint detail they contain. (a) Using
maximum gain and maximum EV compensation, (b) using auto-gain and auto-EV setting. In (a) 70 % of pixels
have a value of 0 (no data recorded). Images such as this may be suitable for enhancement via image stacking.
In (b), 98 % of pixels have a value of zero. Images such as these are not likely to benefit from image stacking
244 Joshua Balsam et al.

ImageJ can be used for this purpose, as well as for performing

simultaneous measurements of the selected regions. A suitable
statistical value to use for each ROI measurement must be cho-
sen (e.g., mean, median, maximum). Data can then be exported
into analysis software (e.g., Excel, Minitab).
13. A fast focal ratio (e.g., f/1.2) will enable shorter exposure
time, but focusing will be more difficult and the depth of field
will be reduced.

References

1. Istepanian R, Jovanov E, Zhang YT (2004) 11. Rosado L, Castro R, Ferreira L, Ferreira M

Introduction to the special section on (2012) Extraction of ABCD rule features from
M-Health: beyond seamless mobility and skin lesions images with smartphone. Stud
global wireless health-care connectivity. IEEE Health Technol Inform 177:242–247
Trans Inf Technol Biomed 8(4):405–414 12. Wadhawan T, Situ N, Rui H, Lancaster K,
2. Mudanyali O, Dimitrov S, Sikora U, Yuan X, Zouridakis G (2011) Implementation
Padmanabhan S, Navruz I, Ozcan A (2012) of the 7-point checklist for melanoma detec-
Integrated rapid-diagnostic-test reader plat- tion on smart handheld devices. Conf Proc
form on a cellphone. Lab Chip IEEE Eng Med Biol Soc Conf
12(15):2678–2686 2011:3180–3183
3. Balsam J, Bruck HA, Rasooly A (2013) 13. Boyce Z, Gilmore S, Xu C, Soyer HP (2011)
Orthographic projection capillary array fluo- The remote assessment of melanocytic skin
rescent sensor for mHealth. Methods 63:276 lesions: a viable alternative to face-to-face con-
4. Balsam J, Rasooly R, Bruck HA, Rasooly A sultation. Dermatology 223(3):244–250
(2013) Thousand-fold fluorescent signal 14. Stedtfeld RD, Tourlousse DM, Seyrig G,
amplification for mHealth diagnostics. Biosens Stedtfeld TM, Kronlein M, Price S et al (2012)
Bioelectron 51C:1–7 Gene-Z: a device for point of care genetic test-
5. Balsam J, Bruck HA, Rasooly A (2013) ing using a smartphone. Lab Chip
Capillary array waveguide amplified fluores- 12(8):1454–1462
cence detector for mHealth. Sens Actuators B 15. Bourquin Y, Reboud J, Wilson R, Zhang Y,
186:711–717 Cooper JM (2011) Integrated immunoassay
6. Zhu H, Yaglidere O, Su TW, Tseng D, Ozcan using tuneable surface acoustic waves and lens-
A (2011) Wide-field fluorescent microscopy on free detection. Lab Chip 11(16):2725–2730
a cell-phone. Conf Proc IEEE Eng Med Biol 16. Lee DS, Jeon BG, Ihm C, Park JK, Jung MY
Soc Conf 2011:6801–6804 (2011) A simple and smart telemedicine device
7. Zhu H, Sikora U, Ozcan A (2012) Quantum for developing regions: a pocket-sized colori-
dot enabled detection of Escherichia coli using metric reader. Lab Chip 11(1):120–126
a cell-phone. Analyst 137(11):2541–2544 17. Zhang G, Li C, Lu Y, Hu H, Xiang G, Liang Z
8. Tseng D, Mudanyali O, Oztoprak C, Isikman et al (2011) Validation of a mobile phone-
SO, Sencan I, Yaglidere O et al (2010) Lensfree assisted microarray decoding platform for
microscopy on a cellphone. Lab Chip signal-enhanced mutation detection. Biosens
10(14):1787–1792 Bioelectron 26(12):4708–4714
9. Zhu H, Mavandadi S, Coskun AF, Yaglidere O, 18. Lee D, Chou WP, Yeh SH, Chen PJ, Chen PH
Ozcan A (2011) Optofluidic fluorescent imag- (2011) DNA detection using commercial
ing cytometry on a cell phone. Anal Chem mobile phones. Biosens Bioelectron
83(17):6641–6647 26(11):4349–4354
10. Wang S, Zhao X, Khimji I, Akbas R, Qiu W, 19. Hay Burgess DC, Wasserman J, Dahl CA
Edwards D et al (2011) Integration of cell (2006) Global health diagnostics. Nature
phone imaging with microchip ELISA to 444(Suppl 1):1–2
detect ovarian cancer HE4 biomarker in urine 20. Urdea M, Penny LA, Olmsted SS, Giovanni
at the point-of-care. Lab Chip MY, Kaspar P, Shepherd A et al (2006)
11(20):3411–3418 Requirements for high impact diagnostics in
 Smartphone-Based Fluorescence Detector for mHealth 245

the developing world. Nature 444(Suppl detection of Botulinum Neurotoxin A Activity.

1):73–79 Sens Actuators B 146(1–8):297–306
21. Yager P, Domingo GJ, Gerdes J (2008) Point- 26. Balsam J, Ossandon M, Bruck HA, Rasooly A
of-care diagnostics for global health. Annu Rev (2012) Modeling and design of microma-
Biomed Eng 10:107–144 chined optical Soller collimators for lensless
22. Balsam J, Ossandon M, Bruck HA, Lubensky CCD-based fluorometry. Analyst
I, Rasooly A (2013) Low-cost technologies for 137(21):5011–5017
medical diagnostics in low-resource settings. 27. Irawan R, Tjin SC, Yager P, Zhang D (2005)
Exp Opin Med Diagn 7(3):243–255 Cross-talk problem on a fluorescence multi-
23. Balsam J, Bruck HA, Kostov Y, Rasooly A channel microfluidic chip system. Biomed
(2012) Image stacking approach to increase Microdevices 7(3):205–211
sensitivity of fluorescence detection using a low 28. Hawkins KR, Yager P (2003) Nonlinear
cost complementary metal-oxide- decrease of background fluorescence in poly-
semiconductor (CMOS) webcam. Sens mer thin-films - a survey of materials and how
Actuators B 171–172:141–147 they can complicate fluorescence detection in
24. Balsam J, Bruck HA, Rasooly A (2014) microTAS. Lab Chip 3(4):248–252
Thousand-fold fluorescent signal amplification 29. Sapsford KE, Sun S, Francis J, Sharma S,
for mHealth diagnostics. Biosens Bioelectron Kostov Y, Rasooly A (2008) A fluorescence
51:1 detection platform using spatial electrolumi-
25. Sun S, Francis J, Sapsford KE, Kostov Y, nescent excitation for measuring botulinum
Rasooly A (2010) Multi-wavelength spatial neurotoxin A activity. Biosens Bioelectron
LED illumination based detector for in vitro 24:618
 Chapter 17

Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary

Valves for mHealth Medical Diagnostics
Joshua Balsam, Hugh Alan Bruck, and Avraham Rasooly

Abstract
There is a new potential to address needs for medical diagnostics in Point-of-Care (PoC) applications using
mHealth (Mobile computing, medical sensors, and communications technologies for health care), a
mHealth based lab test will require a LOC to perform clinical analysis. In this work, we describe the design
of a simple Lab-on-a-chip (LOC) platform for mHealth medical diagnostics. The LOC utilizes a passive
capillary valve with no moving parts for fluid control using channels with very low aspect ratios cross sec-
tions (i.e., channel width ≫ height) achieved through transitions in the channel geometry via that arrest
capillary flow. Using a CO2 laser in raster engraving mode, we have designed and fabricated an eight-
channel LOC for fluorescence signal detection fabricated by engraving and combining just two polymer
layers. Each of the LOC channels is capable of mixing two reagents (e.g., enzyme and substrate) for vari-
ous assays. For mHealth detection, we used a mobile CCD detector equipped with LED multispectral
illumination in the red, green, blue, and white range. This technology enables the development of low-cost
LOC platforms for mHealth whose fabrication is compatible with standard industrial plastic fabrication
processes to enable mass production of mHealth diagnostic devices, which may broaden the use of LOCs
in PoC applications, especially in global health settings.

Key words LOC, Microfabrication, Microfluidics, Valve, Laminated object manufacturing, Laser
engraving, CCD, LED, Fluorescence

1 Introduction

The emergence of mHealth (mobile computing, medical sensors,

and communications technologies for health care) technology has
been facilitated by rapid advancements in smartphone and Lab-on-
a-chip (LOC) technologies. One of the simplest techniques for
fabricating LOC platforms is Laminated Object Manufacturing
(LOM). It utilizes layers of adhesive-coated laminates successively
bonded together for prototyping and as well as commercial pro-
duction. In LOM, polymer sheets are precisely cut through their
thickness [1–5] which enables a rapid single step (generally only a
few minutes) fabrication of an individual layer followed by the lay-
ers assembly and bonding by an adhesive [3] or heat [4, 5] to

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_17, © Springer Science+Business Media New York 2015

247
248 Joshua Balsam et al.

produce a 3D object. To realize LOC platforms, polymer films

such as acrylic, polycarbonate or polyester can be cut with a CO2
laser, followed by assembly and bonding.
However, current microfluidics designs using LOM are often
complex flow networks requiring many layers of polymers and
adhesives to fabricate complex fluidic structures. Reducing the
number of layers can improve the optical performance and reliabil-
ity of the device, while reducing the cost of fabrication, because it
reduces the number of interfaces that can potentially weaken and
disrupt the flow in the LOC. The laser is normally used in vector
mode for cutting through the entire depth of the layer and assem-
bly of 3D devices is achieved by stacking multiple layers. However,
using micromachining approaches (i.e., engraving to variable
depth within the plane of the layer) to realize 3D features, it is pos-
sible to fabricate microfluidic elements on both sides of the layer,
which increases the efficiency of fabrication and simplifies design.
An associated issue with reducing the number of layers is flow
control. In complex LOCs, this is not a major issue because high
aspect ratios increase flow resistance and naturally limit flow.
Simplified designs use less complex, shorter channels that lower
flow resistance and cause uncontrolled capillary flow that can only
be overcome only if a simple valve can be developed.
In this work, we describe a new eight-channel LOC that uti-
lizes the new passive capillary valve design to control flow in chan-
nels with low aspect ratios based on sharp transitions in the channel
geometry that act as a barrier to arrest capillary flow. Capillary
force valves [6] act to prevent liquid from entering or leaving a
capillary. They are triggered passively (i.e., passive capillary valves)
using large pressure differentials to overcome the surface tension
of the liquid. The LOC platform is fabricated with a simple LOM
fabrication approach using just two layers of polymer with laser
engraved 3D features, which is also compatible with standard
industrial plastic fabrication processes (e.g., metal mold master or
injection molding) for mass production of mHealth diagnostic
devices. For mHealth detection, we used a mobile CCD detector
equipped with LED multispectral illumination in the red, green,
blue and white range. Therefore, this new design employing pas-
sive capillary valves can broaden the use of LOC platforms in PoC
applications, especially in global health settings.

2 Material

2.1 Reagents 1. Black and clear 3.2 mm poly-methylmethacrylate (PMMA)

and Materials for LOC (Piedmont Plastics, Beltsville, MD).
Fabrication 2. Clear 0.25 or 0.5 mm polycarbonate film (PC) (Piedmont
Plastics, Beltsville, MD).
 Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves for mHealth Medical… 249

3. 3 M 9770 adhesive transfer tape (Piedmont Plastics, Beltsville, MD).

4. Fluorescein (Sigma-Aldrich, St. Louis, MO).

2.2 Equipment 1. Fisher (FS-14) Sonicator (Fisher Scientific, Pittsburgh, PA).

2. Beckman mini centrifuge (Beckman, Fullerton, CA).
3. Epilog Legend CO2 65 W computer controlled laser cutter
(Epilog, Golden, CO).
4. Corel Draw 11 (Corel Corp. Ontario, Canada).
5. SXVF-M7 Cooled CCD (Adirondack Video Astronomy,
Hudson Falls, NY).
6. An alternative low cost camera: Sony PlayStation® Eye web-
cam was used as an inexpensive CMOS photodetector
(Amazon, Seattle, WA).
7. An alternative camera with improved sensitivity is a C mount
1.3 Mp gray scale CCD camera CMLN 13S2M-CS (Point
Grey Research, Richmond BC Canada).
8. Tamron manual zoom CCTV 4–12 mm, f/1.2 lens (Spytown,
Utopia, NY).
9. Alternatively, a smartphone, webcam, or digital camera can be used.
10. Green emission filter HQ535/50 M (Chroma Technology
Corp Rockingham, VT).
11. Blue excitation filter D486/20× (Chroma Technology Corp,
Rockingham, VT).
12. Illumination: LED illumination box containing red, green,
blue, and white LEDs was custom built by Luminous Film
(Shreveport, Louisiana, www.luminousfilm.com/led.htm).
13. As an alternative to LED illumination, low cost lasers equipped
with line generator can be used (from various ebay vendors).

2.3 Data Analysis 1. Image analysis software: ImageJ, free NIH software, (http://
rsb.info.nih.gov/ij/download.htm) (see Note 1).
2. Data analysis software: Excel (Microsoft, Redmond, WA) and
Sigma plot (Sigma plot, Ashburn, VA).

3 Methods

The LOCs were prepared using a rigid 3.2 mm thick thermoplastic

PMMA and more compliant 0.25 mm thick thermoplastic polycar-
bonate film (Piedmont Plastics, Beltsville, MD). Fluorescein and
water are used to demonstrate mixing. The eight-channel LOC
used in this study was designed in Corel Draw 11 (Corel Corp.
Ontario, Canada) and micromachined in 1/8 in. clear acrylic using
a computer controlled Epilog Legend CO2 65 W laser cutter
250 Joshua Balsam et al.

(Epilog, Golden, CO). Before cutting, layers of PC or PMMA were

coated with 3 M 9770 adhesive transfer double-sided tape
(Piedmont Plastics, Beltsville, MD). For fluorescence detection we
used a CCD detector which was described previously [7]. An eight-
channel LOC was designed and fabricated using only two layers
with engraved 3D features capable of mixing two reagents for fluo-
rescence detection in various enzymatic and chemical assays.

3.1 General LOM Corel Draw computer aided drawing (CAD) software was used to
Approach design the patterns and features of the device. Before cutting, layers
of PC and PMMA were coated with 3 M 9770 adhesive transfer
double-sided tape. The cutting and engraving was done with a CO2
laser cutter as described in previous work [7–9] designed to work
with thin sheets of polymer film, such as acrylic, polycarbonate, and
polyester. The layers were assembled and bonded by successive lam-
ination with adhesive [3] to produce a 3D device [1–5].
Laser cutting: Depending on the material being cut, different
power levels and speed settings must be used to ensure a clean cut (see
Note 2). The appropriate settings are arrived at by trial and error. It
is good practice to use the lowest necessary power for cutting to
minimize burning of the material and localized heating of the work-
piece. The power settings for the vector mode used for fabrication
depend on the cutting objective. In most cases, vectors cuts are
designed to be through cuts which require higher power and slower
beam speeds. It is essential to clean debris from the cuts after laser
cutting as well to avoid blocking the channels by polymer and adhe-
sive debris. After cutting, the edge of the PC, PMMA and double-
sided tape needs to be cleaned using a fine brush and any debris in the
fluid channels should be removed using a needle or by sonication.
Laser Engraving: To reduce the number of layers, we have
sought to utilize a new variant on the laser cutting approach where
both surfaces of the PMMA can be engraved to include the func-
tional elements of the LOC. The technical challenge in this approach
is appropriate control of the Z-axis of the features, which is how
deep the features are engraved. This was achieved by empirically
determining the relationship between the power of the laser and the
depth of the engraving for a given polymer. In addition, the features
must satisfy the clearance requirements of the engraving process
(i.e., undercuts are not possible as with processes like etching).
When engraving both sides of the PMMA, both sides must
align precisely so fluids can flow from the top side wells to the bot-
tom side fluidics (see Note 3). This precise alignment was
achieved by:
1. Engraving the top side.
2. Flipping the layer and placing it in a custom frame built to hold
the machined layer in a precise position. The frame is designed
so that the PMMA layer will snap-fit into it for repeatable
Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves for mHealth Medical… 251

alignment. The frame is then aligned to built-in guides on the

laser cutter stage.
Bonding: The polycarbonate layers were bonded to the acrylic
with double-sided pressure-sensitive adhesive transfer tape. The
transfer tape can be bonded to the acrylic prior to machining so
that during machining the adhesive can be removed in the area of
contact with the fluid. Thus, the sample has minimal contact with
the adhesive during LOC operation. Device layers are aligned and
then bonded by pressure.
The type of adhesive used for LOC systems is very critical.
Therefore, several adhesive tapes can have been explored for bond-
ing, including 3 M 9770, 3 M 9690 and 3 M 501FL. 3 M 9770 is
an adhesive transfer tape with 2.0 mil (51 μm) hi-strength acrylic
adhesive 330MP on a 4.2 mil (106 μm) polycoated Kraft liner.
According to the manufacturer, the adhesive will hold securely
after exposure to numerous chemicals including oil, mild acids and
bases. The tape is stable for short periods (up to an hour) at tem-
peratures up to 120 °C. The adhesive transfer tape is normally
applied on one or both sides of a particular layer. To apply the tape,
the liner is first removed from one side followed by bonding to the
polymer surface (e.g., polycarbonate). Once the polymer/adhesive
assembly is created, it can be cut in the laser cutter. Final assembly
then involves removing the remaining liner, alignment of the layers
and application of pressure to make the bond. The main challenge
with this adhesive tape is that the adhesive may flow when pressure
is applied during bonding. For designs involving smaller channels,
the flowing adhesive can block the small channels, which limits the
utility of this adhesive tape.
The double coated tape 3 M 9690 uses 330MP adhesive 2.8
mil (70 μm) on a clear 2 mil (50 μm) polyester carrier. The adhe-
sive in this system flows less than 3 M 9770, resulting in less block-
ing of channels, although the potential still exists. The polyester
carrier can be used as one of the layers for the microfluidics
replacing a polycarbonate layer, thereby simplifying production.
However, the polyester carrier used in 3 M 9690 exhibits auto flo-
rescence, so it is less useful when florescence detection is desired.
The 3 M 501FL is called ultraclean laminating adhesive trans-
fer tape, and it comes without polyester backing. The 25 μm adhe-
sive layer may reduce flow into the channels; it uses a 50 μm PET,
silicone liner which seems to produce cleaner cuts, especially when
engraving is used.
To achieve strong bonding using these tapes, the following
procedure is employed (see Note 4):
1. Remove ~1 cm of the liner from one side of the double-sided
adhesive tape roll.
2. Align and attach this portion of the tape with the polymer sur-
face to be bonded (e.g., polycarbonate film or acrylic layer).
252 Joshua Balsam et al.

a b
1
2
3
4
5
6

7
8

Fig. 1 Functional elements of multichannel LOC system. (a) A schematic of an eight-channel microfluidic LOC plat-
form where the reagents reservoirs (1) and (2) are joined via a V-junction to the joining channel (3). The mixed
reagents are monitored in the detection wells (4) followed by a mixing trap (5). The eight-channel negative pressure
distribution manifold (6) is connected to a waste chamber (7) to the outlet, which is connected to a pump or syringe
(8). (b) A clear LOC device connected to a syringe loaded with food coloring to enhance the LOC features

3. With a ruler (or a roller), press the tape against the polymer
surface and slowly unwind the tape and press it against the
acrylic surface with the ruler to prevent air bubbles.
4. Cut the tape.
5. To create a polymer surface with adhesive on both sides, repeat
for the other side of the polymer.
Assembly: For LOM, the liner is removed from the adhesive on
each layer. Each layer is then aligned using guide holes machined
in all layers. A simple plate with two pins is used to align the layers
(Fig. 1) which are first joined together with gentle pressure.

3.2 Design Figure 1a shows the functional elements of the LOC system which
and Fabrication are similar to our previous design of a six layer microfluidic device
of Two-Layer LOC developed for botulinum neurotoxin A (BoNT-A) activity analysis
[9]. Each of the channels is designed to mix two different reagents
(e.g., enzyme and substrate) in order to generate a signal for detec-
tion. As shown in Fig. 1a, each channel is based on a V-junction
design [9] that joins the two reagents reservoirs (1, 2) passing into
a 0.5 mm joining channel (3). The mixing and incubation of
reagents is performed and monitored in detection wells (4) (see
Note 5). At the end of the channel, a trap chamber (5) prevents
diffusion among the connected channels during incubation. All
the eight channels are connected to a negative pressure distribu-
tion splitter (6), which leads to a waste chamber buffer (7).
 Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves for mHealth Medical… 253

Fig. 2 Outline of layers and the passive valve design for the two-layer LOC platform. Outline of the layers for
the two-layer LOC platform. Layers are numbered in order from the top layer down. Both layers are 3.2 mm
PMMA. Layer I has top side (I-T) and bottom side engraving (I-B). The second layer has top side engraving only
(II). Image III is the combined device shown from the top and image IV is the side view of the LOC showing the
fluid flow between the layers. The passive capillary valve with a sharp transition in cross-sectional geometry
is used to arrest capillary flow

The outlet of the device (8) is connected to a pump or syringe as

the prime fluid mover. When a syringe is used, no electric power is
needed. In Fig. 1b, a clear PMMA LOC device was photographed
with the syringe moving fluids. The clear PMMA enables optical
monitoring of the assay.
As previously discussed, the unique aspect of our LOC design is
the engraving of features into both sides of the PMMA to enable
better utilization of the surfaces. The two-layer configuration is
shown in Fig. 2, the first layer (Fig. 2-I) with features on both the
top (Fig. 2-I-T) and bottom (Fig. 2-I-B). The top side of the
engraved PMMA (Fig. 2-I-T) layer providing rigidity to the LOC
and sufficient volume for the reagent reservoirs. This layer includes
two reagent wells (Fig. 1a-1 and 2) engraved in the plastic. In the
bottom of each well a small hole (e.g., 0.25 mm) is machined
through the plastic to enable flow between the two sides of the layer.
Also in this layer, an outlet hole machined through the layer provid-
ing an outlet connection for a pump or syringe (Fig. 1-A-8).
The bottom side of the first layer (Fig. 2-I-B) includes addi-
tional engraved features of the top of a V-junction connected to
the detection well with a channel. This layer also includes the nega-
tive pressure splitter (Fig. 1-A-6) connecting all the channels to the
waste reservoir (Fig. 1-A-7) and outlet for a pump or syringe
(Fig. 1-A-8).
The second layer fabricated with double-sided adhesive tape on
the top and includes the V-junction the detection wells and a mixing
trap (Fig. 1-A-5). As shown in Fig. 2 IV, in this design, there is no
direct flow from the V-junction to the detection wells on layer II,
instead the flow is through layer I-B which enables better flow
control and minimizes flow during the filling of the reagent wells.
254 Joshua Balsam et al.

In addition to the two-layer design, we also tested a similar three-

layer design. The top first layer is identical; however, the second layer
fabricated with thin 0.25 mm thick polycarbonate (PC) film is not
engraved but cut through and sealed by a third with no features.

3.3 Flow Control One critical issue with microfludics is control of capillary flow (e.g.,
Using Passive during filling wells), which is mainly a function of channels diam-
Capillary Valves eter, length, and fluid viscosity (i.e., Hagen–Poiseuille’s Equation).
Fluid control in a LOC can be achieved by using a pump and valves
with moving parts, which complicates fabrication. In multilayer
LOCs, a 6-layer design that we have previously developed was used
[9] to create channels with high aspect ratios which can naturally
limit uncontrolled “spontaneous” flow [9].
To reduce the number of layers, we incorporated a new passive
capillary valve design, shown in Fig. 2-III with all layers superim-
posed. A side view is shown in Fig. 2-IV with the PMMA layer
engraving shown from the sides, the top (I-T), and the bottom
(I-B) and the top engraved layer II. To prevent uncontrolled flow
during filling of the reagent, there is no direct flow between sample
reservoirs and detection wells, with flow instead via channels in
layer II. In addition, the passive capillary valve shown in the circled
area in Fig. 2-IV marked with an arrow contains a sharp transition
in cross-sectional geometry to restrict capillary flow. Fluid must
transition from layer II up to a channel in I-B. This step change in
channel geometry creates a point where the fluid contact angle is
undefined and additional work (in the form of increased pressure)
is required to overcome this. The result is a passive capillary valve
which minimizes flow during filling of the reagent wells.
Figure 2-IV demonstrates the challenges of engraving both sides
of the PMMA, both sides have to be precisely aligned so fluids can
flow from the top side wells to the bottom side fluidics. This precise
alignment was achieved by first engraving the top side and then flip-
ping the layer and engraving the bottom side using a custom frame
built to hold the machined layer in a precise position. The frame was
designed so that the PMMA layer would snap-fit into it for reliable
and repeatable alignment. The frame is then aligned to built-in
guides on the laser cutter stage. Another technical challenge is the
control of the Z-axis, which is how deep the features are engraved.
As previously discussed, this was achieved by empirically varying the
power of the laser to determine the relationship to the depth of
engraving for a given polymer. By combining engraving and lamina-
tion, we have been able to reduce the number of layers required to
achieve microfluidic flow in a LOC platform from 6 to 2, substan-
tially simplifying the fabrication of a LOC while maintaining preci-
sion. The reduced number of layers also reduces the number of
interfaces, which significantly enhances the reliability of these devices
as these interfaces can potentially weaken and disrupt the flow in the
LOC eventually rendering it inoperable.
 Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves for mHealth Medical… 255

1 2 3 4 5 6 7 8
a b c
1
2
3

4
5
6

Fig. 3 Fluorescein mixing and detection using the LOC platform. (a) The main elements of CCD based fluorescent
detector: CCD camera (1) equipped with a manual zoom 4–12 mm, f/1.2 C-mount lens (2) and a green pass band
emission filter (3). The eight-channel LOC (4) is placed above the blue band pass excitation filter (5) which is
mounted on the top of the multiwavelength LED illuminator (6). (b) The image of the two-layer LOC after mixing,
the florescent dyes in the detection wells are marked with circle and (c) the 3D ImageJ analysis of the wells

3.4 Fluorescence The use of the LOC for mHealth requires an appropriate detection
Detector technique, where optical detection is most commonly used. The
basic optical configuration of such mHealth fluorescence detector
[10–12] is shown schematically in Fig. 3a. The main elements of
this CCD based fluorescent detector are: CCD camera (1) equipped
with a manual zoom 4–12 mm to capture the image, f/1.2
C-mount lens (2) and a green pass band emission filter (3) which
block the excitation light (see Note 5). The eight-channel LOC
described above (4) is placed above the blue band pass excitation
filter (5) which is mounted on the top of the multiwavelength
LED illuminator (6) used for the excitation of the fluorophores
(see Note 6). An example of the image of the two-layer LOC with
a florescent dye in the detection well (marked with circle) is shown
in Fig. 3b and in Fig. 3c the 3D ImageJ analysis of the wells.

3.5 LOC Fluorescein The LOCs were tested for mixing and detection of fluorescein, a
Mixing and Detection medically relevant and widely used fluorescent tracer. To measure
the mixing uniformity of the eight-channel device, one well in each
channel was loaded with ~10 μl fluorescein and the other with
10 μl water (see Note 7). After applying low pressure at the outlet
port using a syringe, the fluids are drawn into the mixing wells. We
were able to control the flow with the simple passive capillary valve
(Fig. 2-IV) which prevented uncontrolled flow and filling of the
channels.
Fluorescein in the mixing wells was detected by measuring the
signal from the wells excited at 480 nm and detected at 523 nm by
256 Joshua Balsam et al.

the CCD detector (see Note 8). The schematics of the CCD detec-
tor [13] are shown in Fig. 3a, with the main elements consisting of
an SXVF-M7 CCD camera (1) equipped with a Tamron manual
zoom CCTV 4–12 mm, f1.2 C-mount lens (2) that has a green
pass band emission filter (3) mounted on the end of the lens. The
eight-channel LOC (4) is placed above the blue band pass excita-
tion filter (5) which is mounted on the top of a multiwavelength
LED illuminator (6). The image of the two-layer LOC after mix-
ing is shown in Fig. 3b and the 3D ImageJ analysis of the wells is
shown in Fig. 3c.
For the two-layer design (Fig. 3b) the average signal is 196
ADU and the standard deviation of the measurements of 25
ADU, which is ~13 %. For the three-layer device (not shown),
the average signal is 86 ADU which is lower than the signal of
the two-layer design, most likely due to replacing the thin poly-
carbonate film with the approximately order of magnitude
thicker PMMA (~1/3 of it is the engraved channels and wells).
The standard deviation of the measurements was 16 ADU,
which is ~19 % of the average. The issue of uniformity can be
attributed to the use of the CO2 laser in engraving raster mode.
When laser pulsing for engraving, heating of the plastic surface
is less uniform than during raster cutting. This nonuniformity
introduces surface roughness that can affect the uniformity of
the measurement. For cheap, mass production of the prototype,
replication processes using a metal mold master or injection
molding could improve surface quality and therefore increase
the uniformity of the measurements. Therefore, these results
suggest that the trade-off for simplicity, using two-layer fabrica-
tion, is a lower signal and lower uniformity among channels.
However, the lower signal may not be a major issue because it
can be compensated for by increasing exposure time, provided
the background noise does not increase with time.
A critical issue in microfluidic LOC platforms is cross talk (see
Note 9), which occurs when there is mixing or diffusion between
channels, especially in assays requiring long incubation times. Such
cross talk reduces measurement reliability, decreases sensitivity, and
limits the usefulness of the LOC. To overcome this problem we
use traps (Fig. 1a-5) at the end of channels to separate measuring
wells (Fig. 1a-4) from the connected channels (Fig. 1a-6). To mea-
sure the design’s effectiveness, sample and reagent reservoirs of
alternating channels were loaded with either water or Fluorescein.
The solutions were drawn to fill the detection wells and the traps,
but not the connected channels. Measurements of the fluorescent
signals suggested that there was no significant cross talk between
channels loaded with fluorescein to the channels loaded with water,
even after 2 h of incubation, suggesting that this design can elimi-
nate cross talk or diffusion between channels.
 Two-Layer Lab-on-a-Chip (LOC) with Passive Capillary Valves for mHealth Medical… 257

4 Notes

1. Enhanced image visualization: For data analysis, ImageJ can be

used for high contrast visualization, which will not affect the
measured values but will enable easier visual identification of
their locations.
2. Engraving: The laser power and speed for cutting and engrav-
ing polymers has to be determined empirically. It is recom-
mended to use the minimum laser power to reduce overheating
or burning the material.
3. Alignment: In assembling two engraved surfaces, it is very crit-
ical that both sides be precisely aligned so fluids can flow prop-
erly from the top side wells to the bottom side fluidics. The
custom frame built to hold the machined layer in a precise
position must hold the layer in a fixed position with little toler-
ance for errors so the frame must be identical to the part
machined, the best is to use the remaining PMMA used for the
layer as a frame. The frame must include a guide for the place-
ment of it on the laser cutter stage so the laser beam will start
machining at precisely the same point.
4. Bonding: For strong bonding, remove 1 cm of the adhesive
tape cover, align the tape with acrylic surface, and attached the
exposed tape to the acrylic surface. With a ruler, press the tape
to the surface and slowly move the ruler across the tape with
little pressure in order to prevent air bubbles. When bonding
the tape, it should be aligned with the acrylic. For assembling
of the assay plate, remove the protective cover from the other
side of the double side adhesive (taped to the acrylic) and
remove the protective cover from the polycarbonate. Then,
align the two pieces, and apply pressure to fully bond the PC
and PMMA.
5. Imaging conditions: For taking images, a fully open aperture
(e.g., f/1.2) will enable shorter exposure time, but focusing
will be more limited and reduce the sharpness of the image.
6. Quality of optical filtering: For the CCD detector to measure the
effectiveness of the filters, it is recommended to perform two
long exposures (e.g., 3 min) without the assay plate, one with
the EL on and one with the EL off. Ideally, the two measure-
ments should be very similar (the blue filters pass only blue light
which is blocked by the green filters). The difference between
measurements may suggest that the blue filters do not block all
green light and/or the green filters do not block all blue light.
7. Sample Size: The “holes” for the wells on the assay plate are
3 mm in diameter, which allows for analysis of 20 μl samples
(only ~13 μl used). Smaller holes make the loading of the
258 Joshua Balsam et al.

sample less reproducible because the fluid meniscus within the

wells causes light diffraction, which complicate quantification.
8. Uniformity of optical signal: For the CCD detector to measure
light uniformity, it is important to measure the light with a
long enough exposure time (e.g., 60 s) when all of the wells
are loaded with the same fluorescein sample. If the lighting is
not uniform, correction values for each well can be calculated
and used for measurement compensation.
9. Controlling noise: Fluorescence emission and scattered excita-
tion light can propagate through the chip, causing cross talk
between adjacent channels. This can become a major source of
optical noise in the system [1, 7, 14], which reduces the sensi-
tivity of the measurements. To limit the effect of fluorescence
background, PC, and not Mylar, which is a commonly used
material for most lamination based fabrication, was used as the
main fabrication material due to its lower fluorescence back-
ground [7]. Using black material decreases the noise. Adding
air gaps between channels did not reduce the background noise.

References

1. Irawan R, Tjin SC, Yager P, Zhang D (2005) 8. Sapsford KE, Francis J, Sun S, Kostov Y,
Cross-talk problem on a fluorescence multi- Rasooly A (2009) Miniaturized 96-well ELISA
channel microfluidic chip system. Biomed chips for staphylococcal enterotoxin B detec-
Microdevices 7(3):205–211 tion using portable colorimetric detector. Anal
2. Schilling EA, Kamholz AE, Yager P (2002) Bioanal Chem 394(2):499–505
Cell lysis and protein extraction in a microflu- 9. Sun S, Ossandon M, Kostov Y, Rasooly A
idic device with detection by a fluorogenic (2009) Lab-on-a-chip for botulinum neuro-
enzyme assay. Anal Chem 74(8):1798–1804 toxin a (BoNT-A) activity analysis. Lab Chip
3. Munson MS, Hasenbank MS, Fu E, Yager P 9(22):3275–3281
(2004) Suppression of non-specific adsorption 10. Balsam J, Bruck HA, Rasooly A (2013)
using sheath flow. Lab Chip 4(5):438–445 Capillary array waveguide amplified fluores-
4. Rossier JS, Schwarz A, Reymond F, Ferrigno cence detector for mHealth. Sens Actuators B
R, Bianchi F, Girault HH (1999) Microchannel 186:711–717
networks for electrophoretic separations. 11. Balsam J, Bruck HA, Rasooly A (2013)
Electrophoresis 20(4–5):727–731 Orthographic projection capillary array fluo-
5. Rossier J, Reymond F, Michel PE (2002) rescent sensor for mHealth. Methods 63:276
Polymer microfluidic chips for electrochemical 12. Balsam J, Rasooly R, Bruck HA, Rasooly A
and biochemical analyses. Electrophoresis (2013) Thousand-fold fluorescent signal
23(6):858–867 amplification for mHealth diagnostics. Biosens
6. Pompano RR, Platt CE, Karymov MA, Ismagilov Bioelectron 51C:1–7
RF (2012) Control of initiation, rate, and rout- 13. Sun S, Francis J, Sapsford KE, Kostov Y,
ing of spontaneous capillary-driven flow of liquid Rasooly A (2010) Multi-wavelength Spatial
droplets through microfluidic channels on slip- LED illumination based detector for in vitro
chip. Langmuir 28(3):1931–1941 detection of Botulinum Neurotoxin A Activity.
7. Sapsford KE, Sun S, Francis J, Sharma S, Sens Actuators B 146(1–8):297–306
Kostov Y, Rasooly A (2008) A fluorescence 14. Hawkins KR, Yager P (2003) Nonlinear
detection platform using spatial electrolumi- decrease of background fluorescence in poly-
nescent excitation for measuring botulinum mer thin-films - a survey of materials and how
neurotoxin A activity. Biosens Bioelectron they can complicate fluorescence detection in
24(4):618–625 microTAS. Lab Chip 3(4):248–252
 Chapter 18

Spectrometry with Consumer-Quality CMOS Cameras

Alexander Scheeline

Abstract
Many modern spectrometric instruments use diode arrays, charge-coupled arrays, or CMOS cameras for
detection and measurement. As portable or point-of-use instruments are desirable, one would expect that
instruments using the cameras in cellular telephones and tablet computers would be the basis of numerous
instruments. However, no mass market for such devices has yet developed. The difficulties in using mega-
pixel CMOS cameras for scientific measurements are discussed, and promising avenues for instrument
development reviewed. Inexpensive alternatives to use of the built-in camera are also mentioned, as the
long-term question is whether it is better to overcome the constraints of CMOS cameras or to bypass them.

Key words Spectrophotometry, Array detector, Luminescence spectrometry, Commercial off-the-

shelf instrumentation, Portable instruments

1 Introduction

Absorption, fluorescence, chemiluminescence, and reflectance

spectrometries have been employed for biomedical characteriza-
tion for many decades. While vibrational spectroscopy is commonly
used for qualitative identification of chemical species, ultraviolet,
visible, and near-infrared spectroscopies are most commonly used
for quantification. Silicon photodetectors are natively responsive
between 400 and 1,100 nm; the human eye responds in the range
400–700 nm [1]. Thus, cameras optimized to produce photo-
graphs that appeal to the human eye are likely useful for quantita-
tive spectrometry over a 300 nm range. The quirks of human
perception complicate such use. Nevertheless, with the wide pro-
liferation of digital cameras, cellular telephones equipped with
cameras, and tablet computers similarly equipped, there has been
increasing interest in development of analytical methods that
exploit such devices. If consumers already have the most expensive
part of a spectrometer in-hand and then add low-cost components

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_18, © Springer Science+Business Media New York 2015

259
260 Alexander Scheeline

to increase chemical and medical relevance, it is plausible that all

humans could become their own laboratory technicians. Here we
review existing methods, note limitations to such methods, and
extrapolate to likely technical and sociological developments.

2 Materials

1. Light sources. As for any portable instrument, low power and

small size are important. Common sources include: ambient
light (room light, sun), flash from the camera or cell phone,
independently powered LEDs, or laser pointers. Two types of
LEDs are common. The first is an array of red, green, and blue
LEDs, each with 20–30 nm bandwidth. While these can gener-
ate light that appears to be white to humans, there are gaps in
the spectrum for doing measurements. Such sources can be
employed as triple colorimeters. “True white” LEDs use a blue
gallium nitride LED (typically with peak emission ~450 nm)
and phosphors (rare earth oxides) to cover the entire visible
spectrum [2]. Battery powered ultraviolet sources (black-
lights/mercury lamps) have been used for fluorescence mea-
surements or wavelength calibration (Fig. 1).
2. Focusing optics. Whatever lenses are built into consumer devices
are employed by default. A common characteristic of spec-
trometry with consumer devices is the absence of external
focusing optics. Limitations are widely understood: absent col-

Fig. 1 Spectra of typical LED light sources. Three narrowband photodiodes, emitting blue, green, and red, can
be set to various relative intensities which the human eye perceives as any of a continuum of colors. Chemical
species respond only to the actual emitted wavelengths; there is almost no light from narrowband LEDs
between 580 and 610 nm. A fourth “True white” LED has no gaps in the spectrum from 430 to 730 nm.
Redrawn from images by superbrightleds.com
 CMOS/Cell Phone Spectrometry 261

limation of incoming light, dispersion devices are inefficient at

wavelength separation. Collimation with singlet lenses suffers
from chromatic aberration; achromats or superachromats over-
come chromatic aberration to a degree, but the better the lens,
the higher the cost. Mirrors do not have chromatic aberration,
but unobstructed collimation requires off-axis optical elements
that are expensive, at least in small quantities.
3. Dispersing media. Three approaches are common: gratings,
prisms, and filters. Because compact disks and DVDs have reg-
ularly spaced (though circular) grooves, they are a convenient
material from which to cut diffraction gratings. Holographic
or imprinted transparent film is also inexpensively available [3,
4]. Replica diffraction gratings purpose-made for spectroscopy
have typically been regarded as too expensive for this market.
Prisms suffer from nonlinear dispersion which makes wave-
length calibration more difficult than with gratings. While lin-
ear variable filters are commercially available in dedicated near
infrared devices [5, 6], they have not, to date, been used with
consumer-grade cameras.
4. Detectors. Electronic cameras began as vidicon tubes in the late
1920s. Affordable electronic video and still cameras came into
use in the 1990s, and have become ubiquitous in the current
century. Charge-coupled devices and CMOS detectors are
both widely employed, with CMOS most likely to dominate in
coming years because of the vast infrastructure supporting this
prolific technology [7]. Charge storage in CMOS and CCD
devices is typically on the order of 20,000 electron–hole pairs
per square micron. Because measurement of light intensity
from incoherent sources is a Poisson process, these detectors
are typically shot-noise limited, i.e., the uncertainty in number
of photons detected is the square root of the number of
detected photons. For a 1 megapixel detector 1 mm square,
the precision of measuring intensity half-way to “full well” (the
full 20,000 electrons in the 1 μm square pixel) is 10,0001/2 = 100
or a precision of 1 % (see Note 1). For humans, whose eyes
respond nonlinearly to light intensity, 1 % precision is more
than adequate (see Note 2). Further, common color encodings
(BMP, JPG) use 8-bit intensity ranges for each of red, green,
and blue intensity (24 bit color). Cyan magenta yellow key
(CMYK) is a continuous intensity/color mapping, but is lim-
ited to 8-bit resolution when paired with RGB values [8]. The
encoding of intensity is thus typically noted with resolution of
1 part in 256. For scientific measurements, 1 % precision (shot
noise) or 0.4 % (encoding) is a significant error in many
(though not all) cases. It follows that large array detectors
(e.g., the 41 megapixel detector in the Nokia Lumia 1020,
announced while this chapter was being written) are more
262 Alexander Scheeline

likely to be developed than are larger pixel, deeper well, more

precise devices. If cellcam/
webcam-based devices are to have significance, the limited pre-
cision and dynamic range of individual pixel response and
encoding must be overcome.
An example of the limitations of CMOS cameras is shown in
Fig. 2, using data obtained from a illuminating a MightEx
BCE-030U 3 MByte CMOS camera with light dispersed from
an LED. Because different pixels had varied illumination, a
100 ms exposure gave a variety of responses, recorded on the
ordinate. The response for each pixel to a 200 ms exposure was
then recorded, and the second response plotted as a function
of the first. When camera internal gain was set to unity, the
response was linear only for short-exposure response up to 64
counts (giving approximately 128 counts in the longer expo-
sure). Above 128 counts, readout was quite noisy (as shown by
the band of response in Inset A of Fig. 2), and saturated well
below 255 counts. Setting the internal gain to 2 (given
MightEx’s drivers and hardware encoding, a binary setting of
8 in the gain word), response closer to the ideal was obtained
(Inset B, Fig. 2). While still not perfectly linear or with every
pixel responding identically, the output was closer to that of an
ideal detector. Note that while the signal extrapolates to the
origin, there are no readings below ten counts. In common
with all analog detectors, there is an offset or bias that must be

255 255
Byte Value Image 2

Byte Value Image 2

0 0
0 255 0 255
Byte Value Image 1 Byte Value Image 1

Unity Gain Gain of 2

Fig. 2 Effect of gain setting on blue channel response of MightEx 3 Mpix CMOS camera. (a) Gain set to 1×. (b)
Gain set to 2×. Note gain saturation at less than full scale and noisy value for saturation in (a), while signal fills
entire scale in (b)
 CMOS/Cell Phone Spectrometry 263

measured and compensated. Such “dark subtraction” removes

the measurement offset, but does not expand the data encod-
ing range. Thus, this camera has a useful range of only 245
counts per color per pixel.
5. Data processing. Consumer devices typically produce data in
such common file formats as JPG, BMP, or TIFF. Less com-
mon are custom formats that allow greater color or intensity
bit depth than the 8 bits per color. While 8 bits may seem
adequate for spectrometry, this is somewhat misleading. The
raw data must be decoded to give intensity as a function of
wavelength, taking into account the response characteristics
of the detector and optical system. This requires software.
Photoshop® has many of the necessary capabilities. ImageJ [9]
also has the capabilities, but is not novice-friendly. The author
has found that free software mated with the capability to
interpret raw images, even awkwardly, is important if non-
specialists are to employ consumer cameras for scientific pur-
poses [10]. While the cited software was designed for
absorption spectrometry (calculating A(λ) = −log10(I(λ)/I0(λ))
across the visible spectrum, a number of high school teachers
and even an occasional geologist has asked if one might use a
transmission holographic grating and cellcam to do atomic
emission spectrometry. Since the naked eye can easily discern
atomic sodium (589 nm), copper (several green lines), bar-
ium (553 nm), or additional alkali elements (Li 670 nm, Cs
672 nm, or Sr 689 nm) through such gratings, one would
expect qualitative measurement to be possible since consumer
cameras respond in the this wavelength range (see Notes 3
and 4). One would not anticipate that quantitative results
will be feasible, as dynamic range is (again) too limited,
atomic lines from transition elements are too closely spaced
to allow adequate resolution with short focal length, inex-
pensive equipment, and many elements only emit in the
ultraviolet of infrared regions of the spectrum. Nevertheless,
for application to the alkalis, alkaline earths, and a few select
transition metals, a modification of existing software is cur-
rently being coded by the author. It will eventually be posted
at or linked to http://scheeline.scs.illinois.edu/CPS.
6. Assembled Instrument. An example of an instrument using a cell-
cam detector is shown in Fig. 3, discussed in the next section.

3 Methods

3.1 Perspective Chemical or biological measurements using cellular telephones

may be regarded as measurements made by transducers/instru-
ments linked to the phone via some communications port
264 Alexander Scheeline

Fig. 3 An open-air spectrometer made of inexpensive components (CR2032 battery, RL-W5020 LED, 500 line
per mm transmission grating, Motorola Droid RAZR Max running Android 4.1.2, 6 Mpix camera, normal expo-
sure, printed baseplate designed for classroom use.). Inset (a) shows the layout with component labels and
overlays of the optical path. Dispersion and saturation depend on diffraction order and exposure. Inset (b) shows
an overexposed image when the grating/camera distance is 0.12 m and the grating normal/camera observation
angle is approximately 30°. While no sample cuvette is present in Inset (a), the position in which it might be
placed is indicated. Inset (a) has been heavily retouched to remove distractions. Note that room light, allowing
the setup to be photographed, is an example of stray light if one is trying to perform spectroscopy

(BlueTooth or USB), or as a measurement made with a low

dynamic range camera attached to a computer. For spectroscopy
using the built-in camera, a wide range of arrangements and appli-
cations are covered by a patent [11]. Accurate rendering of color is
a challenge for any photography. Using a cellular phone camera for
scientific measurement is particularly challenging as all the factors
 CMOS/Cell Phone Spectrometry 265

influencing laboratory instrument measurement stability, precision,

accuracy, calibration, traceability, and dynamic range are influenced
by the diversity of environments in which the device is deployed,
the lack of temperature control, unpredictability of ambient light-
ing, possibly large amounts of dust, and unknowable cleanliness of
the lens. Additionally, a consumer device as the hub of a measure-
ment scheme suggests to the user that a measurement peripheral
should be small, light, inexpensive, yet resilient. Is it any wonder
that cellular instruments are only appearing slowly?

3.2 Spectrometry Whitesides, at Harvard, has been pursuing the use of cell phones
with Cellular for carrying out telemedicine [12, 13]. As an early adopter of this
Telephones technology, the group faced programming challenges and a rapidly
evolving platform; while current cellcams feature multiple mega-
pixels, the cameras available at the time of the cited work were 1
megapixel or smaller. Further, lighting conditions are difficult to
control, so interpretation of colors from raw images can be mis-
leading. The group advocated transmission of pictures to a central
base or authority for interpretation using tools such as PhotoShop®
to normalize response and extract chemically, environmentally, or
medically important information. With the explosive growth of
phone-born computing power, it is no longer obvious that central
processing is required. The extent to which systems can be hard-
ened against user limitations without real-time oversight by experts
is also unknowable as yet.
An early example of a spectrometer, developed by Mitch Nelson,
using a consumer-grade CCD camera is still accessible on the web
[14]. It employs a 35 mm diagonal holographic transmission grat-
ing, an approach that has come into favor due to the astonishingly
low cost of such dispersion. Also in common with other cellcam and
webcam-based systems, intensity calibration is not addressed.
At about the same time as Nelson’s development, a sleeker
device of similar layout was published by Field [15]. Billed as “high
resolution,” it is not clear how such a term should be defined. In
conversation with the author, Field noted that he did wavelength
calibration by aligning components so that undispersed light (zero
order) is centered in the instrument focal plane. Plus and minus
first order for specific wavelengths show up symmetrically on oppo-
site sides of zero order. Because fluorescent lamps emit strongly at
the mercury 546.1 nm green atomic emission line, pointing the
input slit of the instrument at a fluorescent light generates the req-
uisite light. Silence in the writeup about response linearity and off-
set leads one to believe that intensity measurements were accepted
at face value, without considering dark current or relative response
as a function of wavelength. The quantum efficiency of a detector
(number of measureable charges generated per incident photon) is
always wavelength-dependent. Thus, absent a traceable standard,
only relative measurements between similar sources is possible.
266 Alexander Scheeline

A crowd-sourced startup [16] has introduced a webcam-based

spectrometer through Public Lab [17, 18]. The instrument uses
a slice of DVD-ROM as a diffraction grating, with open source
software to read and assist interpretation of data [19]. Claimed
resolution is 10 nm or better. While the website does not discuss
intensity calibration, precision, or accuracy, one may expect some
member of the Public Lab community to address these issues at
some point. Public Lab maintains a website of uploaded spectra
[20]. This kit appears to be a reasonable means for widely dis-
seminating the ability to obtain fluorescence and emission spec-
tra. Whether it is sufficiently robust for precise measurement is
less clear. The instrument has been used for environmental sur-
veys. Lacking peer-reviewed publication, one hesitates to evaluate
the utility or scope of the measurements. What is clear, however,
is that citizens, not schooled in optical design, mechanical fabri-
cation, or software coding, can now purchase a spectrometric
instrument at sufficiently low cost so that its widespread use is
plausible. While commendable for educational and survey pur-
poses, one hesitates to use such an instrument for medical pur-
poses. Not only would it need to pass FDA approval for such use,
but one ought know what measurement pathologies are present
when non-specialists employ it. Example pathologies include:
stray light, temperature-dependent alignment, wavelength cali-
bration, and intensity response, corrosion of reflective coatings
on gratings or clouding of lenses due to atmospheric contami-
nants, and any degradation from moisture. Any of these might
lead to inaccurate results, but would not provide some obvious
indication of problems.
The author’s involvement in this field was secondary to teach-
ing instrumental analysis to students at Vietnam National University
of Science—Hanoi, Faculty of Chemistry. The circumstances are
described elsewhere [21]. A lab suitable for high school or college
students to learn the basics of spectrometry, including supporting
software, is available online [22]. Briefly, a polymeric holographic
transmission grating is used to disperse light from a battery-
powered white LED. If a cuvette is interposed between the light
source and grating, and can do crude absorption spectrometry,
provided the camera is stably positioned with respect to the grating
and cuvette and the exposure can be controlled. The design was
intentionally poor to assist students in seeing the parameters that
must be controlled in designing spectrometers. Because of the sim-
plicity, low cost, and free availability of software, the system has
been used on at least four continents, and has been adopted by a
number of secondary schools for teaching spectrometry as an
inquiry-based exercise. A rigorous discussion of its optical design,
precision, and other characteristics has been reviewed [10]. The
component layout is shown in Fig. 3a, with an overexposed photo
 CMOS/Cell Phone Spectrometry 267

of the dispersed LED light in Fig. 3b. This highlights the need to
embed software control in an “app” if serious work is to be done
on cell phones; the default exposure tends to saturation at least
somewhere in the image. Because of this unregulated default per-
formance, further detail is avoided here. No medical use should be
made of an instrument of this design. Nevertheless, one of the first
students to employ this “toy instrument” combined the gratings in
such a way that a novel stacked diffraction grating arrangement,
potentially useful for absorption, fluorescence, and reflectance
spectrometry, was devised. A patent for this grating arrangement is
now pending [23].
Other efforts to get inexpensive spectrometry into classrooms
has exploited the wide availability of cellcams and webcams.
A lightbox or computer screen can provide reasonably uniform back-
lighting for an array of cuvettes so that an entire working curve can
be obtained in a single photographic frame [24]. Spectral resolution
can either be from the RGB image plane resolution of the camera or
from placing a bandpass filter in front of the camera. The authors
used PhotoShop® to extract intensity data, and kept absorbance below
1 to ensure adequate digitization resolution. Published data showed
little evidence of stray light(see Note 5). In a related approach, an
array of LEDs backlit a multiwell plate. The RGB response of a web-
cam measured absorbance in each of the wells [25].

3.3 Example An example of how access to the idea of an inexpensive spectrom-

eter can lead to wide involvement in the scientific enterprise were
two projects by then high school student Lakshmi Raju. Soon after
publicity concerning the instrument appeared [26], Raju translated
the Pascal code for the instrument into Matlab, and earned recogni-
tion in the Alabama State Science Fair [27]. She also created a visi-
ble spectrometer using a cellcam to create spectral images. Further
development the following year resulted in measurement of atmo-
spheric water vapor [28]. In the latter work, the rapid development
of cellcams figured in instrument design. Early cellcams had simple
lenses that transmitted light throughout the visible and near infra-
red; more recent cameras include filters that block near infrared
wavelengths not visible to the human eye. Because water vapor has
significant absorbance near 970 nm, silicon detectors can measure
absorption, provided no filters are in the lenses, and an external
bandpass filter is used, a bandpass filter specific to this wavelength is
part of the lens assembly. A convenient infrared source for testing
camera performance was an electric stove element set to “high,”
until it glowed orange (see Fig. 4). Photographing through a near
infrared filter results in an image of appropriate color; absent the
filter, the element appears violet (see Note 6).
RGB data was combined to form a grey-scale nephelometer by
da Silva and coworkers [29]. While the cited work also employed
268 Alexander Scheeline

Fig. 4 Two cell cams look at a hot stove. (a) Samsung Omnia Windows CE phone; camera has no near infrared
filter. Note purple tone to heating element. (b) Motorola Droid phone with near infrared filter. Color of an orange
plastic jar top included for comparison; the top in (b) as well as the burner color there is approximately the
same as the author’s perception. Not only does this illustrate the need for proper filtering to avoid interference
from the near-infrared, but also the difficulty of accurately measuring color from camera images

infrared absorption to measure the concentration of adipic acid

whose solubility was being determined, the webcam measurements
are applicable to any precipitable substance. While the authors cite
a source where they describe their software [30], attempts to
download the cited article were unsuccessful.
Recent work continues to expand the use of the 8-bit per color
CMOS cameras in cell phones for colorimetry. Why not use the
phone display LEDs as the illumination source? Iqbal and Eriksson
have done just that [31]. By viewing clear solution and sample in
the same frame, they obtain backlight intensity data simultaneously
with transmittance data. They employ not only pure channel RGB
information but also the overlap channels where pairs of colors
respond to a sample simultaneously. Because of the large number
of pixels uniformly illuminated in a field of view, precision is sur-
prisingly good. A mechanical stand of appropriate dimensions is
required so that the field of view of the camera and the region of
measurement interest optimally overlap. Similarly, Oncescu and
coworkers image pH paper with a reference strip adjacent to quan-
tify sweat and saliva pH [32]. They convert RGB to hue in their
smartphone app as response is more nearly linear. Precision is a few
tenths of a pH unit. A number of plots appear to ignore measure-
ment precision as time series are plotted with “connect the dots”
lines, with little indication that the point-to-point variations are
statistically significant.
A Netherlands-based spectrometer company, Avantes, in the
spring of 2013 sponsored a collaborative science project, in which
inexpensive spectropolarimetric devices were distributed to interested
 CMOS/Cell Phone Spectrometry 269

citizens in Leuven. Impingent light was split into two beams,

p-polarized light transmitted from one beam while s-polarized light
transmitted via the other, and then dispersed by a transmission grat-
ing. The two dispersed beams were then directed to the participants
cell phone camera. At a particular moment, all participants were to
take a video, starting at the zenith, and gradually scanning to the
horizon. From the angular and spectral dependence of scattered
light, particulate loadings in the air were deduced [33, 34]. While
the dynamic range, alignment, and calibration of individual instru-
ments were uncontrolled, the central limit theorem suggests that
some, perhaps biased, result will be obtained. After comparison to
simultaneously obtained measurements from better characterized
instruments, an example of the value of crowd-sourced data and of
the cost/volume/quality trade-offs will be available.
CMOS and CCD cameras are dimensionally precise regardless
of pixel response. With suitable calibration, they can thus be used
for wavelength measurement more readily than they can be used
for intensity measurement. An example is the wavemeter of White
and Scholten [35]. A CCD with 3.5 μm wide pixels was placed at
the focal plane of a 200 mm focal length lens. Over a narrow wave-
length range (6.5 nm), the CCD/lens/grating combination dis-
played dispersion of 3.5 pm/pixel and resolution for finding the
centers of symmetrical atomic emission lines of 0.7 pm near
λ = 780 nm. Calibration required a quadratic correction to disper-
sion, and changes in calibration from temperature drift in the grat-
ing and refractive index changes in air as a function of humidity
and temperature were required. Given the short focal length of
intact cellcams and webcams, resolution at this level is unlikely to
be common in consumer devices. See Note 7 for details.
Barrel distortion (apparent curvature of straight line objects
towards the edges of an image) prevents one from gaining a multi-
plex/parallel observation advantage from simply dispersing light
along one axis of a camera and summing along columns or rows to
increase signal in the perpendicular direction. The distortion results
in wavelength blurring and degraded resolution. By breaking up
the field of view into separate rows, each calibrated independently
for wavelength dispersion, one can obtain information from several
different sources simultaneously. Stacks of fiber optics placed in the
object plane of a webcam, with a diffraction grating near the camera
lens, is a means to this end [36]. For the particular focal length and
grating reported, a wavelength range from 400 to 655 nm could be
simultaneously observed. Orthogonally, five fiber-optic channels
were imaged. The blur in the fiber optic images limited resolution
to 10 nm, but allowed significant signal averaging (Fig. 5).
Coarse wavelength resolution using only the color separation
of the RGB planes in the webcam detector allows fluorescence
imaging in a microflow batch reactor and, by implication, in any
microfluidic system with suitable light access [37]. Webcams typi-
270 Alexander Scheeline

Fig. 5 Images from two figures in ref. [36]. (a) Layout of fiber optics and webcam detector. (b) Zero order (right
side) and first order (left side) for (a) 465, (b) 591, and (c) 639 nm for light through the five fiber optics. Note
comatic blur, field curvature, and offset. When using a minimum of additional optics with consumer cameras,
software must compensate for non-ideal imaging. Images copyright by the Society for Applied Spectroscopy.
Used with permission

cally image at 30 Hz or slower, so flow cytometry (where time

resolution of 1 ms or less is often required) may not be feasible.
However, imaging of labeled micro-liquid chromatography peaks
or capillary electrophoresis peaks may work. Such rapid, low reso-
lution, low dynamic range measurements are the opposite trade-
off from that chosen in recent work by this book’s editor and
coworkers [38]. They summed multiple images over an extended
time to accumulate measureable signals from low level fluores-
cence. Summing background-subtracted images improves signal in
proportion to the number of images, and signal-to-noise ratio as
the square root of the number of images if shot noise is limiting.
Surprisingly, signal-to-noise ratio improved in direct proportion to
signal. This is consistent with readout limitation; the digitization
resolution of the camera was too coarse to see low level signals. As
noted long ago by Horlick [39, 40], there are times when adding
noise can actually improve overall measurement precision, and this
appears to be one of them.
 CMOS/Cell Phone Spectrometry 271

Non-spectroscopic scientific use of cellcams has been a focus of

the Ozcan group at UCLA [41–43]. Imaging, perhaps through a
filter, rather than spectral analysis appears to be the main theme of
this group’s work. Wavelength resolution is exclusively determined
by the filter (within the response window of silicon). A sample may
be placed in contact with a CMOS detector so that diffraction by
micron-scale objects leads to Airy patterns in an image. One can
count bacteria or dust motes in this manner, since the individual
pixels are smaller than the objects under study (e.g., an E. coli bac-
terium may be 10 μm long, stretching across 5–8 pixels, depending
on the camera). Dead pixels influence only isolated points in the
image. Up-to-date information is available online [44].
One senses that mobile spectroscopy in specific and personal
scientific measurement in general is neither in its infancy nor close
to mature. The computational capabilities of cell phones and tab-
lets are immense and, largely, underexploited. The capabilities of
miniature cameras are substantial, at least in terms of numbers of
pixels if not in intensity dynamic range. The trend towards indi-
vidualized and personalized products of all types is widely noted
[45]. It follows that individuals will want personally obtained mea-
surements to drive their personally specified choices. Cost pres-
sures in medicine auger for as much laboratory work as practical to
be done by individuals so that they need not consume professional
time and resources for routine, or even acute, diagnosis. Blood
glucose measurement by diabetics has been personalized on a mass
scale for decades. Additional testing becomes practical only if the
instrumentation is widely and cheaply available. While one way to
do this is to turn many measurements into glucose measurement
[46], another is to turn every cell phone into a laboratory.
Electrochemical sensors have been interfaced to any system with a
USB port [47]. The technology discussed here suggests that opti-
cal measurements, at least with resolution coarser than 5 nm in the
visible region of the spectrum, may also be universally available in
the near term.

4 Notes

1. Noise in any spectrometric measurement is the root mean

square average of the sum of assorted noise contributors
including digitization (or readout) noise, shot noise, sample
positioning variability, and light source flicker. See [48–51].
2. Digital data is represented as binary numbers. Converting binary
numbers to conventional, base 10, numbers gives the discrete
encodings available for digital data. A bit can have values 0 or 1
only. A nybble (4 bits) can have values 0, 1, …, 14, or 15. A byte
has values 0, 1, …, 254, 255, which is 256 different levels.
272 Alexander Scheeline

Words (2 bytes, 4 nybbles, 16 bits) have values 0, 1, …, 65,535.

Scientific cameras are typically designed to encode data with as
much resolution as the corresponding detector integrated cir-
cuit allows. If one can store 108 electrons in a single photodiode
in a diode array [52], then the shot noise limited precision of
each diode is (108)1/2 = 1 part in 104. Representing 104 in binary
notation requires 14 bits, which can encode integers from 0 to
16,383. Because digitization itself is noisy, one ensures that the
digitization is not a significant contributor to uncertainty for
such detectors by using a 16 bit analog to digital converter. For
CMOS detectors with 5 × 104 electrons per pixel, shot noise lim-
ited precision is 1 part in (5 × 104)1/2 or 1 part in 224 (rounding
the square root to the nearest integer). This can be represented
in one byte. Because CMOS full well capacity in small cameras is
rarely even as large as 5 × 104 electrons, cellcams and webcams
would waste memory to encode each color in more than one
byte. BMP and JPG files concatenate 3 bytes to encode the
intensities of the red, green, and blue light sensed in each pixel.
Cameras with larger detectors and thus deeper wells to hold
more electrons encode in a variety of RAW formats with more
than 8 bits per color per pixel.
3. For plane grating spectrometers, one typically assumes light is
d l d cos b
collimated so that nλ = d(sin α + sin β) and = , with
db n
n = diffraction order, λ = wavelength of interest, d the spacing
between grating grooves or lines (and, alas, also the usual
mathematical meeting for “differentiate”), α the angle between
the normal to the plane of the grating and incoming direction
of the light wavefront, and β the angle between the normal and
the exiting, observable ray. For a focal length f collecting dif-
fracted light and focusing it on a detector with pixels of width
Δx, the nominal range of wavelengths striking the pixel is
d cos b
l= x . If the entrance aperture is imaged to a spot of
nf
width Δx or smaller, instrument resolution is nominally 3Δλ. If
the entrance aperture’s image is larger than Δx, resolution and
Δx are inversely proportional.
4. Calibrating any spectrometer for wavelength requires observa-
tion of at least two wavelengths, a particular wavelength in two
or more orders, or one wavelength in a non-zero order plus
that same wavelength in zero order. For small angles β, one
may approximate dispersion as independent of wavelength,
d
i.e., l ~ x . One may calculate the error of this lineariza-
nf
tion by integrating the dispersion equation with respect to β.
Because optical aberrations in lenses and mirrors also distort
the spectrum, it is common to empirically rather than theoretically
carry out calibration. Fitting dispersion to a cubic dependence
on wavelength is frequently adequate [53–55].
 CMOS/Cell Phone Spectrometry 273

5. Stray light is any light reaching the detector that has not
followed a path that corresponds to the physics and chemistry
expected for an experiment. Room light, light from junctions
on integrated circuits acting as weak LEDs, and light scattered
from dust that sprays around the interior of an instrument are
examples of stray light.
6. Webcam and cellcam detection can be extended to the X-ray
region using a 5 μm aluminum foil transducer to convert X-rays to
avalanches of “dark counts” in nearby pixels [56]. The CCD or
CMOS transducer must be adjacent to the foil, so the foil must
be placed inside the camera housing. Backscatter from nearby elec-
tronics or scattering from housing components may blur images.
7. What resolution is feasible with a cellcam? Lens focal length is
4–8 mm. Readily available gratings have 300–3,600 lines per
mm, so an example at 1,200 lines per mm is useful. First, if the
light directed at the grating is collimated, using the relation-
ship in Note 3 results in dispersion of 0.06 to 0.13 nm/pixel
for 1.5 μm pixels, ignoring aberrations (imperfect imaging)
and the blur from the (de)magnified image of the entrance
aperture. Realistically, with a 100 μm diameter fiber optic as
light source and a 100 mm focal length collimation lens or
mirror, the source image diameter would be between 8/100
and 4/100 × 100 μm in diameter, or 3–5 pixels, corresponding
to 0.2–0.8 nm resolution. This is consistent with the author’s
measurements using a 50 μm pinhole entrance aperture,
200 groove/mm grating, 25 mm focal length collimating
optic, and 8 to 12 mm focal length zoom lens on the MightEx
camera mentioned elsewhere in this chapter.

Acknowledgments

The work reported here was supported, in part, by NIH Grants

U01DE017855 (Bau, Mauk) and K25AI099160 (Liu), and a
grant from the Commonwealth of Pennsylvania’s Ben Franklin
Technology Development Authority through the Ben Franklin
Technology Partners of Southeastern Pennsylvania (Bau, Sadik).

References
1. Rieke G (2003) Detection of light: from the log/displayproduct.cfm?productID=1490 ,
ultraviolet to the submillimeter, 2 edn. (downloaded 12/27/2009)
Cambridge University Press, Cambridge 4. Rainbow Symphony Inc (2009) Rainbow
2. Su J-C, Song S-F, Chen H-S (2011) Symphony Store Holographic Gratings. http://
Chromaticity stability of phosphor-converted www.rainbowsymphonystore.com/scienanded1.
white light-emitting diodes with an optical fil- html, (downloaded 12/27/2009)
ter. Appl Opt 50:177–182 5. Coates J (2010) New miniaturized spectral
3. Edmund Scientific diffraction grating page, measurement platforms covering the visible to
http://www.edmundoptics.com/onlinecata- the mid-IR, http://depts.washington.edu/
274 Alexander Scheeline

cpac/Activities/Meetings/Fall/2010/docu- 22. Kelley KD, Scheeline A (2009) Cell phone

ments/ NeSSI_Workshop_November_2010_ spectrophotometer. J Analyt Sci Digital Lib.
Coates.pdf, (downloaded 10/7/2013) entry 10059. Reprinted in m-science: sensing,
6. JDSU (2013) MicroNIR spectrometer, computing, and dissemination, Cannesa E,
http://www.jdsu.com/en-us/Custom- Zennaro M (ed) (The Abdus Salam
Optics/Products/a-z-product-list/Pages/ International Centre for Theoretical Physics,
m i n i a t u r e - n e a r- i n f r a r e d - S p e c t r o m e t e r. November, 2010)
aspx?rcode=micronir, (downloaded 23. Scheeline A, Bui TA (2011) Energy dispersion
7. Holst GC, Lomheim TS (2007) CMOS/CCD device. pending, 10/12/2011 (provisional),
sensors and camera systems. JCD, Winter Park, FL 8/28/2012 (utility), 4/18/2013 (published)
8. rapidtables.com (2013) CMYK to RGB color 24. Kehoe E, Penn RL (2013) Introducing colori-
conversion, http://www.rapidtables.com/ metric analysis with camera phones and digital
convert/color/cmyk-to-rgb.htm, (down- cameras: an activity for high school or general
loaded 10/7/2013) chemistry. J Chem Ed 90:1191–1195
9. Rasband W (2013) ImageJ: image processing 25. Castillo J, Gutierrez H, Vitta Y, Martinez M,
and analysis in Java, http://rsbweb.nih.gov/ Fernandez A (2007) Development and testing
ij/, (downloaded 10/8/2013) of multi-well plates absorbance reader for clini-
10. Scheeline A (2010) Focal point: teaching, cal analysis using inexpensive web-cam. Proc
learning, and using spectroscopy with com- SPIE Int’l Soc Opt Eng.
mercial, off-the-shelf technology. Appl 6755:67550W-67551–67550W-67559
Spectrosc 64:256A–268A 26. Ahlberg E (2010) Can you analyze me now?
11. Wang SX, Zhou XJ (2008) Spectroscopic sen- Cell phones bring spectroscopy to the class-
sor on mobile phone. U.S. Patent 7,420,663, 2 room, http://www.news.illinois.edu/
Sept 2008 news/10/1007scheeline_spectrophotometry.
html, (downloaded 10/7/2010)
12. Martinez AW, Phillips ST, Carrilho E, Thomas
SW III, Sindi H, Whitesides GM (2008) 27. Raju L (2011) Development of an Inexpensive
Simple telemedicine for developing regions: Spectrophotometer and a Matlab Spectral
camera phones and paper-based microfluidic Analysis Program. http://spie.org/Documents/
devices for real-time, off-site diagnosis. Anal Courses/Education_Outreach/Lakshmi-
Chem 80:3699–3707 Raju-PR-2013.pdf; http://asef.uah.edu/
Documents/2011%20ASEF%20Special%20
13. Whitesides GM (2009) Paper diagnostics – using Awards.pdf, (downloaded 10/13/2013)
first world science in developing economies. In:
Pittsburgh conference on analytical chemistry 28. Raju L (2012) Development of a Low Cost
and applied spectroscopy, Chicago, IL Infrared Spectrophotometer and a Matlab
Program to Detect Terrestrial and Extra-
14. Nelson M (2010) WinSpec: CCD spectrome- terrestrial Water Vapor. http://www.uab.edu/
ter on a budget, http://www.materialintelli- carsef/2012resuilts; http://www.googlelu-
gencellc.com/mitchnelson/W inSpec/ , narxprize.org/teams/jurban/blog/more-lak-
(downloaded 10/9/2013) shmi-rajus-spectrophotometer, (downloaded
15. Field SQ (2010) High resolution spectro- 10/13/2013)
graph. Make 24:58–61 29. da Silva APM, de Oliveira PB, Bandini TB,
16. Warren JW (2012) Public Lab DIY spectrome- Barreto AG Jr, de Sena RC, Cajaiba de Silva JF
try kit, http://www.kickstarter.com/projects/ (2013) Low-cost system based on image analy-
jywarren/public-lab-diy-spectrometr y-kit , sis to determine solubility curves. Sens Actuat
(downloaded 10/9/2013) B 177:1074–1077
17. (2013) Public Lab website, http://publiclab. 30. Sena RC, Soares M, Pereira MLO, Silva RCD,
org/, (downloaded 10/9/2013) Rosário FF, Silva JFC (2011) A simple method
18. (2013) Public Lab Store, http://store.publi- based on the application of a CCD camera
clab.org/products/desktop-spectrometry-kit, as a sensor to detect low concentrations of
(downloaded 10/9/2013) barium sulfate in suspension. Sensors 11:
19. (2013) Spectral Workbench, http://publiclab. 864–875
org/wiki/spectral-workbench, (downloaded) 31. Iqbal Z, Eriksson M (2013) Classification and
20. PublicLab (2013) Spectral workbench DIY quantitative optical analysis of liquid and solid
material analysis, http://spectralworkbench. samples using a mobile phone as illumination
org/, (downloaded 10/9/2013) source and detector. Sens Actuat B
21. Scheeline A (2010) Teaching in Hanoi: good 185:354–362
mornings in Vietnam. Anal Bioanal Chem 32. Oncescu V, O'Dell D, Erickson D (2013)
398:2751–2753 Smartphone based health accessory for colori-
 CMOS/Cell Phone Spectrometry 275

metric detection of biomarkers in sweat and 44. Ozcan A (2013) Ozcan Research Group
saliva. Lab Chip 13:3232–3238 Refereed Journal Publications, http://inno-
33. Hoving W (2013) Environmental monitoring/ vate.ee.ucla.edu/refereed-journal-
spectroscopy: crowd-sourced measurements publications.html, (downloaded 10/09/2013)
offer a unique view of pollution’s effects. 45. Titlow JP (2011) Why user-customized prod-
BioOptics World, vol 6. http://www.biooptic- ucts are the future of business (for real this
sworld.com/ar ticles/print/volume-6/ time), http://readwrite.com/2011/04/15/
issue-5/features/environmental-monitoring- user-customized-products-future-of-business#
spectroscopy-crowd-sourced-measurements- awesm=~okg94zZ2kkPR4l, (downloaded
offer-a-unique-view-of-pollution-s-effects.html 10/14/2013)
34. University L (2013) Measuring particulates 46. Xiang Y, Lu Y (2011) Using personal glucose
with your smartphone, http://www.research. meters and functional DNA sensors to quantify a
leiden.edu/news/measuring-particulates- variety of analytical targets. Nat Chem 3:697–703
with-your-smartphone.html, (downloaded 47. Lillehoj PB, Huang M-C, Truong N, Ho C-M
10/10/2013) (2013) Rapid electrochemical detection on a
35. White JD, Scholten RE (2012) Compact dif- mobile phone. Lab Chip 13:2950–2955
fraction grating laser wavemeter with sub- 48. Ingle JD, Crouch SR (1972) Evaluation of
picometer accuracy and picowatt sensitivity precision of quantitative molecular absorption
using a webcam imaging sensor. Rev Sci spectrometric measurements. Anal Chem
Instrum 83:113104 44:1375–1386
36. Subriddetchkajorn S, Intaravanne Y (2012) 49. Rothman LD, Crouch SR, Ingle JD (1975)
Home-made N-channel fiber-optic spectrome- Theoretical and experimental investigation of
ter from a web camera. Appl Spectrosc factors affecting precision in molecular absorp-
66:1156–1162 tion spectrophotometry. Anal Chem
37. Lima MB, Andarde SE, Barreto IS, Almeida 47:1226–1233
LF, Araujo MCU (2013) A digital image-based 50. Ingle JD, Crouch SR (1988) Spectrochemical
micro flow batch analyzer. Microchem J analysis. Benjamin, New York
106:238–243 51. Skoog DA, Holler FJ, Crouch SR (2007)
38. Balsam J, Bruck HA, Kostov Y, Rasooly A Principles of instrumental analysis. Thomson
(2012) Image stacking approach to increase Higher Education, Belmont, CA
sensitivity of fluorescence detection using a low 52. Inc. HS (2013) VS7000-PDA Miniature PDA
cost complementary metal-oxide- Spectrometer, http://www.horiba.com/fil-
semiconductor (CMOS) webcam. Sens Actuat eadmin/uploads/Scientific/Documents/
B 171–172:141–147 OEM/VS-7000-PDA-2013.pdf, (downloaded
39. Horlick G (1975) Reduction of quantization 12/24/2013)
effects by time averaging with random added 53. Hsu W-H, Sainz MA, Coleman DM (1989) An
noise. Anal Chem 47:352–354 aberration-corrected time- and spatially-
40. Belchamber RM, Horlick G (1981) Use of resolved spectrometer for studies of transient
added random noise to improve bit-resolution discharges. Spectrochim Acta 44B:109–121
in digital signal averaging. Talanta 28:547–548 54. Miller DL, Scheeline A (1993) A computer
41. Tseng D, Mudanyali O, Oztoprak C, Isikman program for the collection, reduction, and
SO, Sencan I, Yaglidere O et al (2010) Lensfree analysis of echelle spectra. Spectrochim Acta
microscopy on a cell-phone. Lab Chip 48B:E1053–E1062
10:1787–1792 55. Sadler DA, Littlejohn D, Perkins CV (1995)
42. Zhu H, Yaglidere O, Su T, Tseng D, Ozcan A An automatic wavelength calibration proce-
(2010) Cost-effective and compact wide-field dure for use with an optical spectrometer and
fluorescent imaging on a cell-phone. Lab Chip array detector. J Anal Atom Spec
11:315–322 10:253–257
43. Mudanyali O, Dimitrov S, Sikora U, 56. Lane DW (2012) X-ray imaging and spectros-
Padmanabhan S, Navruz I, Ozcan A (2012) copy using low cost COTS CMOS sensors.
Integrated rapid-diagnostic-test reader plat- Nucl Instrum Methods Phys Res B Beam
form on a cellphone. Lab Chip 12:2678–2686 Interact Mater Atoms 284:29–32
 Chapter 19

Mobile Phone Based Electrochemiluminescence Detection

in Paper-Based Microfluidic Sensors
Jacqui L. Delaney and Conor F. Hogan

Abstract
The development of simple, inexpensive paper-based sensors for medical diagnostics and other applications
is now an important emerging area in the field of biosensors; however, the electronic instrument or reader
used to interrogate such sensors adds significantly to the cost of the analysis. In this chapter we describe the
design and construction of novel, low-cost disposable electrochemiluminescent (ECL) sensors based on
screen printed carbon electrodes and paper-based microfluidics. Moreover, a method to interrogate these
sensors using only a mobile phone is articulated. This is realized by exploiting the audio output of the device
to achieve electrochemical control, while using the camera to detect the resulting light emitted during the
ECL reaction. The combination of cell phone technology with low-cost paper microfluidic sensors dramati-
cally reduces the cost of sensing and has the potential to enhance health-care outcomes by exploiting the
functionality, connectivity, and close to worldwide penetration of mobile phone technology.

Key words Paper microfluidics, Low-cost sensing, Mobile phone-based sensing, Electro-
chemiluminescence

1 Introduction

The growth in mobile phone usage since the late 1990s has been phe-
nomenal. Moreover, their capability has grown dramatically in that
time. Even more remarkable is the projection that virtually all of the
inhabitants of this planet (including those in the developing world) will
have access to such a device within the next 5–10 years [1]. We believe
that paper-based microfluidic sensors combined with the power and
ubiquity of mobile phones can be harnessed to facilitate a revolutionary
approach to medical diagnostics in the developing world.
Microfluidic paper-based analytical devices (or μ-PADs) have
the significant advantage of not requiring any external means of
fluid transport which occurs via capillary action. Also, they require
only small sample volumes, the paper may filter or otherwise
separate the components of the sample, they are easy to store and
transport and are readily disposed of safely via incineration. While

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_19, © Springer Science+Business Media New York 2015

277
278 Jacqui L. Delaney and Conor F. Hogan

most of the research in this area has focused on colorimetric [2]

and electrochemical [3] detection platforms, we have sought to
develop electrochemiluminescence (ECL)-based strategies because
of the inherent advantages of electrochemical control coupled with
the sensitivity of a luminescence-based technique.
The three main contributors to the cost of a sensor are: the
materials, the fabrication, and the detection hardware. Therefore,
we have sought to develop a sensing strategy incorporating very
simple paper-based fluidic substrates, which can be easily produced
without recourse to expensive fabrication facilities. A device such as
an inkjet printer or an office laminator can be used to produce paper
microfluidic substrates which are combined with screen printed
electrodes (SPEs) to create simple, cheap, disposable sensors.
Detection strategies used with paper-based sensors to date
have been predominantly colorimetric in nature [2] with quantita-
tion achieved by analysis of color intensity using flat-bed scanners
or cameras. We have shown ECL to be a highly promising tech-
nique [4–7] particularly for low-cost sensing, because the built-in
camera of a mobile phone can function as a luminescence detector
[8] and, as ECL is performed in the dark it is independent of ambi-
ent light, giving it the capacity for far greater sensitivity. Apart from
cost reduction, an important aspect of mobile phone based sensing
is the possibility for it to facilitate telemedicine by transmitting
analytical results to a central database where it can be reviewed by
a medical practitioner. Importantly, such a transmission would
contain a variety of other important data such as a date and time
stamp as well as GPS triangulation data.
However, ECL detection is also an electrochemical technique
and therefore requires an application of a potential in order to
achieve the electrolysis which triggers light emission. Unfortunately,
the extra hardware required to achieve this may undermine the
goals of simplicity and keeping costs extremely low. We have over-
come this problem by exploiting another familiar feature of mobile
phones: their ability to play sound. By making an electrical connec-
tion to the electrodes via the phones audio jack and playing an
appropriate .wav or .mp3 file, we found that the working electrode
could be suitably polarized to initiate the electrochemical reaction
leading to light emission [9].

2 Materials

2.1 Chemicals 1. All solutions were made up using deionized water with a resis-
and Materials tivity of at least 18 MΩ/cm.
2. 0.1 M phosphate buffer solutions (pH 7.5 unless otherwise
stated), were prepared using sodium phosphate dibasic
(≥99 %, Sigma-Aldrich) and sodium phosphate monobasic
(≥99 %, Sigma-Aldrich).
 Mobile Phone Based Electrochemiluminescence Detection… 279

3. Tris(2,2′-bipyridyl)ruthenium(II) chloride hexahydrate

(Ru(bpy)3Cl2) (99 %) was purchased from Sigma-Aldrich or
Strem Chemicals. The desired concentration was made in
phosphate buffer solution (pH 7.5, 0.1 M).
4. 2-(dibutylamino)ethanol (DBAE) (99 %) was purchased from
Sigma-Aldrich. The desired concentration was made in phos-
phate buffer solution (pH 7.5, 0.1 M).
5. β-nicotinamide adenine dinucleotide in its reduced form
(NADH) (94 %) was purchased from Sigma-Aldrich. Solutions
were made fresh to the desired concentration using phosphate
buffer solution (pH 7.5, 0.1 M).
6. L-Proline (99 %) was purchased from Sigma-Aldrich. Solutions
were made fresh daily to the desired concentration using phos-
phate buffer solution (pH 10, 0.1 M).
7. Potassium chloride (≥99.5 %) was purchased from Sigma-
Aldrich. 3 M solutions of KCl in deionized water were used
for plating of counter-electrodes.
8. Silver nitrate (99.9 %) was purchased from M&B Laboratory
Chemicals. 10 mM solutions of AgNO3 in deionized water
were used for plating of counter-electrodes.
9. Alkenyl ketene dimer (AKD) (Precis 900, Hercules Australia
Pty Ltd.) was used as the cellulose hydrophobization agent for
the paper microfluidics to create the hydrophobic barrier.
2 % v/v solutions of this compound in n-Heptane were used.
10. Paper used for the fabrication of the paper microfluidics was
Whatman No. 4 filter paper.

2.2 Instruments 1. The pH of the phosphate buffers was tested using a MEP
Instruments, Metrohm 827 pH Lab Meter and an MEP
Instruments’ Metrohm 6.0228.010 electrode.
2. Cyclic voltammetry was performed using a CH Instruments
(Texas) potentiostat (660B or 620C).
3. Acquisition and analysis of voltage and ECL intensity signals
was performed using an eDAQ e-corder 401 model ED401.
4. Low-cost screen printed electrodes (SPEs) were manufactured by
Zensor R&D (Taiwan) (Zensor TE100 screen printed voltam-
metric electrodes, Cat no. ET077). SPEs contain a 3 mm diam-
eter working electrode and an arc shaped auxiliary electrode
(both made of graphite carbon powder) and a Ag/AgCl pellet
reference electrode all on a 50 × 13 mm plastic substrate (Fig. 2c).
5. ECL optimization experiments were performed using an Eco
Chemie, μ-Autolab type II potentiostat and an Electron Tubes
Ltd. (model 98285B) photomultiplier tube (PMT) in a
custom-built light-tight Faraday cage. A high voltage power
supply was used biased at 500 V (unless otherwise stated).
280 Jacqui L. Delaney and Conor F. Hogan

The PMT signal was amplified using either a custom-made

amplifier or using a Ames Photonics Inc. amplifier model
TA-GI-74 (D7280).
6. Sensor lamination was performed using a GBC Heatseal H212
office laminator. There were two types of pouches used; either
A5 GBC or Ibico document pouches (80 μm × 2 thickness) or
GBC A5 Peel n Stick Pouches (75 μm × 2 thickness).
7. The template for the design of the paper microfluidics (when
used) was prepared using Adobe illustrator. The channel was
7 mm long by 1.5 mm wide, and the detection zone was 8 mm
in diameter (Fig. 2b).
8. Printing of the AKD-heptane hydrophobization solution was
done using a reconstructed commercial digital inkjet printer
(Canon Pixma ip4500). The reconstruction consisted of
replacing the contents of the ink cartridge with the AKD-
heptane solution as described in ref. 10.
9. The sensor holder (Fig. 1a) was fabricated using a 3D printer (UP!
3D printer, purchased from http://3dprinting.co.nz/, New
Zealand) using black ABS plastic to minimise light interference.
10. A Samsung I8910 HD icon mobile phone was used as a
photodetector in the earlier proof-of-concept stages of the
work. When this phone was used, the camera setting called
“fireworks” was activated which provided the best detec-
tion results. The fireworks setting provided slower shutter
speeds with a high ISO which allowed more of the emitted
light to be captured.
11. A Samsung Galaxy S (i9000) smartphone running Google
Android(TM) version 2.2 was used to initiate the ECL reac-
tion, detect the light emission and analyze the results. The
phone was rooted to attain privileged control (known as “root
access”) and a custom kernel was installed to allow installation
of Voodoo Sound (http://project-voodoo.org/) which
enabled greater control over the phone’s audio hardware.
12. A Java based Android application was written to initiate,
detect, and analyze ECL intensities on the phone. The appli-
cation was developed using Eclipse(TM) and the default
Android development kit by Google(TM).

3 Methods

3.1 Fabrication We will describe the two main approaches to the fabrication of the
of Paper Microfluidic paper microfluidic substrates used in this work. The first is the
Sensors method of Li et al. [10] where the fluidic channel is defined by a
hydrophobic barrier printed into the paper using inkjet printing.
The second referred to here as the “cut and seal” method, is where
the paper fluidic element is cut or punched to the desired shape
 Mobile Phone Based Electrochemiluminescence Detection… 281

Fig. 1 (a) is a CAD drawing of the sensor holder used to interface the sensor with the mobile phone camera
and exclude ambient light. The photographs in (b) without the phone and (c) with the phone show how the
holder is used. From ref. 9

and the perimeter of the paper is sealed by laminating, similar to

the approach described by Fenton et al. [11].

3.1.1 Fabrication 1. A design of the paper microfluidics was constructed in Adobe

of Inkjet Printed Paper illustrator. The design was then printed onto Whatman filter
Microfluidic Sensors paper (No. 4) using AKD. Figure 2a shows an example of the
design used for the printing process. The black area denoted the
areas where the hydrophobic AKD solution was to be printed.
282 Jacqui L. Delaney and Conor F. Hogan

Fig. 2 The fabrication and operation of the paper-based microfluidic ECL sensor. Paper microfluidics were
produced in bulk using a conventional inkjet printer (a). Once the fluidics were loaded and dried they were
individually cut to size (b). The paper microfluidics were then aligned and fixed onto the face of the SPE via
lamination (c). A drop of analyte was introduced via a small incision at the bottom of the channel. Once the
paper was fully wetted, the sensor was placed in front of the camera of the mobile phone. A potential was
applied and the resulting ECL emission was captured on the mobile phone (d). (From ref. 8)

2. The printed sheets were placed in a 100 °C oven for 8 min to

cure the AKD onto the cellulose fibers.
3. The paper microfluidics were then loaded with 13 μL of
10 mM Ru(bpy)32+ and left to air dry for 60 min or longer.
4. Once the Ru(bpy)32+ was dry, the paper microfluidics were
individually cut out and laminated to a screen printed elec-
trode (SPE). Care was taken so that the detection zone of the
paper microfluidic lined up perfectly to cover the working,
auxiliary and reference electrode of the SPE. The top of the
SPE was not laminated to allow electrical connection to the
potentiostat or phone.
5. The SPE and the paper microfluidic laminated together consti-
tuted the sensor; up to 20 sensors could be simultaneously lami-
nated in this way. Each sensor was cut out of the laminate sheet
leaving approximately 1 mm on all sides to ensure a tight seal.

3.1.2 Fabrication 1. The paper microfluidics were cut to shape using a simple hole
of “Cut-and-Seal” Paper punch with a 6 mm diameter from Whatman filter paper (No.
Microfluidic Sensors 4). In this case the laminate itself defined the perimeter of the
fluidic zone rather than a hydrophobic chemical agent.
2. The 6 mm empty discs were loaded with 7 μL of Ru(bpy)32+
and left to air dry for approximately 60 min.
3. Once dry, the discs were placed in a warm oven (≈50 °C) for
approximately 15 min to ensure all moisture was removed.
4. The dry Ru(bpy)32+ loaded discs were placed in GBC A5 Peel
n Stick Pouches which have an adhesive back.
 Mobile Phone Based Electrochemiluminescence Detection… 283

Laminate Paper fluidic

Dummy disc Working

Adhesive back SPE electrode

1 2 3 4
Fig. 3 Steps involved in fabrication of the paper microfluidic sensors. (1) The paper fluidic element (6 mm
Ru(bpy)32+ loaded disc) is laminated into an adhesive backed laminating pouch with a dummy disc under the
paper fluidic. (2) The adhesive back of the laminating pouch is exposed. (3) The dummy disc is removed to
expose the Ru(bpy)32+ loaded disc. (4) The laminated paper microfluidic element is adhered to the screen
printed electrode with the detection zone aligned with the working electrode [9]

5. Each Ru(bpy)32+ loaded disc was placed on top of a blank

(unloaded) disc of the same size inside the Peel n Stick pouch.
The purpose of this “dummy disc” is to ensure a good seal
around the paper fluidic element and intimate contact with the
slightly protruding working electrode surface (Fig. 3).
6. Once the discs were aligned perfectly they were laminated.
7. The discs were cut out of the laminate using scissors in a square
shape allowing approximately 3 mm around all sides. The
adhesive back was then removed exposing the sticky back.
8. Touching the adhesive as little as possible, the dummy discs
were then removed via a scalpel blade as shown in Fig. 3.
9. With the dummy disc cut out, the Ru(bpy)32+ loaded disc was
exposed. The adhesive on the back of the laminate all around
the detection zone was then adhered firmly to the SPE. With
the Ru(bpy)32+ loaded disc firmly contacting the working elec-
trode, the sensor was then ready for use.

3.2 Detection The sensors described above may be interrogated in two ways:
and Analyses Firstly, a potentiostat or other external voltage source may be used
to initiate ECL emission which is then detected and quantified
using the camera of a mobile phone to photograph the lumines-
cence. Secondly, the mobile phone itself can be used to provide the
electrochemical control while the ECL is simultaneously moni-
tored using the camera in video mode.

3.2.1 Mobile Phone 1. To ensure even, consistent pressure was applied to all sensors a
as Detector with External perspex “clamp” was made. Perspex was chosen because it is a
Potentiostatic Control sturdy, transparent plastic material. The clamp consisted of two
pieces of perspex (10 mm thick) held together by four screws
(one in each corner). The sensor was placed in between the
pieces of perspex and the screws were tightened (Fig. 4).
284 Jacqui L. Delaney and Conor F. Hogan

Fig. 4 An example of how a perspex clamp was used. Pressure is applied evenly
over the sensor

2. Once the sensor was in the clamp a small incision was made at
the bottom of the paper microfluidic channel where the ana-
lyte solution would be introduced into the sensor.
3. A drop of the sample was applied to the incision whereupon it
was transported to the detection zone by capillary action
(<10 s).
4. Once the detection zone was fully wetted, the sensor, still in
the perspex clamp was connected to the potentiostat and
aligned under the PMT (for optimization studies) or in front
of the mobile phone camera (for proof-of-concept detection).
Example cyclic voltammograms and ECL profiles produced
using the sensors for optimization work, are shown in Fig. 5.
5. Using DBAE as a model analyte, an LOD of 0.9 μM was
obtained using a conventional light detector (PMT), while the
LOD for NADH was determined to be 72 μM.
6. For mobile phone detection the camera was positioned 10 cm
away from the sensor in a darkened room.
7. The settings on the mobile phone were changed to “fire-
works” and a timer was started for 10 s. When the timer hit
the 3 s mark the potentiostat was started to initiate the ECL
reaction.
8. The potential step used by the potentiostat was 0 to 1.25 V for
2 s. The bright red emission was then captured in a photo-
graph from the mobile phone. (See Fig. 6b).
 Mobile Phone Based Electrochemiluminescence Detection… 285

Fig. 5 Typical sensor responses for (a) 5 mM DBAE and (b) 5 mM NADH. The
voltametric (current) responses are dashed blue, and the ECL emission profiles
are in red. Both experiments were run at a scan rate of 0.2 V/s and were versus
a Ag/AgCl reference electrode. (From ref. 8)

9. The picture was then analyzed via a simple program scripted in

Python which analyzed the red pixel intensity for each picture
and gave a value that represented the intensity of the ECL emis-
sion. There was a strong relationship between analyte concen-
tration and pixel intensity as shown by the linear trend (Fig. 6a).
10. A calibration curve (Fig. 6a) was obtained using the model
analyte DBAE. An LOD of 250 μM was obtained.
286 Jacqui L. Delaney and Conor F. Hogan

Fig. 6 (a) Calibration curve between 0.5 and 20 mM for DBAE using paper microfluidic ECL sensor with mobile
camera phone as the detector. The magnitude of the ECL signal is proportional to the intensity of the red pixels
in the digital image. (b) Digital photographic images of ECL emission from the paper fluidic sensor obtained for
various concentrations of DBAE using camera phone [8]

3.2.2 Both Potentiostatic 1. In order to test the voltage obtainable from the mobile phone
Control and Detection (in this case a Samsung Galaxy S (i9000)), a lead connected to
Performed by Mobile a stereo 3.5 mm audio jack on the one end and two alligator
Phone clips (that were once ear phones) on the other was plugged in
 Mobile Phone Based Electrochemiluminescence Detection… 287

Fig. 7 Comparison of ECL responses produced using voltage excitation signal

produced by phone (upper graph) and a conventional potentiostat (lower graph).
The detector in both cases was a photomultiplier tube and the test solution con-
tained 0.5 mM Ru(bpy)32+ and 0.1 mM DBAE [9]

to the mobile phone audio socket. With the Voodoo sound

application running, the phone was capable of producing a
potential of ±1.8 V with maximum audio settings (Fig. 7).
2. Because there was no reference electrode the potential was
subject to change and drift. To solve this issue the counter-
electrode of the SPE was coated with Ag/AgCl. This was done
by electroplating in a 10 mM AgNO3 solution for 30 s at
−1.5 V vs. Ag/AgCl followed by a 3 M KCl for 10 s at 1.0 V
vs. Ag/AgCl. This stabilized the potential so that ECL could
be observed at a lower apparent applied potential. Au plating
the counter-electrode is an alternative to Ag/AgCl plating;
however, Ag/AgCl provided more consistent results.
288 Jacqui L. Delaney and Conor F. Hogan

Fig. 8 Arrangement used for mobile phone based ECL sensing. The audio jack
supplies the potential to the paper microfluidic sensor, while the resultant emis-
sion is detected by the camera. Both the excitation and detection processes are
controlled by a software application. The black plastic holder positions the sen-
sor adjacent to the camera and blocks ambient light. (From ref. 9)

3. The sensor was placed in the sensor holder (shown in Fig. 1)

and attached to the phone. The holder positioned the SPE
consistently in front of the camera and excluded ambient light.
4. The audio plug was then inserted into the phone and the
alligator clips were connected to the sensor via the working
and counter-electrode connectors (Fig. 8).
5. An android software application was used to initiate the ECL
reaction by playing a square waveform while simultaneously
capturing a video of the luminescence via the camera. The
application then analyzed the red pixel intensity of each frame
(for more information on the application please refer to ref. 9).
As before, the red pixel intensity was proportional to the
concentration of analyte (DBAE or L-proline). There are many
options for analysing pixel intensity from the frames, including
freeware programs such as ImageJ [12].
6. Calibration curves for DBAE and L-proline were obtained
using the phone to initiate, detect, and analyze the ECL reac-
tion. For DBAE an LOQ of 100 μM was obtained and for
 Mobile Phone Based Electrochemiluminescence Detection… 289

L-proline an LOQ of 100 μM was also obtained. The mobile

phone analysis can be run more than once on each sensor for
replicate results. However, here they were used as single-use
sensors.

References

1. http://data.worldbank.org. Accessed 28 May 2013 for selective excitation in mixed electrochemi-

2. Martinez AW, Phillips ST, Carrilho E, Thomas luminescent systems. Chem Sci 4:977–982
SW III, Sindi H, Whitesides GM (2008) 7. Barbante GJ, Hogan CF, Mechler A, Hughes
Simple telemedicine for developing regions: AB (2010) Electrochemiluminescence of
camera phones and paper-based microfluidic surface bound microparticles of ruthenium
devices for real time, off-site diagnosis. Anal complexes. J Mater Chem 20:891–899
Chem 80:3699–3707 8. Delaney JL, Hogan CF, Tian J, Shen W (2011)
3. Dungchai W, Chailapakul O, Henry CS (2009) Electrogenerated chemiluminescence detec-
Electrochemical detection for paper-based tion in paper-based microfluidic sensors. Anal
microfluidics. Anal Chem 81:5821–5826 Chem 83:1300–1306
4. Barbante GJ, Hogan CF, Wilson DJD, 9. Delaney JL, Doeven EH, Harsant AJ, Hogan
Lewcenko NA, Pfeffer FM, Barnett NW, CF (2013) Use of a mobile phone for poten-
Francis PF (2011) Simultaneous control of tiostatic control with low cost paper-based
spectroscopic and electrochemical properties in microfluidic sensors. Anal Chim Acta
functionalised electrochemiluminescent tris 790:56–60
(2,2′-bipyridine) ruthenium (ii) complexes. 10. Li X, Tian J, Garnier G, Shen W (2010)
Analyst 136:1329–1338 Fabrication of paper-based microfluidic sensors
5. Piper DJE, Barbante GJ, Brack N, Pigram PJ, by printing. Colloids Surf B Biointerfaces
Hogan CF (2011) Highly stable ECL active 76:564–570
films formed by the electrografting of a diazo- 11. Fenton EM, Mascarenas MR, Lopez GP,
tized ruthenium complex generated in situ Sibbett SS (2009) Multiplex lateral-flow test
from the amine. Langmuir 27:474–480 strips fabricated by two-dimensional shaping.
6. Doeven EH, Zammit EM, Barbante GJ, ACS Appl Mater Interfaces 1:124–129
Francis PF, Barnett NW, Hogan CF (2013) A 12. http://rsbweb.nih.gov/ij/index.html .
potential-controlled switch on/off mechanism Accessed 13 Feb 2014
 Part II

mHeath Technologies for Physiological

and Anatomical Measurements
 Chapter 20

iStethoscope: A Demonstration of the Use of Mobile

Devices for Auscultation
Peter J. Bentley

Abstract
iStethoscope Pro is the first piece of software (an “App”) produced for iOS devices, which enabled users
to exploit their smartphones, music players, or tablets as stethoscopes. The software exploits the built-in
microphone (and supports externally added microphones) and performs real-time amplification and filter-
ing to enable heart sounds to be heard with high fidelity. The software also enables the heart sounds to be
recorded, analyzed using a spectrogram, and to be transmitted to others via e-mail. This chapter describes
the motivation, functionality, and results from this work.

Key words Stethoscope, Digital auscultation, Audio processing, iPhone app, iStethoscope Pro

1 Introduction

According to the World Health Organization, cardiovascular dis-

eases (CVDs) are the number one cause of death globally: more
people die annually from CVDs than from any other cause. An
estimated 17.3 million people died from CVDs in 2008, represent-
ing 30 % of all global deaths. Of these deaths, an estimated 7.2
million were due to coronary heart disease [1, 2]. It has been pre-
dicted that the number of people who will die from CVDs will
increase to reach 23.3 million by 2030 [1, 3], with CVDs remain-
ing the single leading cause of death [3]. Any method which can
help to detect signs of heart disease could therefore have a signifi-
cant impact on world health.
This article describes the creation of a novel method designed
to help address these significant issues. We provide a demonstra-
tion of the use of mobile computing devices such as smartphones
for auscultation, using a new piece of software developed by the
author: iStethoscope Pro. The method illustrates of how today’s
technology is capable of enabling the general public to gather their
own heart sounds using readily available technology and
communicate that data electronically.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_20, © Springer Science+Business Media New York 2015

293
294 Peter J. Bentley

The concept of exploiting everyday consumer electronics for

auscultation became feasible in 2009 when it was discovered
through experimentation that the latest smartphones had remark-
ably high quality audio capabilities. This was particularly evident
in Apple devices, for this company had previously focused on
consumer music players (“iPod” MP3 devices) that required
excellent digital to analog conversion. This high-quality audio
capability was inherited by the new devices when the company
expanded into mobile phones, and the addition of excellent qual-
ity microphones and associated analog to digital hardware,
resulted in a cell phone that could both sample audio at remark-
able quality, and play it back. The powerful processors within the
devices made real-time audio signal processing feasible, and the
high connectivity of the devices (Bluetooth, cellular data, and
Wi-Fi) made it possible for uncompressed high-quality audio files
to be transmitted from the mobile device to any other connected
computer in the world. Finally, the App Store was created and
Application Programming Interface (API) libraries were released,
opening the devices to software engineers and developers world-
wide, and enabling the rich resources of their hardware to be
exploited for new purposes.
The confluence of these technologies and needs resulted in the
development of a piece of software (or “App”) designed to enable
these smartphones to behave in a similar fashion to digital stetho-
scopes. The method is software-based: once the software is run-
ning on a suitable device and the microphone of the device is
placed appropriately, audio is sampled. The software samples the
audio, filters and records in real time, enabling live monitoring or
playback, visualization via spectrogram and transmission of the
audio. The software is designed to be configurable to enable users
adjust amplification, filtering and other settings to suit their devices
and body types. There is also an option of e-mailing the audio to
an automatically anonymized collection server, enabling the
“crowd sourcing” of audio from devices around the world to form
a large database of heart audio, suitable for subsequent research.
The software was initially released as a simple demonstration of
the technology with no ability to transmit the audio (there was
initially no API to allow such functionality). Free to download, it
was one of the first “Medical Apps” for the iPhone and received
large numbers of downloads. Cardiologists and other heart special-
ists soon trialed this new technology and made contact with the
author asking for new features, resulting in new versions of the
software which recorded and enabled the visualization and trans-
mission of the audio. A later modification also added the ability to
crowd-source heart audio, storing the anonymized data on a server
of the collaborating cardiologist (Prof Glenn Nordehn), stored
with ethical permission.
 iStethoscope: A Demonstration of the Use of Mobile Devices for Auscultation 295

2 Materials

The iStethoscope software is designed to run on Apple iOS devices

that have built-in microphones or have the capability to accept
plug in microphone peripherals. To date this enables iPhones,
iPods and iPads to be used with the software, although it is
designed specifically for iPhones. The software was originally opti-
mized for 3G devices, but has been shown to work effectively on
4, 4S, 5, 5C, and 5S iPhones.
The software is made available through the online distribution
platform known as the App Store, which hosts the package under
the “Medical” category and enables users of the compatible devices
to connect via the Internet, purchase and download the software.
The lowest non-zero price possible was chosen in order to make
the software widely accessible while enabling future updates to be
supported by income generated (1USD or 0.69GBP).
The software is designed to make use of the internal micro-
phone of iPhones where possible. For best quality audio playback,
good quality headphones with frequency response 20 Hz to 20 Kz
enables the lower frequencies of the heartbeat to be heard effec-
tively. Most devices are not sold with such headphones, so it is
necessary for users to procure this additional equipment.
It is common for users of portable devices to keep them in
protective covers, sleeves or cases during everyday use. The iStetho-
scope software requires the microphone to be placed directly
against the skin and thus all such covers must be removed from the
devices.
Optionally, the software may be used in conjunction with other
physical and software-based equipment. Palm et al. [4] describe
the development of a protocol with the iStethoscope software
installed on a device and used in conjunction with a traditional
stethoscope. To enable more detailed analysis of the sampled
audio, the audio data may be e-mailed from the device and
imported into audio processing software (such as Audacity [5])
which enables the plotting of detailed spectrograms in combina-
tion with filtering.
Machine learning classifiers have been created to segment and
classify the audio produced by the app [6, 7]; however, to date
these are still under development and are not sufficiently advanced
to be incorporated into the app.

3 Methods

3.1 Launching The iStethoscope Pro app should first be located on the App Store,
the App downloaded to an appropriate device, and launched. On first
launching the user will be prompted to activate the app by going to
296 Peter J. Bentley

Settings and agreeing to the terms of use. These terms make the
user aware that the device may interfere with implanted medical
equipment such as pacemakers and defibrillators and that it should
not be used near people with such implanted equipment. Once acti-
vated, the app can be launched and will run normally from then on.
On launching, the app shows an initial “splash page” which
contains a link to the help system. After 3 seconds the main screen
is shown, see Fig. 1. The user may return to the splash page by
pressing “i” at the bottom of the main page at any time.

3.2 Auscultation Figure 2 provides an overview of the algorithm of iStethoscope

with the App Pro. To begin auscultation with the app, the user selects a listen-
ing mode by pressing the “mode” button. Each mode corre-
sponds to a different set of user-defined amplification and
filtering settings, which include one or more of the following
procedures in combination:
Low pass filter. Audio is filtered using Eq. 1. The low pass filter
removes frequencies higher than the cutoff frequency, in order to
remove background noise while preserving the low-frequency
heart sounds.
x ( t ) - y ( t - 1)
y ( t ) = y ( t - 1) + (1)
a

Fig. 1 Example screenshots for iStethoscope Pro. (a) The main screen showing the mode button with text indicating
the current audio mode, the stethoscope image, the gain control, and the “i” information button. (b) The playback
screen showing a spectrogram of the sound. (c) Some of the user controllable Settings for the program
 iStethoscope: A Demonstration of the Use of Mobile Devices for Auscultation 297

show splash page containing help

button; after 3 second pause show
main screen, mode: muted

has app been give instructions on activation

no
activated? and prevent further use of app

yes

show main screen with stethoscope image;

If mode button is pressed, cycle to next mode;
perform audio calculation depending on mode:

Mode: Heartbeat Mode:

Mode: Heartbeat Mode:
Mode: Mute pure Mode: Clear Sound Accelerometer
filtered Conversation
no audio is sampled audio is sampled and audio is sampled and input from
audio is sampled and audio is sampled and
or played back filtered using low filtered using low accelerometers is
filtered using low filtered using low
pass filter and pass filter, high pass transformed into a
pass filter, high pass pass filter, high pass
amplification, AVC is filter, and sinewave with
filter, and filter, and
disabled amplification, AVC is frequency
amplification, AVC is amplification, AVC is enabled proportionate to
enabled enabled scaled G forces

add new audio to 8-second audio queue (when full,

begin writing at the start of queue)

has device been shaken, or no

stethoscope image pressed?

yes

take first 7 seconds of audio queue, calculate fft; switch

to playback screen; display spectrogram and audio

playback audio with moving bar, loop to beginning at end

has device been shaken, or

yes
back button pressed?

no

has email button been

no pressed?

yes

convert audio queue into AIF format file; convert spectrogram into
bitmap image; create blank email with both files as attachments;
enable user to modify, send or cancel email

Fig. 2 Flow diagram showing internal processing by iStethoscope Pro app

298 Peter J. Bentley

where: y(t) is the filtered audio value at time t

x(t) is the unfiltered input audio value at time t
a is a constant value, default 125 (determines the cutoff
frequency)
High pass filter. Audio is filtered using Eq. 2 (always in combi-
nation with Eq. 1). The high pass filter removes frequencies lower
than the cutoff frequency, in order to remove background rumble
while preserving higher frequency sounds such as breathing.
x ( t ) - x ( t - 1)
y ( t ) = y ( t - 1) + (2)
b

where: y(t) is the filtered audio value at time t

x(t) is the unfiltered input audio value at time t
b is a constant value, default 1.0001 (determines the cutoff
frequency)
Amplification. The filtered audio is amplified by increasing the
amplitude of the audio signal according to the setting of the on-
screen slider control.
Volume. The filtered and amplified audio is output through the
audio output device (headphones or built-in speaker) at the vol-
ume defined by the device’s current volume level.
Auto volume control (AVC). If the volume of the audio output exceeds
a user-defined value, the AVC will dampen the volume to safe levels.
When the audio output is no longer too loud, the AVC will gradually
increase the volume until it reaches normal levels. This prevents the
user of the app from being deafened by excessive noise caused by
knocking or scraping the microphone during positioning.
Ignore last second. If active, then on playback, only the last 7
out of 8 s of audio are played. This is because there is usually a
short burst of white noise as the device is removed from the skin,
caused by friction.
Listening Modes. Five listening modes are provided:
1. Heartbeat pure—only the low pass filter is provided. Studies
have shown that most of the key data needed for clinical diag-
nosis is contained in frequencies lower than 400 Hz [4] so this
mode enables the user to discard higher frequencies (the exact
cutoff frequency may be defined by the user in the Settings).
Many higher frequencies also correspond to unwanted back-
ground noise so it is advantageous to filter them out. AVC is
disabled as low frequencies rarely become too loud for the user.
2. Heartbeat filtered—for some models of phones the microphone
was found to be inferior and so it was useful to enable the user
to sample audio in at slightly higher frequency band. To achieve
this the low pass filter is used in combination with a high pass
filter. AVC is enabled to avoid excessive loudness for the user
caused by the microphone brushing against skin or clothing.
iStethoscope: A Demonstration of the Use of Mobile Devices for Auscultation 299

3. Conversation—some users with hearing loss requested a listen-

ing mode suitable for everyday use to help them hear voice
conversations. This mode uses low and high pass filters to
remove noise and focus only on frequencies needed to under-
stand voices, around the 300–3,400 Hz plain old telephone
service (POTS) range.
4. Clear Sound—both low and high pass filters are used, but the
cutoff frequencies are much lower and higher respectively
enabling much more of the original sound to be heard com-
pared to the other modes. With a high quality microphone,
this mode can enable both good heart sounds and breathing
sounds to be heard. AVC is enabled.
5. Accelerometer—the accelerometers of the phone are transformed
into a sine wave with frequency corresponding to the motion of
the phone, where larger motion causes a higher pitched motion.
This setting was an attempt to see if the vibration of a chest
could be sufficient to cause measurable motion by the phone;
while this may be possible in the future it was found that other
normal motions of the body (and even vibrations of traffic
nearby) made this setting of limited use for auscultation.
With the appropriate mode chosen, the user must place the
microphone of the device at the desired auscultation point on the
body, see Fig. 3. All filtering, amplification and volume settings can
be modified until audio of desired quality is obtained. The app
performs all audio manipulation in real time so that the user can
listen as they place the device as with a traditional stethoscope.

Fig. 3 Optimal positions on the body for placing the microphone

300 Peter J. Bentley

Fig. 4 Spectrogram image of audio produced by iStethoscope Pro. (The first S1

and S2 has been annotated for this article)

When the user is content with the audio quality they can quickly
remove the device from the body and shake it sharply, or touch the
stethoscope image. This initiates playback mode.

3.3 Audio Playback In playback mode a spectrogram of the audio is shown and the last
7 or 8 s is played continuously. The spectrogram plots the audio
using a Fast Fourier Transform [8], see Fig. 4. The x-axis is time,
the y-axis is frequency, brighter colors indicate volume. Although
the S1 and S2 components are visually easy to pick out because of
the higher frequencies, most of the key information is found in the
lower frequencies, at the bottom of the image.
When in this mode the user may press the mail symbol. This
triggers the creation of an audio file containing the audio being
played back, and a bitmap image of the spectrogram. The two files
are attached to a blank e-mail and the user is then able to send the
data to another computer (or save the data in their own e-mail out-
box for later). Figure 5 illustrates an example of normal and abnor-
mal heart sounds as plotted by Audacity on a desktop computer.

3.4 Crowd In the Settings for the app, the user may choose to activate a “med-
Sourcing Data ical trial” setting. This overrides all other settings, and forces spe-
cific filtering, amplification and AVC settings that were chosen
through experimentation by Nordehn’s group [4]. After ausculta-
tion, if the user chooses to send an e-mail containing their heart
sounds, the resulting mail is automatically addressed to a computer
at the University of Minnesota Duluth, which anonymizes the data
and stores the audio in an archive for later processing.
 iStethoscope: A Demonstration of the Use of Mobile Devices for Auscultation 301

Fig. 5 Data analysis spectrogram plots by Audacity software (sounds have been gathered using a low-pass
filter at 400 Hz). Left: normal heart sound. Right: Aortic regurgitation. The difference in the sounds is immedi-
ately clear to a clinical expert accustomed to reading such spectrogram displays

Most recently, a subset of the heart sounds from the database

has been released for researchers. A machine learning challenge was
created for computer science researchers [9]. The focus was the
creation of the first level of screening of cardiac pathologies at home
by the patient (using a mobile device). The problem was of particu-
lar interest to machine learning researchers as it involved classifica-
tion of audio sample data, where distinguishing between classes of
interest was nontrivial. Data was gathered in real-world situations
and frequently contained background noise of every conceivable
type. The differences between heart sounds corresponding to dif-
ferent heart symptoms could also be extremely subtle and challeng-
ing to separate. Success in classifying this form of data required
extremely robust classifiers. Despite its medical significance, to date
this was a relatively unexplored application for machine learning.
The challenge was split into two parts. The first challenge was to
produce a method that could locate S1(lub) and S2(dub) sounds
within audio data, segmenting the Normal audio files in both data-
sets. S1 corresponds to the noise of the atrioventricular valves of the
heart; S2 corresponds to the semilunar valves. The identification of
these key sounds is of great clinical importance as they enable the
subsequent identification of additional anomalous heart sounds
(e.g., lub-lub dub or lub dub-dub) or murmurs where the position
of the abnormal sound relative to the lub or dub sound has signifi-
cant diagnostic value. The second challenge was Heart Sound
Classification: to produce a method that could classify real heart
audio (also known as “beat classification”) into one of four catego-
ries: Normal, Murmur, Extra Heart Sound, Artifact.
Software that can successfully perform triage and determine
the likelihood of specific kinds of heart disease from heart sounds
could have a significant impact on world health. The result of this
challenge was several novel signal processing and machine learning
methods that were able to segment and classify the real heart audio
with reasonable accuracy [6, 7].
302 Peter J. Bentley

4 Notes

4.1 Limitations iStethoscope Pro is not intended to be used as a medical device.

and Future Work The app is a demonstration of the technology and is intended to
improve awareness of health issues relating to heart disease in the
general public worldwide. As such, while physician users have had
considerable success using the app for auscultation, people
untrained in auscultation techniques often find it challenging to
place the microphone at the appropriate locations on the body to
enable a satisfactory audio signal to be generated.
These issues were addressed by the creation of a help system
for the app [12]. Available by pressing the “help” button on the
splash screen, the app quits and launches a Web browser which
shows a complete online help system, including several videos
that demonstrate all aspects of the use of the app, including
microphone placement on the body. With this help system in
place, members of the general public are now able to train them-
selves to perform auscultation and feedback from successful
users has been very positive.
There are several measures that can be taken to improve qual-
ity and consistency of audio gathered by the app. Other micro-
phones can be used to make auscultation easier. The device can be
used in conjunction with a traditional stethoscope, or with other
instruments designed to conduct sound from the body to the
built-in microphone of the device.
Research continues on heart sound classification, using the
data gathered by the app.

4.2 Assessment The main aim of this work was to demonstrate the use of mobile
of Success phones for auscultation and by doing so, increase awareness of
healthy hearts worldwide. To measure success it is necessary to
examine the uptake and use of the app. So far the reception of the
app has been remarkable, receiving substantial attention world-
wide. Two UK broadsheet articles (Independent and Guardian
[10, 11]) triggered significant media attention at the beginning of
September 2010. The author was then interviewed for over 20
radio shows, and many television programs and news channels
worldwide. The story was picked up by most national papers in the
UK and worldwide, in addition to magazines, leaflets and online
media. In 1 week the iPhone app had been downloaded a further
200,000 times. Out of the 250,000 programs available for these
devices, this became the number one bestselling app in the App
Store. It is hard to estimate the number of people worldwide that
have been made aware of this work, and through it, highlighted
the importance of healthy hearts, but given the extensive world-
wide television, radio, and print coverage, the number is likely to
be several hundred million.
 iStethoscope: A Demonstration of the Use of Mobile Devices for Auscultation 303

The story has been highly positive for all those with an interest
in technology and health. Feedback includes comments from doc-
tors who say the use of the program in training will save lives.
Others suggest that now heart sounds can be captured from
patients in remote regions of the world and e-mailed to the experts
in the cities. The publicity has also led to many new possibilities for
further funding, trials of the technology remote areas of the world
including India, China, and Peru, and licensing of the technology
by medical companies. One user described how the app helped to
save the life of her unborn child as it suffered complications.
The iPhone application has been a huge success, with two to
three million downloads from over 100 countries, and a resulting
large database of useful heart sounds. The response from the medi-
cal profession and public has been overwhelmingly positive; indeed
an article has been published independently by cardiologists in
USA, detailing the use of the app for heart sound collection [4].

References
1. WHO (2011) Global status report on non- PASCAL heart challenge workshop, AISTATS
communicable diseases 2010. World Health 2012, La Palma, 25 Mar 2012
Organization, Geneva 7. Gomes EF, Bentley PJ, Coimbra M, Pereira E,
2. WHO (2011) Global atlas on cardiovascular Deng Y (2013) Classifying heart sounds:
disease prevention and control. World Health approaches and results for the PASCAL
Organization, Geneva Challenge. In Proc. 6th International
3. Mathers CD, Loncar D (2006) Projections of Conference on Health Informatics, HealthInf
global mortality and burden of disease from 2013, Barcelona, Spain, Feb 2013
2002 to 2030. PLoS Med 3(11):e442 8. Bracewell RN (1986) The Fourier transform
4. Palm D, Burns S, Pasupathy T et al (2010) and its applications, vol 31999. McGraw-Hill,
Artificial neural network analysis of heart New York
sounds captured from an acoustic stethoscope 9. Bentley P, Nordehn G, Coimbra M, Mannor S
and emailed using iStethoscopePro. J Med (2011) The PASCAL classifying heart sounds
Device 4(2):027531–027531-1. challenge. http://www.peterjbentley.com/
doi:10.1115/1.3443737 heartchallenge/index.html
5. Audacity. Available under the terms of the 10. Laurance J (2009) 59p stethoscope on iphone
GNU General Public License (GPL) as pub- proves a hit. Article in Independent Newspaper,
lished by the Free Software Foundation: 14 Dec 2009. The Independent (London)
http://audacity.sourceforge.net/ 11. Hill A (2010) iPhone set to replace the stetho-
6. Deng Y, Bentley PJ (2012) A robust heart scope. The Guardian (London), Monday 30
sound segmentation and classification algo- Aug 2010.
rithm using wavelet decomposition and 12. Bentley P. iStethoscope Prop Help. http://
spectrogram. Extended abstract in the first www.peterjbentley.com/istethoscopepro.html
 Chapter 21

iPhysioMeter: A Smartphone Photoplethysmograph

for Measuring Various Physiological Indices
Kenta Matsumura, Peter Rolfe, and Takehiro Yamakoshi

Abstract
iPhysioMeter is a new smartphone application (“App”) for the Apple iPhone and iPod touch that allows
photoplethysmography (PPG) to be implemented without the need for any additional devices. The resulting
signal, the photoplethysmogram, allows the calculation of basic but valuable and frequently used physiological
indices such as heart rate (HR) and pulse volume (PV). The design of iPhysioMeter has very much been
influenced by a consideration of usability, as is immediately evident from ones first experience with it. However,
its apparent simplicity in use should not disguise the need for correct operation, which otherwise might lead
to collection of invalid or inaccurate data. There are several unexpected pitfalls that might not only produce
inaccurate values, but, under some circumstances, could also damage the device or present a hazard to the user
or subject. We therefore describe here, firstly, the core technology that makes it possible to perform PPG and
to calculate HR and normalized PV (NPV) from the photoplethysmogram using only a smartphone, secondly,
the correct and optimum methods and procedures for using iPhysioMeter that will help to ensure safety and
the derivation of valid data under real operational conditions. We hope that these descriptions will help facili-
tate any activities related to physiological measurement when using only a smartphone.

Key words Smartphone, Physiological indices, Photoplethysmography, Heart rate, Pulse volume,
iPhysioMeter, CMOS Camera

1 Introduction

Physiological variables associated with the rhythmical contraction

and pumping action of the heart, such as heart rate (HR), the
number of beats per minute, and pulse volume (PV), arterial blood
volume changes caused by cardiac contractions, have been used for
many decades in a wide spectrum of research and routine clinical
care [1, 2]. For example, in the fields of psychology, physiology,
and biomedical engineering, HR and PV are used as indices of
emotional stress and/or arousal [3–9] because these indices dif-
ferentially reflect autonomic nerve activity [10]. In the field of
clinical care these indices are used for diagnosis [11], in the assess-
ment of cardiovascular health [12, 13], or for monitoring the
effectiveness of treatment [14–17].

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_21, © Springer Science+Business Media New York 2015

305
306 Kenta Matsumura et al.

When measuring HR and PV, the optical method of photople-

thysmography (PPG) is very useful because both of these variables
can be measured using only one device. Key components of PPG
instrumentation are a light source, which could be a laser or a light
emitting diode (LED), and a photo detector, such as photo-diode
(PD). The source and detector are attached to the subject, light
being directed into tissue, such as a finger or an earlobe, and this
light then emerging from the tissue being collected by the detec-
tor. As light passes through the tissue it is attenuated by absorption
and scattering phenomena and such attenuation is particularly
influenced by soft and hard tissues and, importantly, by blood.
Cardiac contractions produce cyclical variations in blood volume,
which in turn produce corresponding variations in the intensity of
the light emerging from the tissue. During the systolic phase of the
cardiac cycle blood volume in the interrogated tissue is at a maxi-
mum and so the detected light intensity is at a minimum. After this
systolic peak blood volume then falls to the end diastolic level
where light is minimally attenuated and so the detected light is at
a maximum. Thus, HR can be calculated by counting the peaks or
troughs in the PPG signal, whereas PV can be calculated from the
magnitude of the light intensity change (peak to trough) over each
cardiac cycle. The detected light is considered to consist of two
components. Firstly, there is a relatively constant light level related
to absorption by the non-blood containing tissues and this is
termed the DC, or direct current, component. Then there is a
cyclically varying light level produced by the cardiac-related beat-
by-beat blood volume changes in the tissue vasculature, termed
the AC, or alternating current, component. In a PPG instrument
there are filtering means able to separate the AC and DC compo-
nents from the PPG signal.
Although such measurement with PPG has hitherto required
dedicated photoplethysmograph devices, a new instrumentation
option has recently emerged in the form of the smartphone [18, 19].
PPG has been shown to be possible using a smartphone without
any additional devices (Fig. 1a). Here, the photoplethysmogram,
which comprises of the DC component and the very small AC
component that is related to peripheral pulse volume, is measured
by the combination of the smartphone embedded pseudo-white
color LED flash as a light source and the complementary metal
oxide semiconductor (CMOS) camera as a photo detector or sen-
sor (Fig. 1b). The changes in light intensity in each red, green, and
blue light color are recorded by the CMOS camera, respectively
(Fig. 1c).
Inspired by these pioneering ideas, Matsumura and Yamakoshi
developed a new smartphone application (“App”) called
iPhysioMeter that can measure HR and normalized PV (NPV; an
index of α-adrenalin-mediated sympathetic activity [20]) to pro-
vide a broad field of researchers with readily available and easy-to-
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 307

Fig. 1 (a) An illustration of photoplethysmography performed using only a smartphone. (b) Schematic drawing
illustrating a section of the measuring site. The white light emitted by the LED flash propagates into the finger,
being scattered and absorbed by blood and other hard and soft tissues, and finally reaches the camera. The
light intensity captured by the camera varies in accordance with arterial blood volume pulsations; that is, pulse
wave. The longer wavelength has deeper penetration into the tissue, thereby probing deeper regions. (c) An
example of an actual image captured by the CMOS camera from the finger. The light intensity change involved
in arterial pulsations could be measured using any of the three light colors, that is red (R), green (G), or blue
(B), available from the CMOS camera. The color discrimination is achieved by the camera

use smartphone PPG measurement tool (Fig. 2). The main features
of iPhysioMeter are: (1) It shows good agreement with results of
HR and NPV from dedicated laboratory devices [21, 22]; (2) It is
available worldwide for free at iTunes App Store by Apple, Inc.; (3)
It allows green light measurement, which has consistently shown
its superiority over red and blue PPG in terms of the relative free-
dom from motion artifacts and high signal/noise ratio [22–25];
(4) It is designed to measure not only the time-related of measure
of HR (to be precise, pulse rate) but also the amplitude related
index NPV (to be precise, cutaneous PV) (see Note 1); (5) It is
equipped with an on-line beat-by-beat HR and NPV auto analysis
function with automatic outlier detection; (6) It is provided with a
data management function that enables users to display, name, and
delete a set of beat-by-beat data, and to send these data in a comma-
separated value (CSV) format via e-mail; (7) It is designed to mea-
sure three-axis g-forces, which is useful for judging whether there
308 Kenta Matsumura et al.

Fig. 2 An example of a measurement procedure in progress, with both clean pulsations and motion artifacts in
the “Measure” window. (A) Measurement meter, with colored segments ranging between 0 (empty) and 13
(full). 0 (empty) = measurement off, 1 = preparing for measurement (during camera calibration or when finger
is not ready), 2–12 = measurement in progress, 13 (full) = completion of the one-shot measurement or begin-
ning of continuous measurement. (B) The mean heart rate (HR) over the preceding (approximately) 10 s during
measurement, the mean HR over the whole period when the measurement is finished; beats per minute (bpm).
(C) The mean logarithmically transformed (ln) normalized pulse volume (NPV) over the preceding (approxi-
mately) 10 s, the mean (ln) NPV over the whole period when the measurement is finished; arbitrary unit (a.u.).
(D) Measured pulse wave. (E) Real-time color image of the finger-tip captured by the camera. (F) The result of
real-time analysis of the pulse wave. Red (the second brightest in gray scale) circle = peak, Blue (the darkest
in gray scale) circle = bottom, Yellow (the brightest in gray scale) circle (peak) = outlier of HR, Yellow circle (bot-
tom) = outlier of NPV. (G) Elapsed time; seconds (s) and sampling rate; frames per second (fps). (H) Start/Stop
button. Quoted from ref. 29 with some modification

was motion artifact, as well as the PPG; (8) It is controllable via a

play/stop button with a wired or wireless headphone.
These features allow iPhysioMeter, when used with the iPhone
and iPod touch from Apple Inc., to derive accurate HR and NPV
measurements easily and reliably. However, the simplicity of use
does not eliminate the need for care in applying the method since,
under some circumstances, inappropriate operation could possibly
damage the device or even present a hazard to the user or subject.
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 309

We therefore describe here, firstly, the core technology that makes

it possible to make PPG measurements and to calculate HR and
NPV only using a smartphone and, secondly, the correct and opti-
mum methods and procedures for using iPhysioMeter that will
help to ensure safety and the derivation of valid data under real
operational conditions.

2 Materials

1. A smartphone or a music player from Apple Inc. with LED

flash (see Note 2).
2. The iPhysioMeter App, which is available at iTunes App Store
(Apple Inc.). Please search using the word “iPhysioMeter” in
App Store (see Note 3).
3. When developing a smartphone app, Xcode app, the develop-
ment environment, from Apple Inc. and a Mac computer on
which Xcode runs are also needed (see Note 4).

3 Methods

3.1 Measuring PPG 1. Red, green, and blue PPG signals can be measured simultane-
from the ously from the CMOS camera (image sensor). In brief, the
CMOS Camera CMOS camera captures each light by using a set of very small
red-, green-, and blue-optical filters located with each pixel.
2. At the beginning of the measuring sequence, the CMOS cam-
era is set to video capture mode with 8 bits depth for red,
green, and blue light, with 192 × 144 pixels resolution.
3. The sampling frequency is set to 30 frames per second (fps)
(see Note 5).
4. The white balance, focus, and exposure (sensitivity to light)
mode of the CMOS camera are set to “Locked,” “Locked,”
and “ContinuousAutoExposure,” respectively [26].
5. Once actual measurement sequence has started after the LED
flash set to “on,” an image captured by the CMOS camera is
displayed on the measurement screen of iPhysioMeter
(Fig. 2E). Although an image captured by the CMOS camera
is square (Fig. 1c), here it is clipped in a circle and displayed.
All captured images are averaged among all of the 192 × 144 pix-
els to produce one red, green, and blue value per each frame.
6. During measurement, it is most important is to set the light sen-
sitivity of the CMOS camera to an appropriate value that enables
it to measure PPG stably. If sensitivity is too high or too low, it
makes it impossible to make PPG measurements. The image
shown in Fig. 1c was captured with an appropriate sensitivity.
310 Kenta Matsumura et al.

However, the “ContinuousAutoExposure” setting of the exposure

mode does not provide the best light sensitivity at all times.
Nevertheless, all that iOS can do is to adjust the sensitivity of the
CMOS camera by setting the exposure mode of the camera to
either “ContinuousAutoExposure” or “ExposureModeLocked”
at its most extreme (see Note 6). This means that it is virtually
impossible to set the light sensitivity of the CMOS camera freely.
As a last resort, we implemented a somewhat forceful but effec-
tive method to change “ContinuousAutoExposure” setting into
“ExposureModeLocked” setting just at the right moment at the
very first stage of auto adjustment of the sensitivity. This method
is conducted when “Exposure Adj” (adjustment) in “Settings” is
set to “Fixed” (Fig. 3b) and is suitable for all iPhone models. If
“Exposure Adj” is set to “iOS Default,” the exposure mode of
the CMOS camera remains set to the “ContinuousAutoExposure”
during the whole period of the measurement. This setting is suit-
able only for iPhone 4S, 4s and iPod touch.

3.2 Detecting Red and blue signals derived at each frame are used to judge whether
a Finger the finger is on the camera or not. The actual measurement of PPG
will be conducted only when iPhysioMeter detects the finger on the
camera. iPhysioMeter deems the finger to be on the camera if the
red signal is over 180 (minimum = 0 and maximum = 255, due to
each color being 8-bit depth) and the blue signal is below 180.
A pair of these criteria varies depending on the selections of “Finger
Detection” in “Settings” (Fig. 3b). The criteria for “Strict,”
“Moderate,” “Loose,” and “N/A” are 200, 200, 180, and −1 (thus
always “on”) for red and 50, 120, 180, and 256 (thus always “on”)
for blue, respectively. When “N/A” is chosen, the analysis will be
conducted regardless of the finger-camera positioning.

3.3 Calculating HR 1. iPhone 4s and later models have enough CPU power to pro-
and NPV cess the PPG signals from all the three light color at once.
However, green light PPG has superior features as described in
the Introduction. Figure 4 shows a typical example of simulta-
neous recordings of red, green, and blue light PPG signals and
three-axis accelerations (see Note 7) with (Right) and without
(Left) intentionally produced motion artifact. The chart clearly
illustrates the higher tolerance to motion artifact in green light
PPG compared to red and blue light PPG. So only the data
from green PPG are subsequently processed.
2. The PPG signal is smoothed by using a first order IIR-type low
pass filter to extract the DC component of the PPG signal, and
then the AC component is separated from the DC component
by using the same type of high pass filter. Figure 5 shows
frequency characteristics of the filters used in iPhysioMeter.
Although there is no strict rule in filter design, the main band-
width of the PPG power spectrum is known to be less than 5 Hz
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 311

Fig. 3 Links between “Environment” windows. (a) Environment window. You can change the “environment”
that is the set of measured data and the settings. This makes it easy to manage concurrent experiments and/
or measurements of multiple individuals. The “>” button is checked for the currently selected environment.
You can select another environment by tapping its name only when measurement is not ongoing. Measurement
data are stored in the currently selected environment. It may take sometime to change to or from the environ-
ment that has a large volume of data. (b) System and operation settings. This window opens when tapping a
“>” button in a row (a). Although there are many settings, often-used settings are: (1) Sound: the pulse detec-
tion sound (OFF, POP, Morse). (2) “Complete” Vibration: the vibration when measurement stops (ON/OFF). The
setting is invalid for the iPod touch fifth generation. (3) Operation Mode: you can select either “One-Shot,”
measurement automatically stops when the measurement meter (see Fig. 2A) becomes full or “Continuous,”
measurement stops only when you press the Stop button. Quoted from ref. 29 with some modification

in ordinary practice [27, 28]. This finding holds true for PPG
signals derived using iPhysioMeter. Figure 6 shows a typical
example of the power spectra of red, green, and blue light PPG
raw (non-filtered) signals derived at rest using iPhysioMeter.
3. For all of the stored green light PPG data, 300 ms (9 frames at
30 fps) after their capture, it is determined whether or not
there is a peak (systolic) point of the AC waveform. This is car-
ried out using a local maximum search algorithm (Fig. 7).
Specifically, if the value at this time point is a maximum among
the values over the period from 333.33 ms (10 frames at
30 fps) before this point to 233.33 ms (7 frames at 30 fps) after
this point, then the point is judged as the peak. Only after a
peak has been detected is the search for a foot (diastolic) point
begun. The foot point corresponding to this peak point is
312 Kenta Matsumura et al.

1.2
Red PPG
(a.u.)

−1.2
1.5
Green PPG
(a.u.)

−1.5
0.5
Blue PPG
(a.u.)

−0.5
0.5
x-axis A
(G)

−0.5
0.5
y-axis A
(G)

−0.5
−0.5
z-axis A
(G)

-1

−1.5
90 95 100 105 110 115
Time (s)

Fig. 4 A typical example of simultaneous recordings of the alternating-current (AC) components of red, green,
and blue photoplethysmogram (PPG)s and three-axis acceleration (A)s. The left side and the right side of the
chart correspond to in the absence and presence of motion artifact, respectively. Red and blue closed circles
represent the peak and foot points identified by a beat-by-beat auto analysis algorithm equipped with iPhysi-
oMeter. Open circles identify outlier points, that may have arisen from misidentification and/or omission of an
expected preceding point, are also judged by this algorithm. Quoted from ref. 22 with some modification. a.u.
arbitrary units

determined as the first point, before the peak point, at which

the monotonic decrease ceases.
4. A pair of determined peak and foot points is used to calculate
HR and NPV. HR is calculated using the following formula:
HR (bpm) = 60,000 (ms)/peak-to-peak interval (ms).
NPV [20] is calculated using the following formula:
NPV (a.u.) = foot-to-peak amplitude/((DC/255)1/1.8 × 255),
where DC is the mean DC level between foot point and peak
point, and a.u. = arbitrary unit. A logarithmic transformation
was usually applied to the NPV values (i.e., lnNPV) to normal-
ize the distribution.
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 313

−3

Relative Gain (dB)

−6

DC filter
−9
AC filter

−12
0.1 1 10
Frequency (Hz)

Fig. 5 Frequency characteristics of alternating-current (AC) and direct-current

(DC) filters used in iPhysioMeter

5. The formula used in the denominator when calculating NPV is

introduced to aim at attaining the linearity of captured light inten-
sity from the CMOS camera. Because the specification of the
CMOS camera is undocumented, this formula was derived through
a trial and error process. Figure 8 shows lnNPV values derived from
four models of devices using an optical calibrator set to three differ-
ent output levels (see Note 8). As clearly shown, linearity is kept
across the devices, though absolute levels are somewhat different.
6. As values of HR and NPV are calculated, they are checked to
see if they can be regarded as outliers or not. This is achieved
by comparing the present values with those from the preceding
10-s period (Fig. 2B, C). Specifically, if the present lnNPV is
smaller than the mean lnNPV in the preceding 10-s period by
1.5 a.u., the present peak and foot points detected were dis-
carded completely. These are observed as peak and foot detec-
tion omissions, such as is observed in the red and blue PPG
signals in Fig. 4. This is because such a major decrease in
lnNPV is virtually impossible for normal in vivo situations and
as such is highly likely to be due to artifact. Next, if the value
contributes by increasing the standard deviation (SD) of this
period above 8.0 bpm for HR or 0.25 a.u. for lnNPV, respec-
tively, each is judged as an outlier (yellow circles in Fig. 2F and
open circles in Fig. 4). These criteria vary depending on the
selections of “Outlier Check” in iPhysioMeter “Settings” con-
trol. The criteria for “Strict,” “Moderate,” “Loose,” and “Too
Loose” are 4, 8, 12, and 20 bpm for HR and 0.15, 0.25, 0.40,
and 0.80 a.u., respectively.

3.4 Using 1. iPhysioMeter has many system and operation settings, but all of
iPhysioMeter these are set automatically according to the device on which
iPhysioMeter is run. Therefore it is not necessary to set
314 Kenta Matsumura et al.

Red PPG
0.2
HR

0.15

Power (a.u.)
2nd harmonics
0.1
of HR

0.05
3nd harmonics
of HR
0
0 5 10 15
Hz

Green PPG
HR
2

1.5
Power (a.u.)

2nd harmonics
1 of HR

0.5 3nd harmonics

of HR

0
0 5 10 15
Hz

Blue PPG HR
0.15
Power (a.u.)

0.1

2nd harmonics
0.05
of HR

0
0 5 10 15
Hz

Corresponding R-R
intervals of ECG
Fig. 6 A typical example of power spectra of red, green, and blue light photoplethysmography (PPG) raw (non-
filtered) signals. Dotted line around 1.5 Hz represents mean heart rate (HR) expressed in Hz calculated from
corresponding R-R intervals of electrocardiogram (ECG). Data points are 512 and sampling frequency (Hz) are
512 and 30, respectively. Quoted from ref. 22 with some modification. a.u. arbitrary units
iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 315

Time point Now

of analysis
1.2 Peak
300 ms
Local Maximum
Value
0.8

AC signal (a.u.)
0.4

0 Foot

−0.4
333 ms 233 ms
−0.8
91 91.5 92

Tims (s)

Fig. 7 Schematic drawing illustrating the peak and foot detection algorithm
implemented in iPhysioMeter. AC alternating current

Model
4s 5 iPod 5s
−2

−3 L
lnNPV (a.u.)

M
−4
S
−5

−6

−7

Fig. 8 Logarithmically transformed normalized pulse volume (lnNPV) derived

from four models of devices using an optical calibrator set to three different
output levels. L calibrating signal simulating large lnNPV, M calibrating signal
simulating moderato lnNPV, S calibrating signal simulating small lnNPV, 4s
iPhone 4s, 5 iPhone 5, iPod iPod touch 5th, 5s iPhone 5s

anything when conducting basic measurements. However,

before using iPhysioMeter, it is necessary to make sure that
iPhysioMeter is formally supported by your iPhone or iPod touch
and the iOS version that they use. Supporting information is
available at iTunes App Store where iPhysioMeter is distributed
or on iPhysioMeter’s homepage [29]. Although we make every
effort to update iPhysioMeter to be compatible with the latest
products, there is inevitably some delay following the release of
such new products until the matching version of iPhysioMeter is
fully debugged and tested. In this situation, it is a good idea to
316 Kenta Matsumura et al.

use the older iOS products and versions that are proven to be
stable but have enough power to run iPhysioMeter.
2. Whenever possible, it is recommended that you stop any back-
ground Apps before running iPhysioMeter.

3.5 Preparation 1. Firmly hold the iPhone or iPod touch in which iPhysioMeter
for Measurement was installed, so that the tip of the index finger completely cov-
ers both the LED flash and the CMOS camera lens (Fig. 1a).
As described earlier, the PPG signal is calculated by averaging
over the whole area of the captured image. This means that if
changes in background light levels are captured, there may not
be a correct measurement. In addition, it is important to note
that the background light should be constant during the mea-
surements if possible. If the background light level is too
bright, such as under direct sunlight, it might be impossible to
conduct the PPG measurement due to saturation of the cam-
era. Although almost all background green and blue light is
absorbed by the biological tissue [30], red light can be trans-
mitted through the finger and then exert an adverse influence
on the adjustment of the exposure of the CMOS camera.
2. It is important to keep the measuring site as constant as possi-
ble in every measurement, not only across the multiple mea-
surements in one person but also across multiple individuals.
Although the HR calculation is virtually free from this kind of
problem, the NPV is influenced by the measuring site.
3. It is also important to keep the finger—iPhone (or iPod touch)
contact pressure as constant as possible. Fluctuations in con-
tact pressure are one of the major sources of error in the NPV
measurements [21, 22]. So it is recommended to hold the
iPhone firmly, although excessive contact pressure will occlude
arterioles, and it will then be impossible to record the
photoplethysmogram.
4. To obtain reliable measurements of NPV, it is recommended
to hold the iPhone or iPod touch at heart level. Positioning
the device below or above heart level will affect the NPV value.
The HR calculation is virtually free from this kind of
problem.
5. Having dealt with all of the above matters, run iPhysioMeter
by tapping its icon on the home screen (or if already launched,
tap “Play” button), then the measurement will start automati-
cally (Fig. 2).
6. Ensure that all preparations for using iPhysioMeter are made
before the measurement process commences. Of particular
importance is to start the measurement after both the LED
flash and camera lens are covered with the index finger com-
pletely unless “Exposure Adj (adjustment)”is set to “iOS
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 317

Default” (Fig. 3b). In case iPhysioMeter does not start

measurement even though the finger is correctly positioned,
then the sensitivity will need to be adjusted to let iPhysioMeter
detect the finger on the camera. This is done by setting “Finger
Detection” in “Settings” described above. If an error occurs,
then the measurement should be stopped and restarted.

3.6 During 1. Keep the finger and the body as still as possible. Measurement is
Measurement possible under a certain level of motion artifacts because iPhysi-
oMeter benefits by the use of the green light PPG [22–25], but
keeping the finger and the body still produces better results.
2. If the finger is extremely cold, the PPG signal might not be
obtained. In such a case, PPG signals cannot be obtained even
with conventional PPG instrumentation. The only option in
this situation is to wait for re-warming of the finger.
3. The measurement will pause if iPhysioMeter detects the finger
off the camera unless “Finger Detection” in “Settings” is set to
“N/A” (Fig. 3b). Note that when resuming from this pause,
camera calibration will not be carried out when “Exposure
Adj” in “Settings” is set to “Fixed.”
4. Make sure that the PPG measurement is running correctly.
With correct operation the PPG waveform and the results of
on-line beat-by-beat analysis to give HR and NPV should be
displayed together (Fig. 2).
5. During the operation of iPhysioMeter within an experimental
procedure, it is beneficial to cover the iPhone or iPod touch
and measurement site with a black cloth to prevent the partici-
pants from looking at the display. Although this practice also
prevents the experimenters from looking at the display, the
iPhone or iPod touch’s screen can be mirrored to an external
LCD monitor by using a display adapter that is released from
Apple Inc. or third party organizations (e.g., [21]).
6. As described above, outlier judgment by iPhysioMeter is per-
formed on the assumption that HR and NPV are relatively
constant over 10 s periods, and is affected by the “Outlier
Check” selection in “Settings” (Fig. 3b). So this judgment is
not an absolute judgment, and should only be used as a guide.
7. The measurement stops automatically when the measurement
meter reaches full if “Operation Mode” in “Settings” is set to
“One-Shot” (Fig. 3b). You can also stop the measurement at
any time by tapping the “Stop” button (Fig. 2).
8. If “Operation Mode” in “Settings” is set to “Continuous”
(Fig. 3b), you can make continuous measurements indefinitely,
or at least until the battery is fully discharged. However, it is
very important to pay special attention to the inevitable tem-
perature increase in the LED flash. If the measurement period
318 Kenta Matsumura et al.

is very long low-temperature burns could be caused, together

with possible damage to the device itself, or iPhysioMeter may
be forced to quit by iOS if the in-built temperature sensor
detects potential danger. We recommend that one continuous
measurement period is limited to a maximum of 10 min with
set cooling periods between measurements. This is only a rec-
ommendation and is not a guarantee of safety.
9. Even when “Operation Mode” in “Settings” is set to
“Continuous” (Fig. 3b), the measurement process can be
interrupted by the Auto-Lock function within iOS ab initio
and a network event. To avoid this problem, set “Auto-Lock”
(“Home screen” -> “Settings” -> “General”) to “Never” and
“Airplane Mode” (“Home screen” -> “Settings”) to “ON.”
When using these settings, do not forget to turn off the iPhone
or iPod touch or recover the settings after use.
10. You can start/stop the measurement process by using a wired
or wireless Bluetooth headphone by means of its play/stop
button. When using a wireless Bluetooth headphone, note that
about 0.2 s delay will occur. In addition, if you set “Operation
Mode” in “Settings” to “Continuous” (Fig. 3b), it is advisable
to invalidate all networks such as Wi-Fi and 3G or 4G network
(“Home screen” -> “Settings”), except for Bluetooth.

3.7 After 1. Name the measurement that was automatically saved to iPhone
the Measurement Is or iPod touch’s main storage when measurement finishes
Finished (Fig. 9e). A “measurement” is defined as all of the data col-
lected during the period from the “play” button being tapped
to the end of the measurement.
2. Although you do not have to name the measurement, it is highly
recommended to do this in order to reduce mistakes in data
management. In addition, because HR and NPV related mea-
sures can be affected by many factors such as posture, mental
state, time after eating, smoking, exercising or taking medicine,
last day sleep, and air temperature, consider including these sta-
tus items to the name in case these variables are not controlled.
In fact, in our experiments, we make every effort to control
these factors to the maximum extent (e.g., [4, 21, 22, 31]).

Fig. 9 (continued) (c) or tapping “Open…” button in (b) or (d). You can name or rename the measurement,
check, e-mail, and erase each stored dataset on this screen. Data contents include the beat-by-beat data that
can be checked and sent via e-mail. They are (1) Time: the time (s) from the start of the measurement. (2) gTime:
the time (s) from when the start button was tapped. There is an approximately 1.3 s latency between Time and
gTime. (3) HR: the beat-by-beat value automatically calculated. (4) lnNPV: the beat-by-beat value automatically
calculated. (5) AC: AC component of the pulse wave. (6) DC: DC component of the pulse wave. (7) A_x, A_y, and
A_z: three-axis g-forces when the peak was detected. (8) HR_Out: checked when HR is judged as an outlier. (9)
NPV_Out: checked when lnNPV is judged as an outlier. Quoted from ref. 29 with some modification
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 319

Fig. 9 Links among “Graph” windows. (a) Graph window. You can zoom in or zoom out, or scroll the graph, and
can select either heart rate (HR) or normalized pulse volume (NPV) graphs by tapping the right-bottom button
in the display. (b) Popup Summary. When tapping the graph in (a), the popup summary of the nearest place
data appears. The summary includes the following information: serial number; date when the start button was
tapped; name; measurement duration (s); the number of segment, which increases by detaching the finger;
the number of beats; the mean HR and NPV during the whole measuring period; frames per second (fps). (c) A
list of the whole measured data. This data list opens when tapping “Data>>” button in (a). The serial number,
the mean HR and lnNPV during the whole measuring period, the number of beats, and name are also shown.
(d) Popup summary. When tapping the row in (c), the popup summary of the selected data appears. The con-
tents are same as (b). (e) Data management menu. This window opens when tapping a “>” button in a row
320 Kenta Matsumura et al.

3. It is also recommended to delete unnecessary data to reduce

mistakes in data management (Fig. 9e). However, note that
iPhysioMeter does not have the “undo” function that is com-
monly used. So once the measurement is deleted, it is impos-
sible for it to be recovered.
4. Send the “measurement” as an attached file via e-mail (Fig. 9e).
The data format is comma separated value (CSV). You can open
this file using spreadsheet software such as Excel (Microsoft, Inc.).
The contents are the same as shown in the caption of Fig. 9e.

3.8 Other Matters 1. There are likely to be individual differences in the LED flash
Considered Significant and CMOS camera even if the same models of particular
devices are used (for example, two iPhone 4Ss). So if you use
iPhysioMeter in your experiment, use only one device (for
example, your iPhone 4s) for the whole study. If that really is
impossible, then devices should be randomized (for example,
your iPhone 4S and your colleague’s iPhone 4S) across partici-
pants. Mixing usage of models (for example, your iPhone 4S
and colleague’s iPhone 5) should be avoided wherever possi-
ble. These considerations also hold true with respect to the
iOS version being used. Having said that, the HR calculation
is virtually free from this kind of problem.
2. The sampling rate of iPhysioMeter is set to 30 fps (see Note 5).
At this sampling speed, beat-by-beat HR can have in principle
approximately ±2 and ±6 bpm error at 60 and 100 bpm,
respectively (Fig. 10). One of the best practices to overcome
this problem is to average HR values over 20-s or 60-s, for
example, and get one averaged value. This method has been
shown to yield quite accurate values for HR as well as NPV as
compared with those derived from standard laboratory elec-

200

Ȉ = 13.6 bpm
150
Heart Rate (bpm)

100 Ȉ = 5.9 bpm

Ȉ = 2.1 bpm
50

0
0 500 1000 1500 2000

Peak-to-Peak Interval (ms)

Fig. 10 Discrete heart rate values derived from 30 Hz sampling frequency

(=sampling interval = 33.33 ms). bpm = beats per minute, Δ = difference value
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 321

Linear Fit of GMR Fixed Bias

95% CI Limits of Agreement

100 2
y = 1.006 x − 0.288 Bias = −0.12, SD = 0.36
r = 0.9991
HR by iPhone 4s (bpm)

Difference (bpm)
1
80

60
−1

0 −2
0 60 80 100 40 60 80 100
HR by ECG (bpm) Average (bpm)

−6 −5 −4 −3 0 2
0 Bias = 0.01, SD = 0.43
y = 0.936 x − 0.279
lnNPV by iPhone 4s (a.u.)

r = 0.791 1
Difference(a.u.)

−3

0
−4

−1
−5

−2
−6 −6 −5 −4 −3

lnNPV by NIR PPG (a.u.) Average (a.u.)

Fig. 11 Scatter plots (Left) and their Bland-Altman plots (Right; [35]) of heart rate (HR; Upper) and logarithmically
transformed normalized pulse volume (lnNPV; Lower) between iPhysioMeter and standard laboratory instru-
ments. Solid lines and dashed lines represent linear fit and its 95 % confidential interval (CI)s of geometric mean
regression (GMR; [36]), respectively. Quoted from ref. 22 with some modification. ECG electrocardiograph, PPG
photoplethysmograph, NIR=near-infrared, bpm=beats per minute, a.u. arbitrary units, CI confidential interval

trocardiograph (ECG) and PPG instruments [21, 22].

Figure 11 shows the effectiveness of this 20-s averaging tech-
nique. These data were derived from 12 participants at 3-min
rest with and without motion artifact by conducting simulta-
neous recording of green PPG using iPhysioMeter at 30 fps
and standard laboratory near-infrared (NIR) PPG and ECG at
the sampling frequency of 1,000 Hz. This graph clearly shows
that the above-mentioned ±6 bpm error around 100 bpm is
not present despite the presence of motion artifact.
322 Kenta Matsumura et al.

3. We believe that the oxygen saturation (%) in the arterial blood

(SpO2) could, in principle, be measured using iPhysioMeter
although we have not yet made any attempt to measure SpO2.
If you are interested in SpO2, please see other references
(e.g., [19, 32]).
4. Heart rate (pulse rate) variability, HRV, (PRV), can be easily
calculated from the csv files that iPhysioMeter can send via
e-mail (Fig. 9e). We have already reported the standard devia-
tion of normal-to-normal RR intervals (SDNN), which is one of
the basic time domain measures of HRV or PRV, despite only
using data at a very short rest [22]. SDNN derived from iPhysi-
oMeter was higher than that from ECG regardless of PPG light
colors, which is consistent with a previous finding from qualita-
tive review literature [33]. Figure 12 shows the agreement of
low frequency to high frequency ratio (LF/HF), which is one of
the frequently used frequency domain measures of HRV or PRV
(here, LF = 0.04–0.15 Hz, HF = 0.15–0.40 Hz), derived from
iPhysioMeter, laboratory standard NIR PPG, and ECG. These
data were derived from 12 participants at 3-min rest by conduct-
ing simultaneous recordings of red, green, and blue light PPGs
using iPhysioMeter at 30 fps and standard laboratory NIR PPG
and ECG at the sampling frequency of 1,000 Hz. These graphs
clearly demonstrate the good agreement of LF/HF derived
from laboratory NIR PPG and ECG, while those from iPhysio-
Meter and ECG were not satisfactory and LF/HF from iPhysi-
oMeter were underestimated, though the r value for green light
PPG is as high as that of NIR-PPG. Considering the previous
finding that the agreement between HRV and PRV decreases in
accordance with the decrease in sampling frequency [34], 30 fps
would be too low to derive satisfactory agreement. Further
development is obviously needed.

4 Notes

1. Currently, iPhysioMeter adopts the terms “HR (heart rate)”

and “NPV (normalized pulse volume)” as the indices pro-
duced from the photoplethysmogram by applying the appro-
priate calculation formulae [20]. However, strictly speaking, it
is more correct to use the direct names, that is pulse rate (PR)
and cutaneous pulse volume (CPV) respectively. In addition, it
is desirable that these alternative names reflect the fact that the
measurements were obtained by means of a smartphone. So in
the later release version of iPhysioMeter we plan to rename
these indices iPR and iCPV, respectively. Note that this is only
a renaming exercise, so this change does not affect other
aspects of the system or of the data.
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 323

Linear Fit of GMR Linear Fit of GMR

95% CI 95% CI
Red PPG NIR PPG
6 6
y = 0.815 x − 0.522 y = 0.994 x − 0.175
r = .531 r = .982
LF/HF by Red Light

LF/HF by NIR Light

4 4

2 2

0 0
0 2 4 6
0 2 4 6
LF/HF by ECG LF/HF by ECG

Green PPG
6
y = 0.734 x − 0.214
r = .909
LF/HF by Green Light

0
0 2 4 6
LF/HF by ECG

Blue PPG
6
y = 0.626 x − 0.408
r = .615
LF/HF by Blue Light

0
0 2 4 6
LF/HF by ECG

Fig. 12 Scatter plots of low to high frequencies ratio (LF/HF) between iPhysioMeter and electrocardiograph (ECG; Left)
and standard laboratory near-infrared (NIR) photoplethysmography (PPG) and ECG (Right). Solid lines and dashed lines
represent linear fit and its 95 % confidential interval (CI)s of geometric mean regression (GMR; [36]), respectively
324 Kenta Matsumura et al.

2. When this manuscript was written, available models were iPhone

4S, iPhone 4s, iPhone 5, iPhone 5c, and iPhone 5s or music
player iPod Touch fifth generation. Currently, iPod models are
not available because they are without an LED flash.
3. When this manuscript was written, the latest version of iPhysi-
oMeter was 1.2.1 and was supported by iOS 5.1 to iOS 7.0.6.
4. When this manuscript was written, the latest version of Xcode
and OS X were 5.0.2 and 10.9.1, respectively.
5. The combination of iOS 7 and iPhone 5s, the latest versions,
supports 120 fps. We have confirmed the stable operation of at
least 60 fps with this combination.
6. In fact, there is “AutoExpose” mode.
7. g-Force measurement is easily implementable in iOS.
8. This calibrator, build in our laboratory, generates light that
resembles PPG signal using pseudo white color LED. By
bonding red cellophane to the lens of CMOS camera, we can
easily achieve acceptance by iPhysioMeter.

References

1. Parker KH (2009) A brief history of arterial wave responses to stress following alcohol intake in
mechanics. Med Biol Eng Comput 47:111–118. social drinkers who are at risk of hazardous
doi:10.1007/s11517-009-0440-5 drinking. Biol Psychol 93:9–16. doi:10.1016/j.
2. Challoner AVJ (1979) Photoelectric plethys- biopsycho.2012.12.009
mography for estimating cutaneous blood flow. 8. Yamakoshi T, Matsumura K, Hanaki S, Rolfe
In: Rolfe P (ed) Noninvasive physiological P (2013) Cardiovascular hemodynamic
measurements, vol 1. Academic, London, effects of Red Bull(R) Energy Drink
pp 125–151 during prolonged, simulated, monotonous
3. Matsumura K, Yamakoshi T, Yamakoshi Y, driving. SpringerPlus 2:215. doi:10.1186/
Rolfe P (2011) The effect of competition on 2193-1801-2-215
heart rate during kart driving: a field study. 9. Yamakoshi T, Matsumura K, Rolfe P, Hanaki S,
BMC Res Notes 4:342. Ikarashi A, Lee J, Yamakoshi K (2014) Potential
doi:10.1186/1756-0500-4-342 for health screening using long-term cardiovas-
4. Matsumura K, Yamakoshi T, Noguchi H, Rolfe cular parameters measured by finger volume-
P, Matsuoka Y (2012) Fish consumption and oscillometry: pilot comparative evaluation in
cardiovascular response during mental stress. regular and sleep-deprived activities. IEEE J
BMC Res Notes 5:288. Biomed Health Inform 18:28–35.
doi:10.1186/1756-0500-5-288 doi:10.1109/JBHI.2013.2274460
5. McNally RJ, Lasko NB, Clancy SA, Macklin 10. Allen J (2007) Photoplethysmography and its
ML, Pitman RK, Orr SP (2004) application in clinical physiological measure-
Psychophysiological responding during script- ment. Physiol Meas 28:R1–R39.
driven imagery in people reporting abduction doi:10.1088/0967-3334/28/3/R01
by space aliens. Psychol Sci 15:493–497. 11. Pitman RK, Orr SP, Forgue DF, de Jong JB,
doi:10.1111/j.0956-7976.2004.00707.x Claiborn JM (1987) Psychophysiologic assess-
6. al’Absi M, Nakajima M, Hooker S, Wittmers L, ment of posttraumatic stress disorder imagery
Cragin T (2012) Exposure to acute stress is in Vietnam combat veterans. Arch Gen
associated with attenuated sweet taste. Psychiatry 44:970–975
Psychophysiology 49:96–103. 12. Kuvin JT, Patel AR, Sliney KA, Pandian NG,
doi:10.1111/j.1469-8986.2011.01289.x Sheffy J, Schnall RP, Karas RH, Udelson JE
7. Nakajima M, Kumar S, Wittmers L, Scott MS, (2003) Assessment of peripheral vascular endo-
al’Absi M (2013) Psychophysiological thelial function with finger arterial pulse wave
 iPhysioMeter: A Smartphone Photoplethysmograph for Measuring Various… 325

amplitude. Am Heart J 146:168–174. 22. Matsumura K, Rolfe P, Lee J, Yamakoshi T

doi:10.1016/S0002-8703(03)00094-2 (2014) iPhone 4s photoplethysmography:
13. Tanaka G, Yamakoshi K, Sawada Y, Matsumura which light color yields the most accurate heart
K, Maeda K, Kato Y, Horiguchi M, Ohguro H rate and normalized pulse volume in the pres-
(2011) A novel photoplethysmography tech- ence of motion artifact? PLoS One 9:e91205.
nique to derive normalized arterial stiffness as a doi:10.1371/journal.pone.0091205
blood pressure independent measure in the fin- 23. Jonathan E, Leahy MJ (2011) Cellular phone-
ger vascular bed. Physiol Meas 32:1869–1883. based photoplethysmographic imaging. J
doi:10.1088/0967-3334/32/11/003 Biophotonics 4:293–296. doi:10.1002/
14. Matsuoka Y, Nishi D, Yonemoto N, Hamazaki jbio.201000050
K, Matsumura K, Noguchi H, Hashimoto K, 24. Lee J, Matsumura K, Yamakoshi K, Rolfe P,
Hamazaki T (2013) Tachikawa project for pre- Tanaka S, Yamakoshi T (2013) Comparison
vention of posttraumatic stress disorder with between red, green and blue light reflection
polyunsaturated fatty acid (TPOP): study pro- photoplethysmography for heart rate monitor-
tocol for a randomized controlled trial. BMC ing during motion. Conf Proc IEEE Eng Med
Psychiatry 13:8. Biol Soc 2013:1724–1727. doi:10.1109/
doi:10.1186/1471-244X-13-8 EMBC.2013.6609852
15. Matsumura K, Noguchi H, Nishi D, Matsuoka 25. Maeda Y, Sekine M, Tamura T (2011) The
Y (2012) The effect of omega-3 fatty acids on advantages of wearable green reflected photo-
psychophysiological assessment for the second- plethysmography. J Med Syst 35:829–834.
ary prevention of posttraumatic stress disorder: doi:10.1007/s10916-010-9506-z
an open-label pilot study. Glob J Health Sci 26. Apple (2011) AV Foundation Programming
4:3–9. doi:10.5539/gjhs.v4n1p3 Guide. http://developer.apple.com/library/
16. Pitman RK, Sanders KM, Zusman RM, Healy ios/DOCUMENTATION/AudioVideo/
AR, Cheema F, Lasko NB, Cahill L, Orr SP Conceptual/AVFoundationPG/Articles/00_
(2002) Pilot study of secondary prevention of Introduction.html. Accessed 27 Feb 2014
posttraumatic stress disorder with propranolol. 27. Kamal AA, Harness JB, Irving G, Mearns AJ
Biol Psychiatry 51:189–192 (1989) Skin photoplethysmography—a review.
17. Mozaffarian D, Geelen A, Brouwer IA, Comput Methods Programs Biomed 28:257–269
Geleijnse JM, Zock PL, Katan MB (2005) 28. Hayes MJ, Smith PR (2001) A new method for
Effect of fish oil on heart rate in humans: a pulse oximetry possessing inherent insensitivity
meta-analysis of randomized controlled trials. to artifact. IEEE Trans Biomed Eng 48:452–
Circulation 112:1945–1952. doi:10.1161/ 461. doi:10.1109/10.915711
CIRCULATIONAHA.105.556886 29. Yamakoshi T, Matsumura K (2014) iPhysioM-
18. Jonathan E, Leahy M (2010) Investigating a eter.com. http://iphysiometer.com/app/.
smartphone imaging unit for photoplethys- Accessed 27 Feb 2014
mography. Physiol Meas 31:N79–N83. 30. Giltvedt J, Sira A, Helme P (1984) Pulsed mul-
doi:10.1088/0967-3334/31/11/N01 tifrequency photoplethysmograph. Med Biol
19. Scully CG, Lee J, Meyer J, Gorbach AM, Eng Comput 22:212–215. doi:10.1007/
Granquist-Fraser D, Mendelson Y, Chon KH BF02442745
(2012) Physiological parameter monitoring 31. Yamakoshi T, Matsumura K, Rolfe P (2014)
from optical recordings with a mobile phone. Controlled mechanical vibration applied to
IEEE Trans Biomed Eng 59:303–306. driver’s right heel to sustain alertness: effects
doi:10.1109/TBME.2011.2163157 on cardiovascular behavior. Transport Res C
20. Lee J, Matsumura K, Yamakoshi T, Rolfe P, Emerg Technol 38:101–109. doi:10.1016/j.
Tanaka N, Kim K, Yamakoshi K (2013) trc.2013.10.009
Validation of normalized pulse volume in the 32. Karlen W, Ansermino JM, Dumont GA,
outer ear as a simple measure of sympathetic Scheffer C (2013) Detection of the optimal
activity using warm and cold pressor tests: region of interest for camera oximetry. Conf
towards applications in ambulatory monitor- Proc IEEE Eng Med Biol Soc 2013:2263–
ing. Physiol Meas 34:359–375. 2266. doi:10.1109/EMBC.2013.6609988
doi:10.1088/0967-3334/34/3/359
33. Schafer A, Vagedes J (2013) How accurate is
21. Matsumura K, Yamakoshi T (2013) iPhysioM- pulse rate variability as an estimate of heart rate
eter: a new approach for measuring heart rate variability? A review on studies comparing pho-
and normalized pulse volume using only a toplethysmographic technology with an elec-
smartphone. Behav Res Methods 45:1272– trocardiogram. Int J Cardiol 166:15–29.
1278. doi:10.3758/s13428-012-0312-z doi:10.1016/j.ijcard.2012.03.119
326 Kenta Matsumura et al.

34. Giardino ND, Lehrer PM, Edelberg R (2002) methods of clinical measurement. Lancet
Comparison of finger plethysmograph to ECG 1:307–310. doi:10.1016/
in the measurement of heart rate variability. S0140-6736(86)90837-8
Psychophysiology 39:246–253. doi:10.1017/ 36. Ludbrook J (1997) Comparing methods of
S0048577202990049 measurements. Clin Exp Pharmacol Physiol
35. Bland JM, Altman DG (1986) Statistical meth- 24:193–203. doi:10.1111/j.1440-1681.1997.
ods for assessing agreement between two tb01807.x
 Chapter 22

Smartphone Attachment for Stethoscope Recording

Jeff Thompson

Abstract
With the ubiquity of smartphones and the rising technology of 3D printing, novel devices can be
developed that leverage the “computer in your pocket” and rapid prototyping technologies toward
scientific, medical, engineering, and creative purposes. This paper describes such a device: a simple
3D-printed extension for Apple’s iPhone that allows the sound from an off-the-shelf acoustic stetho-
scope to be recorded using the phone’s built-in microphone. The attachment’s digital 3D files can be
easily shared, modified for similar phones and devices capable of recording audio, and in combination
with 3D printing technology allow for fabrication of a durable device without need for an entire fac-
tory of expensive and specialized machining tools. It is hoped that by releasing this device as an open
source set of printable files that can be downloaded and reproduced cheaply, others can make use of
these developments where access to cost-prohibitive, specialized medical instruments are not avail-
able. Coupled with specialized smartphone software (“apps”), more sophisticated and automated
diagnostics may also be possible on-site.

Key words iPhone, Stethoscope, Audio recording, 3D printing, STL, Shapeways, Smartphone, App,
Sound, Selective laser sintering, Fused deposition modeling

1 Introduction

The following is more of a “how to” than a laboratory procedure;

this is worth noting for two reasons. First, 3D printing handles much
of the heavy lifting, creating the component that acts as interface
between the iPhone and the stethoscope. Once printed, the compo-
nent and related pieces require minimal preparation for use. Second,
I am trained and work as a visual artist, and this device is born out of
my creative research. The stethoscope attachment was originally
conceived as a device for making interesting, experimental audio
recordings. The additional use as a medical instrument provides fur-
ther evidence of something I believe strongly: that research in the
arts can have practical and important applications in STEM fields.
Such a device could be used for a variety of medical purposes,
including remote diagnosis of cardiovascular or respiratory issues,
real-time telemedicine between a specialist and a health worker in

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_22, © Springer Science+Business Media New York 2015

327
328 Jeff Thompson

Fig. 1 A schematic drawing of a standard acoustic stethoscope

a remote or underserved region, and, with the addition of

custom-written smartphone applications, performing automated,
audio-based analysis for diagnosis by non-professionals. While
electronic and recording stethoscopes do exist, they are very
expensive (see ref. 3). Coupled with this attachment, even an
inexpensive, store-bought acoustic stethoscope can make record-
ings sufficient for diagnosis.
The central part of the device is a 3D printed interface between
the iPhone (selected because it is the most ubiquitous smartphone
today, though this design could easily be adapted for other devices)
and a standard acoustic stethoscope (see Fig. 1). The part is 3D
printed in selective laser-sintered (SLS) nylon or ABS plastic, and is
friction-fit to the bottom of the phone where the microphone is
located. A barbed fitting at the base of the attachment connects the
stethoscope’s PVC tubing. A hole down the barb carries sound
from the tubing directly into the smartphone’s microphone, the
same way sound in a traditional stethoscope is passed through the
earpiece into the ear.
In order to capture the full range of adult chest sounds, higher-
quality stethoscope heads have two sides: the diaphragm for higher
frequency sounds in the range of approximately 100–1,000 Hz,
and the bell for lower frequency sounds in the range of 20–100 Hz
(see ref. 1). The microphone in the iPhone 4 (released in 2010) has
an attenuated frequency response, as we might guess in a consumer
device not intended for high-fidelity audio. However, it is capable
of recording the entire frequency range outlined above, with a very
flat frequency response past 200 Hz. This suggests that, as an inex-
pensive, portable, and very flexible device, the iPhone is very well
suited for recording the sounds from a stethoscope.
 Smartphone Attachment For Stethoscope Recording 329

Fig. 2 A CAD rendering of the 3D printed attachment, which slides tightly onto the bottom of the iPhone. Note
how the sound from the stethoscope’s tube is funneled directly into the phone’s built-in microphone. Shapes
like this are difficult to manufacture, making 3D printing an ideal solution in such cases

2 Materials

1. Stethoscope head with fitting for standard 3/16–1/4″ (4.75–

6.3 mm) ID tubing.
2. Standard stethoscope tubing (single lumen), 3/8″ OD, 3/16–
1/4″ (4.75–6.3 mm) ID, at least 6″ long (see Note 1).
3. 3D-printed iPhone attachment, either using a commercial ser-
vice printing in laser-sintered nylon (preferable), or locally using
a hard material such as ABS or PLA plastic (see Figs. 2 and 3).
For an overview of 3D printing technologies, see refs. [2–4].
4. iPhone generation four or above.
5. Sound recording app for iPhone such as the preinstalled “Voice
Recorder.”
6. Optional: audio post-production software for filtering record-
ings to improve the clarity of recordings, especially of low-
frequency sounds such as heartbeats; the free, open-source
Audacity is an excellent option for this purpose.

3 Methods

This section describes 3D printing, assembly, and use of the stetho-

scope attachment. The first of these instructions describe printing
the attachment using the commercial service Shapeways; if a 3D
printer is available locally, the attachment can be printed using a
variety of alternative methods outlined in Subheading 3.2
(see Subheading 4 for possible issues and further suggestions).
330 Jeff Thompson

Fig. 3 The completed iPhone stethoscope attachment, 3D-printed in laser-sintered nylon

3.1 Printing Shapeways, a 3D printing service based in the USA and the
the Attachment Using Netherlands, provides very high-quality printing in a variety of
Shapeways’ 3D materials.
Printing Service 1. Visit the following link to order a 3D-printed version of the
stethoscope attachment: http://shpws.me/oQUX.
2. Choose a material color from the “Material Options” drop-
down menu—default is white (see Note 2).
3. Click the “Buy Now” button to order the printed model in the
material specified (see Note 3); at time of writing, commercial
printing of the attachment was approximately $30 USD.

3.2 Alternatively, Inexpensive 3D printers designed for consumer/hobbyist use such as

Print Using a Local 3D the MakerBot or open-source RepRap may be substituted without loss
Printer of audio fidelity, though they may take more work to produce usable
prints.
1. Visit the project’s GitHub repository: https://github.com/
jeffthompson/iPhoneStethoscopeAttachment.
2. Download the necessary files by clicking the “Download ZIP”
button in the right sidebar. This includes the printable stereo-
lithography (STL) files, as well as supporting documents and
sample recordings.
3. Print using a hard material such as ABS or PLA plastic at 100 %
infill; other settings will depend on model of 3D printer
(see Note 4).
4. Clean and sand/file printed model as necessary for a smooth finish.
 Smartphone Attachment For Stethoscope Recording 331

3.3 Prepare A single length of flexible PVC tubing is needed to connect the chest-
Stethoscope Tubing piece to the 3D printed attachment. This can be repurposed from an
existing stethoscope or bought in bulk and cut to size (see Note 1).
1. Cut one section of stethoscope tubing using a sharp knife or
scissors. If using the tubing from an existing stethoscope, cut
below the fork to form a single tube. Length is variable but less
than 15″ (38 cm) can make it difficult to maneuver; very long
tubing will likely cause a reduction in the fidelity of the audio.

3.4 Assemble For an image of the completed assembly, see Fig. 2.

the Attachment
1. Slide one end of the stethoscope tubing over the barb in the
stethoscope head until secure.
2. Insert the other end of the stethoscope tubing over the barb at
the bottom of the 3D printed attachment; the tubing should
fit tightly but can be carefully removed (see Note 5). While the
3D printed parts are sturdy, care should be taken not to break
the barb when attaching the tubing.
3. Slide the 3D printed attachment over the bottom of the phone;
the half-circle cutout in the attachment should align with the
“home” button. Slide the attachment in until the bottom of the
phone hits the bottom of the 3D printed attachment (see Note 6).

3.5 Test and Record 1. Turn on the phone and launch an audio recording app like
Voice Recorder (see Note 7).
2. Set your app’s settings to highest quality possible, if available.
Higher sample rate and bit depth will result in higher-fidelity
recordings. Avoid recording in compressed formats like MP3 if
non-compressed formats such as WAV or AIFF are available.
3. Press record and test the stethoscope. Depending on your
app’s sensitivity settings, loud sounds such as tapping the
stethoscope head should be very clearly picked up as sharp
spikes by the microphone (see Note 8).

3.6 Audio Processing While the recordings from the stethoscope may be used without any
and Analysis processing, some filtering will make the isolation of certain sounds
easier, especially for automated analysis.
1. A variety of adult chest sounds can be heard with a stethoscope
and can be identified by their frequency ranges. These include
“low heart sounds (including first, second, and third heart
sounds) from 20 to 115 Hz; medium/high heart sounds
(including systolic and diastolic murmurs) from 200 to
660 Hz; vesicular (normal) breathing from 150 to 1,000 Hz;
bronchial breathing from 240 to 1,000 Hz; and crepitations
(crackles) greater than or equal to 750 Hz” (see ref. 1).
332 Jeff Thompson

Fig. 4 The waveform from an audio recording of a human heartbeat with no filter applied. While the heartbeat
is visible, there is significant noise in the recording

Fig. 5 The same recording of a human heartbeat as Fig. 4 with a low-pass filter applied. The heartbeat is much
more clearly visible, especially to the untrained eye

Fig. 6 Frequency analysis of a recording of a human heartbeat with no filter applied. While the noise floor for
the iPhone’s built-in microphone is rather high, the low-frequency heartbeat is still clearly visible

2. For lower frequency sounds like heartbeats, a low-pass filter

may be useful for removing noise and more clearly identifying
the beats; this can be applied after the recording is made, or
with a custom app in real-time (see Figs. 4, 5, and 6).
 Smartphone Attachment For Stethoscope Recording 333

4 Notes

1. The design of the attachment’s barb allows for some variation

in the dimensions of the tubing: any tube approximately
3/16–1/4″ (4.75–6.3 mm) inside diameter will fit.
In the interest of accessibility and cost-savings,
“UV-Resistant Black PVC Tubing” can be purchased in bulk
from: http://www.mcmaster.com/#5231k83/=oainxw.
2. Shapeways’ “Strong and Flexible” material is laser-sintered
nylon plastic, resulting in prints that are very strong with some
flexibility. Printed parts can be easily cleaned, are dishwasher
safe, and are heatproof to 80 °C/176 °F. See the following link
for more information: https://www.shapeways.com/materi-
als/strong-flexible.
3. Shapeways offers a variety of other materials for printing. If
you would like to experiment with materials other than the
default laser-sintered nylon, you will need to download the
source files from GitHub and upload them to Shapeways.
4. Depending on your printer’s method and output material, spe-
cific settings may be required for a high-quality and durable
print. For Fused Deposition Modeling (FDM) printing like
the MakerBot, shrinkage may also require reengineering the
model based on trial and error (such considerations are not
needed for laser-sintered nylon). It is suggested that FDM
printers are set to 100% infill with at least 2–3 shells for a
durable print.
5. This design, with the barb extending from the bottom of the
3D printed attachment, allows for the strongest connection
with the tubing. An alternative design, available on the GitHub
repository, simply has a hole in the bottom of the attachment
for inserting the tubing. This is less sturdy but very easy to
disassemble and put back together for cleaning, storage, or
customization.
6. While every effort has been taken to avoid damage to the
iPhone, it should be noted that the 3D printed attachment is
pressure-fit to the phone’s body. Attachments printed using
Shapeways’ laser-sintered nylon and MakerBot-printed ABS
plastic both left no scratches on the phone’s surface, even at a
tight fit and many installations/removals. However, use at
your own risk.
7. Apple’s built-in Voice Recorder app is quite adequate for basic
recording, but offers little manual control. Voice Record Pro
(free, https://itunes.apple.com/us/app/voice-record-pro/
id546983235?mt=8) offers manual quality adjustment (the
“high quality” setting is 44.1 kHz/16-bit at a bitrate of
334 Jeff Thompson

128,000) but is limited to the AAC/MP4/M4A compressed

audio formats. Audio Memos ($9.99 USD with in-app
upgrades, https://itunes.apple.com/us/app/audio-memos-
pro/id290160980?mt=8) has a large level meter and allows
recording in the uncompressed WAV format. Both apps can be
set to auto-upload to Dropbox and similar cloud services.
8. Better-quality audio recording apps may have adjustable gain
for microphone input. If audio level problems persist, check
that the stethoscope tubing and barb of the 3D printed attach-
ment to the iPhone microphone are not blocked.

References

1. Dawson JB (1964) Practitioner 193:315–322 portal/3M/en_US/3M-Littmann/stetho-

2. Faber Acoustical has conducted a variety of high-qual- scope/stethoscope-catalog/catalog/~/All-3M-
ity audio testing on the iPhone 4 and related devices, Products/Brands/3M-Littmann-Stethoscopes/
though finding data on the more recent iPhone mod- All-Stethoscopes?N=5002683+5927679+59322
els did not turn up any reliable testing. For details on 56&rt=c3
the process and graphs of the results, see: http://blog. 4. For an introduction to the technologies of 3D
faberacoustical.com/2010/ios/iphone/ printing, see these useful guides: “What Is 3D
iphone-4-audio-and-frequency-response-limitations Printing? An Overview,” 3D Printer http://
3. Various stethoscope recording devices exist but www.3dprinter.net/reference/what-is-3d-printing
are very expensive and do not include the “3D Printing Technologies Explained,” Shapeways
advanced signal processing functionality Blog https://www.shapeways.com/blog/
that a smartphone app could. For example, the archives/1215-3d-printing-technologies-explained.
3M Littmann electronic stethoscope costs over html?%2Farchives%2F1215-3d-printing-
$300 USD: http://www.littmann.com/wps/ technologies-explained
 Chapter 23

Use of Smartphones and Portable Media Devices

for Quantifying Human Movement Characteristics
of Gait, Tendon Reflex Response, and Parkinson’s
Disease Hand Tremor
Robert LeMoyne and Timothy Mastroianni

Abstract
Smartphones and portable media devices are both equipped with sensor components, such as accelerom-
eters. A software application enables these devices to function as a robust wireless accelerometer platform.
The recorded accelerometer waveform can be transmitted wireless as an e-mail attachment through con-
nectivity to the Internet. The implication of such devices as a wireless accelerometer platform is the experi-
mental and post-processing locations can be placed anywhere in the world. Gait was quantified by mounting
a smartphone or portable media device proximal to the lateral malleolus of the ankle joint. Attributes of
the gait cycle were quantified with a considerable accuracy and reliability. The patellar tendon reflex
response was quantified by using the device in tandem with a potential energy impact pendulum to evoke
the patellar tendon reflex. The acceleration waveform maximum acceleration feature of the reflex response
displayed considerable accuracy and reliability. By mounting the smartphone or portable media device to
the dorsum of the hand through a glove, Parkinson’s disease hand tremor was quantified and contrasted
with significance to a non-Parkinson’s disease steady hand control. With the methods advocated in this
chapter, any aspect of human movement may be quantified through smartphones or portable media
devices and post-processed anywhere in the world. These wearable devices are anticipated to substantially
impact the biomedical and healthcare industry.

Key words Portable media device, iPod, Smartphone, iPhone, Wireless accelerometer, Wearable device,
Quantification, Patellar tendon reflex, Reflex response, Gait, Gait analysis, Parkinson’s disease, Tremor

1 Introduction

Software that runs on smartphone or portable media devices is

gaining mounting presence in the biomedical and healthcare
industry. Robust software enables their application in a patient’s
home setting. These devices consist of integrated sensors, such as
accelerometers, that enable the recording of a subject’s movement
attributes as an acceleration waveform. The acceleration waveform,
stored in a file, can be e-mailed via wireless connectivity.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_23, © Springer Science+Business Media New York 2015

335
336 Robert LeMoyne and Timothy Mastroianni

Subsequently the acceleration waveform file can be obtained by

post-processing resources for analysis. Smartphones or portable
media devices can be deployed for experimentation in one location
and post-processing can occur in another location hundreds or
thousands of miles away [1–5].
The common qualities of smartphones and portable media
devices are their robust operating system. For the designated bio-
medical experiments, the same application for acquiring and wire-
lessly e-mailing the acceleration waveform recording is implemented
for these wireless devices. There are disparate contexts for such
devices regarding their usage in the biomedical and healthcare indus-
try. For example, the iPhone with a mass of 137 g possesses an
Internet connectivity range on the order of cell-phone coverage,
whereas the lighter iPod with a mass of 101 g must remain within the
potentially vast yet relatively limited footprint of a wireless Internet
zone for the capacity to wirelessly convey data though e-mail [1–5].
Since 2010 LeMoyne et al. have applied smartphones or por-
table media devices for acceleration waveform acquisition and sub-
sequent e-mail by wireless transmission through the Internet as a
wearable and wireless accelerometer platform for characterizing an
assortment of human movement attributes. The quantification of
gait features through these devices has been successfully demon-
strated by recording subject’s gait in essentially autonomous envi-
ronments. Implications are the capacity to monitor gait
rehabilitation as a function of therapy strategy with the therapist
and subject in remote locations [2, 3]. Patellar tendon reflex
response has been acquired through these devices enabling a
recording of the reflex response, for which post-processing and
database comparison can enable advanced diagnostics [4, 5].
The most significant applications for these devices are related
to the acquisition of Parkinson’s disease hand tremor features, for
which a clinician can track the nature of the neurodegenerative
disease and advise modifications to therapy strategy. Samplings of
the subject’s hand tremor can be obtained at the convenience of
the subject’s home setting. The data package can be wirelessly con-
veyed to the clinician by connectivity to the Internet. The clinician
can have the tremor acceleration waveform post-processed at a
clinical setting and remotely advise the patient of an optimal ther-
apy. Of particular interest, these devices can be used to measure
hand tremor related to Parkinson’s disease by mounting the smart-
phone to the dorsum of the subject’s hand by a simple glove [1].

1.1 Smartphones/ One of the most evolved biomedical applications for wearable and
Portable Media wireless devices pertains to gait analysis [6, 7]. During the 1950s
Devices for Gait the accelerometer applications were recommended for the role of
Quantification quantifying human movement. Industries associated with the
accelerometer technology space, such as the automotive industry,
Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 337

promoted the evolution of the state of the art. Accelerometer

systems reached a sufficient threshold of technology readiness to
be applied to biomedical and health care applications [7, 8].
The steady evolution of correlated technologies, such as sen-
sors, microelectronics, telecommunications, and data processing,
have enabled the implementation of wireless accelerometers as
wearable systems [6, 7]. Sensor hardware, microelectronics, wire-
less communication architecture, and data analysis strategy are
foundational technologies to wearable applications. The continual
advance in the miniaturization of sensors and electronic circuitry
has significantly impacted the application domain of wearable
devices. Wireless technology has substantially progressed over the
span of the past decade, which has resulted in systems requiring
tethering for sensor data transmission to become outmoded [6].
Wearable systems equipped with an accelerometer providing a
functionally wireless accelerometer capability can substantially
impact the utility of the modern health care system. While provid-
ing optimal autonomy for the patient, wearable applications can
advance monitoring and diagnostic insights. The quality of patient
care can be enhanced, especially in a rural environment [6].
During 2005 Jovanov et al. developed a Wireless Body Area
Network (WBAN) for monitoring patient status with integral accel-
erometer sensors. The device was projected to quantify a patient’s
medical condition, with individual care optimized through a medi-
cal database [9]. During 2006 Saremi et al. and Kavanagh et al.
made significant progress regarding the application of wireless
accelerometer nodes as wearable devices [10, 11]. Saremi et al.
advanced a multi-node accelerometer system as a wearable and
wireless application using highly specific anatomical mounting posi-
tions for evaluating gait parameters. The validity of the wearable
and wireless accelerometer system relative to conventional gait anal-
ysis equipment, such as EMG, a footswitch device, and infrared
markers, through video gait analysis were successfully contrasted
[10]. Kavanagh et al. implemented a multi-node wireless acceler-
ometer system with a similar highly specified anatomical mounting
strategy. The research succeeded in demonstrating that different
examiners or the same examiner could attain minimal reapplication
errors with comparable gait quantification reliability [11]. Ensuing
research utilized a similar application for evaluating the function of
the neck and trunk regarding stability of the head during gait [12].
A significant aspect of the research by Saremi et al. and Kavanagh
et al. was the use of highly defined anatomical mounting positions
for the application of accelerometer nodes [10, 11]. LeMoyne et al.
later revised the mounting technique by using easily identifiable
anatomical positions, such as the lateral malleolus proximal to the
ankle joint, rather than positions requiring a degree of specializa-
tion, such as the L3 spinous process on the trunk [13].
338 Robert LeMoyne and Timothy Mastroianni

Lee et al. conducted validation research for gait analysis using

wireless accelerometers. The research contrasted an integral wireless
accelerometer system to a footswitch. The findings implicate that
the temporal patterns of the acceleration waveform are comparably
valid to the data acquired by a traditional footswitch [14, 15].
Bamberg et al. developed another integrated gait analysis
application called GaitShoe, consisting of accelerometers, gyro-
scopes, force sensors, bend sensors, electric field height sensors,
and pressure sensors. The advantage of the device is the sensor
subsystems are integrated into a shoe, which is a simplified mount-
ing strategy. The sensor package has volumetric constraint on order
of the shoe [16].
Between 2007 and 2009 LeMoyne et al. demonstrated the
capability to quantify hemiplegic gait characteristic through
mounting accelerometer nodes to the lower limb of each leg, such
as either the lateral epicondyle proximal to the knee joint or the
lateral malleolus proximal to the ankle joint [17, 18]. During this
general time frame LeMoyne et al. also successfully demonstrated
the role of wireless accelerometers for real-time biofeedback for
gait rehabilitation [19, 20]. Two highly relevant parameters derived
by the gait acceleration signal are the temporal difference from a
stance phase to the following stance phase and the time-averaged
acceleration from stance to stance. The temporal difference from a
stance phase to the following stance phase is identified by the dis-
parities between the consecutive maximal peaks of the acceleration
waveform, which each represent a stance event. The time-averaged
acceleration from stance to stance is derived from the integral of
the acceleration waveform from two consecutive stance events.
The temporal difference from a stance phase to the following
stance phase and the time-averaged acceleration from stance to
stance represent quantified spatial-temporal parameters that can
infer a level of symmetry regard both legs or a level of consistency
regarding a single leg [2, 3, 7, 17–20].
A single wireless triaxial accelerometer device demonstrated
the capability to differentiate between healthy gait and hemiplegic
gait from the resulting acceleration signal [21]. Another wireless
inertial sensor network incorporating multiple nodes was able to
quantify the disparity between healthy gait and hemiplegic gait
[22]. Another research team applied a wireless accelerometer appli-
cation with transmission to a PDA and then a PC for storage and
post-processing. The application assessed temporal phases of hemi-
plegic gait [23].
Wearable and wireless accelerometer systems have been applied
to fundamental gait research and specific aspects of the gait cycle.
A trunk mounted wireless inertial sensor entailing a triaxial
accelerometer demonstrated partial success in comparison to tradi-
tional gait analysis equipment [24]. Wireless devices with integral
Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 339

accelerometers have been capable of acquiring joint angle [25, 26].

Wireless accelerometer applications have been applied to evaluate
the influence of gait speed with regard to trunk motion [27].
Other subsets of gait such as toe clearance have been illustrated
through wireless systems with accelerometers [28].
Wearable applications have been developed for evaluating gait
for subjects with osteoarthritis. The sensor conveyed accelerome-
ter data wireless from PDA and wireless to PC for post-processing
has successfully quantified compensatory aspects of osteoarthritic
gait [29]. Another wearable application evaluated gait symmetry
for trans-femoral amputee through conveying the accelerometer
signal data wireless to a PC for post-processing [30, 31].
Traditional applications monitoring activity status are based
on the acceleration waveform [7]. Wireless accelerometer/wear-
able applications have been applied to classifying activity status
and monitoring of elderly people in an indoor environment [32, 33].
The intensity of a therapy session in terms of activity time com-
pared to total time has been determined through a WBAN
equipped with accelerometers [34]. Another novel approach uti-
lizes the wireless accelerometer systems of the Nintendo Wii
Sports games for positively influencing chronic stroke patient
health status [35].
Advances in wearable applications have enabled the introduc-
tion of biofeedback as a therapy mode for neurorehabilitation [36].
Wireless accelerometer nodes secured to both legs have been suc-
cessfully tested and evaluated as a rehabilitation mode for hemiple-
gic gait. The system has been termed “Virtual Proprioception,”
which evaluates the relative similarity of the affected leg and unaf-
fected leg acceleration waveforms and enables the subject to adapt
their gait pattern according to audio or visual feedback [19, 20].
Another biofeedback therapy pertains to balance training [36].
A wireless accelerometer node can be worn near the trunk of the
subject. Real time data can evaluate postural control as a function
of fatigue status [37]. Audio biofeedback has assessed postural
sway by means of an accelerometer transmitting to a local PC [38].
Another balance training application involved an iPhone, success-
fully providing real-time vibrotactile feedback [39].
Advanced algorithms can augment wireless accelerometer bio-
medical applications. The acuity of calculating gait kinetics by
wireless accelerometers has been improved [40]. Machine learning
algorithms have established the ability to classify the body segment
mounted to a wireless inertial sensor during gait [41].
LeMoyne et al. pioneered the origins of gait analysis for smart-
phones and portable media devices through successful test and
evaluation. Both smartphones and portable media devices share
essentially the same software platform, so an application used on an
iPod can be used on an iPhone and vice versa. The proper device
340 Robert LeMoyne and Timothy Mastroianni

selection is dependent on the operational context [2, 3, 42, 43].

With the capacity to e-mail the acquired accelerometer waveform
for gait analysis, the post-processing and experimentation site can
literally exist on separate continents, as long as Internet connectiv-
ity is possible. For the preliminary experiment during 2010
LeMoyne et al. conducted the experiment in a region thousands of
miles remote to the post-processing site by conveying the acceler-
ometer waveform package by wireless transmission through e-mail.
The smartphone was mounted proximal to the ankle joint for a
subject with healthy gait. The two parameters calculated were the
time-averaged acceleration from stance to stance and the temporal
difference from a stance phase to the following stance phase. The
post-processed gait parameters were acquired with considerable
accuracy and reliability [2].
The smartphone can also be applied to other anatomical
mounting positions for gait, such as the lumbar-sacral aspect of the
spine near the belt area and lateral epicondyle of the femur proxi-
mal to the knee joint. The parameter of interest was the step dura-
tion cycle, which was accurately and reliably obtained [42, 43]. A
similar gait experiment involved mounting a wireless device, such
as an iPod, to the lumbar-sacral aspect of the spine near the belt
area to accurately and reliably acquire step to step duration [44].
Portable media devices were subsequently tested and evaluated
by LeMoyne et al. during 2011 using the same software applica-
tion to evaluate the acceleration waveform of gait through the
smartphone. The gait parameters for stance to stance temporal
duration and time-averaged acceleration between stance to stance
were accurately and reliably derived [3].
Since 2010 smartphones are gaining more presence in the
health care field. A smartphone gait application has been success-
fully compared with respect to accuracy in regard to the conven-
tional triaxial accelerometer for gait quantification [45]. The
inherent abnormalities of rheumatoid arthritis gait have been
assessed by smartphone applications [46]. The Timed Up and Go
test for mobility assessment has been consolidated into a smart-
phone application [47, 48]. Portable and Accurate Gait Analysis
System (PAGAS) provides footswitch data that is conveyed to a
smartphone, enabling subjects to monitor gait status [49].

1.2 Smartphones/ The tendon reflex is a feature that is inherently integrated with gait
Portable Media cycle [13]. The wireless quantified reflex device methodology was
Devices for Tendon developed by LeMoyne et al. [13, 50]. The intrinsic attribute of the
Reflex Response wireless quantified reflex device is the use of wireless accelerometer
Quantification nodes, with one node secured by elastic band proximal to the ankle
joint and eventually another node mounted to the swing arm of the
impact pendulum. A potential energy impact pendulum mounted
to a reflex hammer was incorporated to precisely evoke the patellar
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 341

tendon reflex at a prescribed and accurate setting. The device was

progressively upgraded from a single accelerometer for measuring
patellar tendon maximum reflex response to tandem accelerometers
also capable of measuring patellar tendon reflex latency [7, 13, 50–
57]. Maximum reflex response and latency are considered to be
clinically significant and objectively quantifiable parameters through
the application of the wireless quantified reflex device methodology,
by contrast to traditional subjective assessments of the reflex
response. During the final design iterations, quantified maximum
reflex response and latency were acquired with considerable accu-
racy, reliability, and reproducibility [7, 13, 50, 52].
During 2010 LeMoyne et al. used a smartphone as a wireless
accelerometer to convey the recorded waveform of the reflex
response by e-mail [58]. In 2011 LeMoyne and Mastroianni suc-
cessfully tested and evaluated the same iPhone software application
using an iPod to acquire the patellar tendon reflex response [59].
For both experiments the quantified parameter was the maximum
acceleration of the reflex response, with the both iPod and iPhone
mounted proximal to the ankle through an elastic band. For each
experiment the patellar tendon reflex was evoked supramaximally
by manual strike with a reflex hammer while accurately and reliably
quantifying reflex response [58, 59].
LeMoyne et al. during 2012 integrated the potential energy
impact pendulum scheme with a portable media device as a wire-
less accelerometer platform. The application successfully acquired
the patellar tendon reflex response acceleration waveform by
mounting the portable media device proximal to the lateral malle-
olus by an elastic band, and the maximum acceleration of the reflex
response was accurately and reliably quantified [4].
The smartphone wireless accelerometer application for the
quantification of reflex response was conducted in a rural mountain-
ous setting to demonstrate the capabilities of the wide cell-phone
scale coverage of the smartphone, without the limited footprint of a
wireless Internet zone. The application likewise integrated the
potential impact pendulum for precise targeting and accurate poten-
tial energy settings. A Matlab software program automated the
acquisition of the maximum response of the patellar tendon reflex.
The smartphone wireless accelerometer application for the quantifi-
cation of reflex response quantified the maximal acceleration of the
reflex response with considerable accuracy and reliability [5].

1.3 Smartphones Parkinson’s disease is neurodegenerative in nature with distinctive

for Parkinson’s features regarding movement disorder. Predominant occurrence of
Disease Tremor Parkinson’s disease is for people of 55 years age and older [60].
Quantification Within the USA approximately one million people have been diag-
nosed with Parkinson’s disease. Degenerating dopaminergic neurons
of the substantia nigra are the cause for Parkinson’s disease [61].
342 Robert LeMoyne and Timothy Mastroianni

Degeneration of the substantia nigra leads to a decline in dopamine

synthesis capacity for neural structures, such as the caudate and
putamen [62]. There are four major movement disorder attributes:
compromise to balance, gait shuffling, rigidity, and resting tremor
[61]. Resting tremor rate is about 4 to 5 per second [61, 63].
Voluntary movement can alleviate resting tremor [63]. Three pri-
mary treatment strategies are drug therapy, pallidotomy, and deep
brain stimulation [61, 64, 65].
Regarding the two surgical procedures the pallidotomy is most
extreme, when all other therapy contingencies have been deter-
mined unsuccessful. The pallidotomy is a neurosurgical procedure
that induces a lesion on the globus pallidus internal aspect [61,
64]. Another neurosurgical procedure involves implanting a deep
brain stimulator to stimulate the subthalamic, thalamic, or pallidal
aspects of the brain. The four deep brain stimulator parameters are
the electrode polarity, amplitude, frequency, and pulse width.
A movement disorder specialist is tasked with ascertaining the opti-
mal setting with over a thousand combinations [65]. Quantified
feedback for deep brain stimulation parameters have been recom-
mended through the use of wireless accelerometer devices for effi-
cacy determination and applied regarding accelerometer systems
[1, 66–68].
Conventional Parkinson’s disease treatment has involved drug
therapy, such as L-dopa [63]. Wireless accelerometers have been
proposed for titration of optimal drug therapy. A patient’s unique
acceleration signal could advance diagnostic acuity for medical spe-
cialists to provide optimal treatment of their patient [1, 6].
The acceleration signal enables diagnostic classification of “on”
and “off” states for Parkinson’s disease subjects [69]. Adverse side
effects, such as levodopa-induced dyskinesia, have been identified
by analysis of the acceleration waveform [70]. The objectivity of the
acceleration signal from an accelerometer can facilitate diagnosis
involving contraindicating factors, such as a drug therapy that may
enhance cognitive function at the expense of amplifying movement
disorder, such as tremor [71]. Advanced numerical analysis of the
acceleration waveform, such as spectral analysis, can be applied for
Parkinson’s disease subjects [72]. Other aspects of Parkinson’s dis-
ease can be analyzed by accelerometers, such as gait [73].
The evolution to wireless accelerometer applications has
advanced the objective quantification of Parkinson’s disease char-
acteristics, such as resting tremor. Wireless accelerometers that are
also wearable have been demonstrated for assessing the health sta-
tus of Parkinson’s disease patients [1, 6, 7]. LeMoyne et al. pro-
vided first proof of concept for the wireless accelerometer node to
quantify Parkinson’s hand tremor features in 2009 [7, 74].
Giuffrida et al. later produced a device called Kinesia™ that
integrated motion sensors, such as the accelerometer, that conveyed
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 343

data by wire line to a wrist-mounted command module for wireless

transmission. Kinesia™ provided automated Parkinson’s disease
tremor evaluation [75]. A potential issue with the device is the wir-
ing may encumber the subject or even be dislodged. Another wear-
able and wireless application quantified standard activities, such as
gait, respective of Parkinson’s disease subjects [76–78].
Another breakthrough technology by LeMoyne et al. was the
implementation of the smartphone as a wireless accelerometer
platform. Through the software application, a sample of a subject’s
Parkinson’s disease hand tremor could be recorded as an accelera-
tion waveform and then conveyed as an e-mail to a file. The experi-
mental sampling site and post-processing locations were positioned
thousands of miles remote to demonstrate the robust and autono-
mous nature of the application [1]. The software application could
also be readily applied to portable media devices [79]. Later in
2011 Kostikis et al. applied smartphone technology for quantifying
Parkinson’s disease features [80]. LeMoyne et al. also contrasted a
statically positioned wireless accelerometer node with another
wireless accelerometer node quantifying stimulated tremor with
statistical significance [81].

2 Materials

The iPod and iPhone constitute wireless accelerometer platforms

that have been applied for the quantification of gait, reflex response,
and Parkinson’s disease hand tremor [1–5]. The usage of either the
iPod or iPhone is specific to the context of the application. The
iPod is lighter with a mass of 101 g compared to the iPhone with
a mass of 137 g [1–5]. The iPod requires a wireless Internet zone
for connectivity [3, 4]. The iPhone connectivity to the Internet is
in the range of cell-phone coverage [1, 2, 5]. Rural settings are
even relevant to the capabilities of the iPhone as a wireless acceler-
ometer platform [5].

3 Methods

For specific experimental protocols of implementing the smart-

phone or portable media device for quantifying gait, reflex
response, and Parkinson’s disease hand tremor the following pub-
lications are advised:
1. “Implementation of an iPhone as a wireless accelerometer for
quantifying gait characteristics” [2].
2. “Wireless accelerometer iPod application for quantifying gait
characteristics” [3].
344 Robert LeMoyne and Timothy Mastroianni

3. “Quantified reflex strategy using an iPod as a wireless acceler-

ometer application” [4].
4. “Implementation of an iPhone wireless accelerometer applica-
tion for the quantification of reflex response” [5].
5. “Implementation of an iPhone for characterizing Parkinson’s
disease tremor through a wireless accelerometer application” [1].

3.1 Implementation The smartphone equipped with a wireless accelerometer applica-

of a Smartphone tion for quantifying the characteristics of gait constitutes a wear-
and Portable Media able device, as presented in Fig. 1. Figure 2 represents a 10 s
Device Wireless recording of the smartphone application’s acceleration waveform
Accelerometer of gait, note that each stance event is characterized by a local maxi-
Application for Gait mum regarding the acceleration waveform. The autonomy of the
Quantification devices was demonstrated by conducting the gait experiment in a
hallway setting location thousands of miles remote to the post-
processing site with the trial acceleration waveforms conveyed
wireless via e-mail. The implication is the gait experiment and post-
processing sites can be situated anywhere in the world [2].
The following experimental protocol is advised for the smart-
phone regarding gait quantification:

Fig. 1 Smartphone (iPhone) wireless accelerometer application for quantification

of gait [2]. Reproduced from 2010 with permission from IEEE
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 345

Trial 1: iPhone wireless accelerometer

mounted to lateral malleolus of left ankle
4.5
4
3.5
Acceleration (g's)

3
2.5
2
1.5
1
0.5
0
0 2 4 6 8 10
Time (seconds)

Fig. 2 Gait acceleration waveform from smartphone (iPhone) application [2]. Reproduced from 2010
with permission from IEEE

1. Place the smartphone proximal to the lateral malleolus secured

by an elastic band.
2. Activate the smartphone for data acquisition of the accelera-
tion waveform.
3. Instruct the subject to walk at a self-selected speed for 10 s.
4. Transmit the trial sample wireless through connection to the
Internet as e-mail [2].
Descriptive statistics are presented in Table 1 for two gait
parameters: time-averaged acceleration of the gait cycle from stance
to stance less the static gravity offset and step duration cycle time
from stance to stance. The initial stance to stance cycle was deleted
from the analysis for emphasis of the steady state features of gait.
The gait cycle time-averaged acceleration mean was 0.63 × g with a
standard deviation of 0.05 × g. The time-averaged acceleration of
gait cycle was bound by a 95 % confidence level with a 5 % margin
of error about the mean. Step duration cycle mean was 0.87 s with
a standard deviation of 0.04 s. The step duration cycle was bound
by a 96 % confidence level using a 4 % margin of error about the
mean. Successful proof of concept from an engineering perspective
warrants expanding evaluation, such as a clinical trial [2].
The observations suggest that the clinician could remotely
evaluate the characteristics of a subject’s gait in an autonomous
environment. The progressive evolution of smartphone support-
ing technology is anticipated to facilitate gait analysis and quanti-
fication. Increased data storage and battery lifetime along with
improved mass properties should facilitate longer durations for
gait experiments with minimal encumbrance. Further automation
346 Robert LeMoyne and Timothy Mastroianni

Table 1
Smartphone (iPhone) quantified gait parameters [2]

Time-averaged acceleration (g’s) Step cycle time (s)

Trial number Mean Standard deviation Mean Standard deviation

1 0.59 0.03 0.87 0.04
2 0.63 0.04 0.84 0.03
3 0.58 0.03 0.91 0.02
4 0.68 0.05 0.85 0.03
5 0.62 0.06 0.83 0.03
6 0.64 0.02 0.87 0.01
7 0.66 0.05 0.88 0.03
8 0.64 0.04 0.90 0.03
9 0.64 0.03 0.87 0.02
10 0.63 0.04 0.89 0.02
Global (all 10 trials) 0.63 0.05 0.87 0.04
Reproduced from 2010 with permission from IEEE

of software processes could minimize the need to remove the

smartphone from its mounting position, such as manual interac-
tion for e-mailing. Machine learning algorithms may advance
diagnostic acuity. The software capabilities of the smartphone may
be applied to also provide real-time gait feedback and modifica-
tion [2, 3]. “Virtual Proprioception” incorporating leg mounted
wireless accelerometers could enable real-time gait feedback and
modification [7, 19, 20].
The portable media device quantified gait as a wireless acceler-
ometer platform. The gait experiment was conducted thousands of
miles remote from the post-processing site, and the gait data was
transmitted wirelessly as e-mail through Internet connectivity. The
remote distance between the gait quantification resources empha-
sizes the functional autonomy of the application. An experimental
protocol that was essentially identical to the smartphone gait quan-
tification was used for the portable media device regarding gait
quantification [3].
The portable media device wireless accelerometer exhibits the
ability to obtain quantified gait parameters with a high level of
accuracy and consistency. The gait cycle time-averaged acceleration
mean was 1.42 × g, and the standard deviation was 0.04 × g. The
mean for the step duration cycle was 1.10 s, and the standard devi-
ation was 0.05 s. Both step cycle time from stance to stance and the
stance to stance time-averaged acceleration of the gait cycle were
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 347

Fig. 3 Portable media device (iPod) wireless accelerometer reflex quantification

system with iPod mounted proximal to the lateral malleolus [4]. Reproduced from
2012 with permission from IEEE

bound with a 96 % confidence level with a 4 % margin of error

about the mean. The portable media device wireless accelerometer
application exemplifies the future utility of wearable devices for
functionally autonomous quantification of gait. Efficacious proof
of concept from an engineering perspective implicates expanded
future test and evaluation of the portable media device as a gait
quantification device [3].

3.2 Implementation Figure 3 illustrates the wireless quantified reflex system that applies
of a Portable Media the portable media device as a wireless accelerometer platform. The
Device anatomical mounting strategy is also exemplified in Fig. 3 with the
and Smartphone portable media device mounted proximal to the ankle and secured
for Quantification by the elastic band of a sock. The potential energy reflex input sys-
of Tendon Reflex tem demonstrates the capacity to accurately and reliably target the
Response patellar tendon with precise and discrete settings of potential energy.
For the test and evaluation experiment of the wireless reflex quanti-
fication system with a portable media device, a potential energy set-
ting of 30° relative to gravity vector was selected [4].
The goal of the portable media device wireless accelerometer
platform for reflex quantification was to provide a simplified alter-
native for quantifying the patellar tendon reflex response.
348 Robert LeMoyne and Timothy Mastroianni

The quantified reflex strategy device alleviates the requirement for

highly specialized resources. Targeting of the patellar tendon and
eliciting of the reflex with discrete levels of potential energy pro-
vides a simplified alternative to traditional methods for evoking the
patellar tendon reflex. The device acquired an acceleration wave-
form of the reflex response, which was conveyed by wireless trans-
mission through e-mail as a file for post-processing [4].
The following experimental protocol is advised for using the
portable media device to quantify the reflex response:
1. Secure the device to proximal to the lateral malleolus through
an elastic band, such as a sock.
2. Target the reflex hammer of the impact pendulum level to the
tibial tubercle.
3. Activate the device application to record the acceleration wave-
form response.
4. Retract the swing arm to a predetermined potential energy set-
ting, such as 30°.
5. Release the swing arm.
6. Transmit the acquired acceleration waveform as an e-mail
attachment by wireless Internet connectivity.
7. Wait a minimum 15 s before conducting the next trial [4].
Based on post-processing for previous wireless quantified reflex
studies, the major parameter was the maximum reflex response,
which was normalized to gravity [51–53]. Figure 4 illustrates a
representative acceleration waveform of the patellar tendon reflex

Fig. 4 Portable media device (iPod) wireless accelerometer reflex quantification system acceleration waveform
of the reflex response [4]. Reproduced from 2012 with permission from IEEE
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 349

Table 2
Portable media device (iPod) wireless accelerometer device maximum reflex response (30 trials) [4]

Parameter Mean Standard deviation Coefficient of variation

Maximum reflex response 2.04 × g 0.18 × g 0.09
Reproduced from 2012 with permission from IEEE

response. Descriptive statistics for the maximum reflex response

parameter regarding the 30 trials are presented in Table 2 [4].
The 30 trials quantifying the patellar tendon reflex response
through a software application demonstrated considerable accu-
racy and reliability. The maximum reflex response had a mean of
2.04 × g, a standard deviation of 0.18 × g, and a coefficient of varia-
tion of 0.09. With regard to the 30 trial sample size, the maximum
reflex response was bound with a 96 % confidence level with a 4 %
margin of error about the mean [4].
Capability to convey an acceleration waveform of the patellar
tendon reflex response by wireless transmission through e-mail
enables evaluation and post-processing sites to exist at remote dis-
tances. The application requires wireless connectivity to a local
Internet zone. Successfully engineering proof of concept merits
expanded testing and evaluation of the portable media device wire-
less accelerometer reflex quantification system. The device’s wire-
less capabilities for quantifying reflex strategy may facilitate future
advances as a preliminary neurological assessment [4].
The primary theme for the smartphone wireless accelerometer
reflex quantification system is to minimize specialized resource
constraints. With the objective of demonstrating the smartphone
application’s broad access to the Internet, the reflex quantification
experiment was conducted in a rural mountainous location. The
reflex response acceleration waveform was transmitted through
e-mail by wireless connectivity to the Internet. Data post-processing
was conducted in a city environment and facilitated by a Matlab
program. The Matlab program performed automated feature
extraction for the maximum acceleration of the reflex response.
The automation software provides substantial process improve-
ment by reducing the time to acquire the desired data features. An
experimental protocol similar to the portable media device was
applied for using the smartphone to quantify the reflex response,
which was nearly identical to the implementation of the portable
media device for reflex response quantification [5].
The maximum response of the tendon reflex acceleration
waveform demonstrated considerable accuracy and consistency for
the span of the experiment’s 30 trials. The patellar tendon reflex
response maximum acceleration had a mean of 1.89 × g, a standard
deviation of 0.16 × g, and a coefficient of variation of 0.08. Based
350 Robert LeMoyne and Timothy Mastroianni

on a sample size of 30 trials, the maximum acceleration of the

reflex response was bound with a 96 % confidence level with a 4 %
margin of error about the mean [5].
Future advances for the portable media device and smartphone
envision evolutions in software applications, data storage, operation
length, and mass properties, which should improve the capability to
quantify the features of tendon reflex response. Advances regarding
portable media device and smartphone software applications can
include the ability to record the gyroscope signal for greater insight
regarding the tendon reflex response. The implementation of
machine learning algorithms may advance diagnostic acuity of neu-
rological status or even derive a functional latency based on the
feature extraction of the tendon reflex response [4, 5].

3.3 Implementation Figure 5 illustrates the functional simplicity for using the smart-
of a Smartphone phone application to quantify Parkinson’s disease hand tremor.
for Quantifying The smartphone is secured to the dorsum of the hand by a glove.
Parkinson’s Disease The experiment and post-processing were conducted in remote
Tremor locations. The experimental test and evaluation of the application
involved a subject with Parkinson’s disease and a subject without
Parkinson’s disease representing a control. The setting may be
selected at the convenience of the subject. The recorded acceler-
ometer signal for both subjects was conveyed by wireless
transmission through e-mail. Post-processing was conducted thou-
sands of miles remote to the experimentation site [1].

Fig. 5 Smartphone (iPhone) wireless accelerometer application for quantifying

Parkinson’s disease tremor, mounted to the dorsum of the hand by a glove [1].
Reproduced from 2010 with permission from IEEE
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 351

iPhone wireless accelerometer signal characteristics of

subject without Parkinson's disease
1.4
1.3
Acceleration (g's)

1.2
1.1
1
0.9
0.8
0.7
0.6
0 2 4 6 8 10 12
Time (seconds)

Fig. 6 Acceleration waveform of static hand condition for subject without Parkinson’s disease [1]. Reproduced
from 2010 with permission from IEEE

The following experimental protocol is advised for using the

smartphone to quantify Parkinson’s disease hand tremor:
1. Secure the smartphone to the dorsum of the subject’s hand
with a glove.
2. Have the subject comfortably sit; and support the forearm on
a pad above a table. Ensure that the padding is thick enough to
prevent the hand from contacting the table.
3. Activate the device application to record the acceleration wave-
form, with the subject’s hand relaxed.
4. Wirelessly transmit the acceleration waveform as an e-mail
attachment by connectivity to the Internet [1].
The control subject demonstrated a statically positioned accel-
eration waveform as exemplified by Fig. 6, with minimal variation
from a steady 1 × g of acceleration. By contrast the Parkinson’s dis-
ease subject revealed a fluctuating acceleration waveform, due to
hand tremor, as illustrated in Fig. 7. For the acceleration waveform
of each trial set time-averaged acceleration less the gravity offset
constituted the quantified parameter [1].
For the control subject the mean time-averaged acceleration
was 8.0 μ-gravities with a standard deviation of 0.6 μ-gravities. The
coefficient of variation was 0.075. Spectral analysis of a representa-
tive trial was presented in Fig. 8, which implies minimal frequency
content for the acceleration waveform. The time-averaged
acceleration of the control subject was bound with a 95 % confi-
dence level using a 5 % margin of error about the mean [1].
352 Robert LeMoyne and Timothy Mastroianni

iPhone wireless accelerometer signal characteristics of

subject with Parkinson's disease
1.4
1.3
Acceleration (g's)

1.2
1.1
1
0.9
0.8
0.7
0.6
0 2 4 6 8 10 12
Time (seconds)

Fig. 7 Acceleration waveform of hand tremor attributes for subject with Parkinson’s disease [1]. Reproduced
from 2010 with permission from IEEE

Frequency response of subject without Parkinson's disease tremor using Blackman window
10

6
Amplitude

0
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)

Fig. 8 Frequency response with Blackman window for the subject without Parkinson’s disease tremor [1].
Reproduced from 2010 with permission from IEEE

The Parkinson’s disease subject exhibited notable hand tremor

with a mean time-averaged acceleration of 19.1 μ-gravities and a
standard deviation of 5.8 μ-gravities. The coefficient of variation
for the Parkinson’s disease subject was 0.307. The mean time-
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 353

Frequency response of subject with Parkinson's disease tremor using Blackman window
10

6
Amplitude

0
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)

Fig. 9 Frequency response with Blackman window for the subject with Parkinson’s disease tremor [1].
Reproduced from 2010 with permission from IEEE

averaged acceleration 2.4 times was greater than the control

subject, and the coefficient of variation was 4.1 times greater than
the control subject. The Parkinson’s disease subject revealed nota-
ble frequency content in Fig. 9, with notable frequencies at 5.3,
7.7, and 10.4 Hz [1].
Inferential statistical analysis was applied to further contrast
the two subjects. A two group independent t-test with unequal
variances with alpha < 0.05 was applied. The time-averaged accel-
eration of the subject with Parkinson’s disease tremor was found to
be statistically significant by comparison to the static control of a
subject without Parkinson’s disease. The subject with Parkinson’s
disease and the subject without Parkinson’s disease were estab-
lished as two independent groups [1].
The results establish the efficacy of the smartphone wireless
accelerometer application to quantify the characteristics of
Parkinson’s disease hand tremor. Future assessment of Parkinson’s
disease hand tremor could be applied over an extended span of
time. For example the cyclical features of Parkinson’s disease may
be sufficiently quantified using spectral analysis [1].
The nature of the demonstrated smartphone wireless acceler-
ometer application for quantifying Parkinson’s disease tremor con-
firms the ability to accomplish remote diagnostics. The acceleration
waveform was transmitted wireless through e-mail. Therefore
medical resources and subject location can literally exist at any
354 Robert LeMoyne and Timothy Mastroianni

remote distance from each other. With the acceleration waveform

hand tremor data as feedback, the medical resources could advise a
medication therapy strategy from anywhere else in the world [1].
Future evolution of the smartphone are projected regarding
hardware capabilities, such as processing speed, storage capacity,
mass, and battery operational duration. Software capabilities are
also predicted to advance, such as the incorporation of machine
learning algorithms, which may advance diagnostics capability. For
example deep brain stimulator parameter settings may be opti-
mized through the smartphone wireless accelerometer application
acceleration waveform as feedback. Software applications can be
developed for personal use or with the assistance of a caregiver [1].
Another platform for quantifying the acceleration waveform
was envisioned for application with the aid of a caregiver. The plat-
form involved two wireless accelerometer nodes with one node
placed in a static position and the other node mounted to the dor-
sum of the hand by a glove. Test and evaluation involved a static
control and a simulated hand tremor, and statistical significance of
the quantified time average of the acceleration waveform was
achieved [81].

4 Notes

The future holds considerable promise for the use of smartphones

and portable media devices as wearable devices, such as a wireless
accelerometer platform. Advances in software and hardware tech-
nology capabilities are anticipated to substantially impact the use
of these devices for the biomedical and healthcare industries. Gait
quantification and patellar tendon reflex response features have
been demonstrated with considerable accuracy and reliability
through the use of both smartphones and portable media devices
as a wireless accelerometer platform. The Parkinson’s disease hand
tremor was quantified with statistical significance compared to a
non-Parkinson’s disease control. The acceleration waveforms
acquired by smartphones and portable media devices were con-
veyed by wireless transmission as an e-mail attachment through
Internet connectivity. The implication of these experimental
methodologies is that the site for experimentation and post-pro-
cessing can be remote and literally anywhere in the world. Based
on the methodologies presented, any aspect of human movement
may be obtained through such devices as a wireless acceleration
platform. Future applications envision the use of the gyroscope
sensor on both devices for a new perspective for quantifying
human movement features.
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 355

Acknowledgement

The author would like to thank IEEE for granting permission to

reuse the content (Figs. 1, 2, 3, 4, 5, 6, 7, 8, and 9 and Tables 1
and 2) of refs. 1, 2, and 4. I would personally like to acknowledge
the contributions of Dr. Grundfest of UCLA Department of
Bioengineering, as his insight and expertise served an instrumental
role for the advance of smartphones and portable media devices as
a wireless accelerometer platform for the quantification of gait,
tendon reflex response, and Parkinson’s disease hand tremor. I
would like to extend my appreciation to Kevin Zanjani, Michael
Minicozzi, and Anthony Hessel for their assistance with the final
preparation of the manuscript.

References
1. LeMoyne R, Mastroianni T, Cozza M, Coroian applications in biofeedback gait rehabilitation
C, Grundfest W (2010) Implementation of an concepts and movement disorder characteriza-
iPhone for characterizing Parkinson’s disease tion. In: Barros de Mello CA (ed) Biomedical
tremor through a wireless accelerometer appli- engineering. Vienna, Austria, Intech, Ch 10
cation. In Proceedings of the 32nd Annual 8. Culhane KM, O’Connor M, Lyons D, Lyons GM
International Conference of the IEEE EMBS, (2005) Accelerometers in rehabilitation medicine
pp 4954–4958 for older adults. Age Ageing 34:556–560
2. LeMoyne R, Mastroianni T, Cozza M, Coroian 9. Jovanov E, Milenkovic A, Otto C, de Groen
C, Grundfest W (2010) Implementation of an PC (2005) A wireless body area network of
iPhone as a wireless accelerometer for quantify- intelligent motion sensors for computer
ing gait characteristics. In Proceedings of the assisted physical rehabilitation. J Neuroeng
32nd Annual International Conference of the Rehabil 2:6
IEEE EMBS, pp 3847–3851 10. Saremi K, Marehbian J, Yan X, Regnaux JP,
3. LeMoyne R, Mastroianni T, Grundfest W Elashoff R, Bussel B, Dobkin BH (2006)
(2011) Wireless accelerometer iPod applica- Reliability and validity of bilateral thigh and
tion for quantifying gait characteristics. In foot accelerometry measures of walking in
Proceedings of the 33rd Annual International healthy and hemiparetic subjects. Neurorehabil
Conference of the IEEE EMBS, Neural Repair 20:297–305
pp 7904–7907 11. Kavanagh JJ, Morrison S, James DA, Barrett R
4. LeMoyne R, Mastroianni T, Grundfest W, (2006) Reliability of segmental accelerations
(2012) Quantified reflex strategy using an iPod measured using a new wireless gait analysis sys-
as a wireless accelerometer application. In tem. J Biomech 39:2863–2872
Proceedings of 34th International Conference 12. Kavanagh J, Barrett R, Morrison S (2006) The
of the IEEE EMBS, pp 2476–2479 role of the neck and trunk in facilitating head
5. LeMoyne R, Mastroianni T, Grundfest W, stability during walking. Exp Brain Res
Nishikawa K (2013) Implementation of an 172:454–463
iPhone wireless accelerometer application for 13. LeMoyne R, Mastroianni T, Coroian C,
the quantification of reflex response. In Grundfest W (2011) Tendon reflex and strate-
Proceedings of 35th International Conference gies for quantification, with novel methods
of the IEEE EMBS, pp 4658–4661 incorporating wireless accelerometer reflex
6. Patel S, Park H, Bonato P, Chan L, Rodgers M quantification devices, a perspective review. J
(2012) A review of wearable sensors and sys- Mech Med Biol 11:471–513
tems with application in rehabilitation. J 14. Lee JA, Cho SH, Lee JW, Lee KH, Yang HK
Neuroeng Rehabil 9:21 (2007) Wearable accelerometer system for
7. LeMoyne R, Coroian C, Mastroianni T, measuring the temporal parameters of gait. In
Opalinski P, Cozza M, Grundfest W (2009) Proceedings of the 29th Annual International
The merits of artificial proprioception, with Conference of the IEEE EMBS, pp 483–486
356 Robert LeMoyne and Timothy Mastroianni

15. Lee JA, Cho SH, Lee YJ, Yang HK, Lee JW 27. Kavanagh JJ (2009) Lower trunk motion and
(2010) Portable activity monitoring system for speed-dependence during walking. J Neuroeng
temporal parameters of gait cycles. J Med Syst Rehabil 6:9
34:959–966 28. Lai DT, Charry E, Begg R, Palaniswami M
16. Bamberg SJ, Benbasat AY, Scarborough DM, (2008) A prototype wireless inertial-sensing
Krebs DE, Paradiso JA (2008) Gait analysis device for measuring toe clearance. In Proceedings
using a shoe-integrated wireless sensor system. of the 30th Annual International Conference of
IEEE Trans Inf Technol Biomed 12:413–423 the IEEE EMBS, pp 4899–4902
17. LeMoyne R, Coroian C, Mastroianni T, 29. Reininga IH, Stevens M, Wagenmakers R,
Grundfest W (2009) Wireless accelerometer Bulstra SK, Groothoff JW, Zijlstra W (2012)
assessment of gait for quantified disparity of Subjects with hip osteoarthritis show distinc-
hemiparetic locomotion. J Mech Med Biol tive patterns of trunk movements during gait-a
9:329–343 body-fixed-sensor based analysis. J Neuroeng
18. LeMoyne R, Coroian C, Mastroianni T (2009) Rehabil 9:3
Wireless accelerometer system for quantifying 30. Tura A, Raggi M, Rocchi L, Cutti AG, Chiari L
gait. In Proceedings of the IEEE/ICME (2010) Gait symmetry and regularity in trans-
International Conference on Complex Medical femoral amputees assessed by trunk accelera-
Engineering (CME), pp 1–4 tions. J Neuroeng Rehabil 7:4
19. LeMoyne R, Coroian C, Mastroianni T, 31. Tura A, Rocchi L, Raggi M, Cutti AG, Chiari L
Grundfest W (2008) Virtual proprioception. J (2012) Recommended number of strides for
Mech Med Biol 8:317–338 automatic assessment of gait symmetry and
20. LeMoyne R, Coroian C, Mastroianni T, Wu W, regularity in above-knee amputees by means of
Grundfest W, Kaiser W (2008) Virtual proprio- accelerometry and autocorrelation analysis.
ception with real-time step detection and pro- J Neuroeng Rehabil 9:11
cessing. In Proceedings of the 30th Annual 32. Hsu CC, Chen JH (2011) A novel sensor-
International Conference of the IEEE EMBS, assisted RFID-based indoor tracking system
pp 4238–4241 for the elderly living alone. Sensors
21. Mizuike C, Ohgi S, Morita S (2009) Analysis 11:10094–10113
of stroke patient walking dynamics using a tri- 33. Yeoh WS, Pek I, Yong YH, Chen X, Waluyo AB
axial accelerometer. Gait Posture 30:60–64 (2008) Ambulatory monitoring of human pos-
22. Guo Y, Wu D, Liu G, Zhao G, Huang B, Wang ture and walking speed using wearable acceler-
L (2012) A low-cost body inertial-sensing net- ometer sensors. In Proceedings of the 30th
work for practical gait discrimination of hemi- Annual International Conference of the IEEE
plegia patients. Telemed J E Health EMBS, pp 5184–5187
18:748–754 34. Choquette S, Hamel M, Boissy P (2008)
23. Prajapati SK, Gage WH, Brooks D, Black SE, Accelerometer-based wireless body area net-
McIlroy WE (2011) A novel approach to work to estimate intensity of therapy in post-
ambulatory monitoring: investigation into the acute rehabilitation. J Neuroeng Rehabil 5:20
quantity and control of everyday walking in 35. Hurkmans HL, Ribbers GM, Streur-
patients with subacute stroke. Neurorehabil Kranenburg MF, Stam HJ, van den Berg-
Neural Repair 25:6–14 Emons RJ (2011) Energy expenditure in
24. Bugané F, Benedetti MG, Casadio G, Attala S, chronic stroke patients playing Wii Sports: a
Biagi F, Manca M, Leardini A (2012) pilot study. J Neuroeng Rehabil 8:38
Estimation of spatial-temporal gait parameters 36. Huang H, Wolf SL, He J (2006) Recent devel-
in level walking based on a single accelerome- opments in biofeedback for neuromotor reha-
ter: validation on normal subjects by standard bilitation. J Neuroeng Rehabil 3:11
gait analysis. Comput Methods Programs 37. McGregor SJ, Armstrong WJ, Yaggie JA,
Biomed 108:129–137 Bollt EM, Parshad R, Bailey JJ, Johnson SM,
25. Watanabe T, Saito H, Koike E, Nitta K (2011) Goin AM, Kelly SR (2011) Lower extremity
A preliminary test of measurement of joint fatigue increases complexity of postural con-
angles and stride length with wireless inertial trol during a single-legged stance. J Neuroeng
sensors for wearable gait evaluation system. Rehabil 8:43
Comput Intell Neurosci 2011: 975193 38. Dozza M, Chiari L, Chan B, Rocchi L, Horak
26. Djurić-Jovičić MD, Jovičić NS, Popović DB FB, Cappello A (2005) Influence of a portable
(2011) Kinematics of gait: new method for audio-biofeedback device on structural
angle estimation based on accelerometers. properties of postural sway. J Neuroeng
Sensors 11:10571–10585 Rehabil 2:13
 Use of Smartphones and Portable Media Devices for Quantifying Human Movement… 357

39. Lee BC, Kim J, Chen S, Sienko KH (2012) 50. LeMoyne R (2010) Wireless quantified reflex
Cell phone based balance trainer. J Neuroeng device, Ph.D. dissertation, UCLA, Biomed.
Rehabil 9:10 Eng. IDP
40. Djurić-Jovičić MD, Jovičić NS, Popović DB, 51. LeMoyne R, Coroian C, Mastroianni T,
Djordjević AR (2012) Nonlinear optimization Grundfest W (2008) Quantified deep tendon
for drift removal in estimation of gait kinemat- reflex device for response and latency, third
ics based on accelerometers. J Biomech generation. J Mech Med Biol 8:491–506
45:2849–2854 52. LeMoyne R, Mastroianni T, Kale H, Luna J,
41. Weenk D, van Beijnum BJ, Baten CT, Hermens Stewart J, Elliot S, Bryan F, Coroian C,
HJ, Veltink PH (2013) Automatic identifica- Grundfest W (2011) Fourth generation wire-
tion of inertial sensor placement on human less reflex quantification system for acquiring
body segments during walking. J Neuroeng tendon reflex response and latency. J Mech
Rehabil 10:31 Med Biol 11:31–54
42. LeMoyne R, Mastroianni T, Cozza M, Coroian 53. LeMoyne R, Coroian C, Mastroianni T, (2009)
C (2010) Quantification of gait characteristics Wireless accelerometer reflex quantification
through a functional iPhone wireless acceler- system characterizing response and latency. In
ometer application mounted to the spine. In Proceedings of 31st Annual International
Proceedings of ASME 2010 5th Frontiers in Conference of the IEEE EMBS,
Biomedical Devices Conference (BioMed), pp 5283–5286
pp 87–88 54. LeMoyne R, Jafari R (2006) Quantified deep
43. LeMoyne R, Mastroianni T, Cozza M, Coroian tendon reflex device, second generation. In
C (2010) iPhone wireless accelerometer appli- Proceedings of 15th International Conference
cation for acquiring quantified gait attributes. on Mechanics in Medicine and Biology
In Proceedings of ASME 2010 5th Frontiers in ICMMB-15 2006, 6–8 December 2006,
Biomedical Devices Conference (BioMed), Furama Riverfront, Singapore
pp 19–20 55. LeMoyne R, Dabiri F, Jafari R (2008)
44. LeMoyne R, Mastroianni T (2012) iWalk, a gait Quantified deep tendon reflex device, second
analysis device incorporating an iPod wireless generation. J Mech Med Biol 8:75–85
accelerometer application. In Proceedings of 56. LeMoyne R, Jafari R, Jea D (2005) Fully quan-
42nd Society for Neuroscience Annual Meeting, tified evaluation of myotatic stretch reflex. In
New Orleans, LA, October 13–17, 2012 Proceedings of 35th Society for Neuroscience
45. Nishiguchi S, Yamada M, Nagai K, Mori S, Annual Meeting, Washington, DC, November
Kajiwara Y, Sonoda T, Yoshimura K, Yoshitomi 12–16, 2005
H, Ito H, Okamoto K, Ito T, Muto S, Ishihara 57. LeMoyne R., Dabiri F., Coroian C.,
T, Aoyama T (2012) Reliability and validity of Mastroianni T., Grundfest W. (2007)
gait analysis by android-based smartphone. Quantified deep tendon reflex device for assess-
Telemed J E Health 18:292–296 ing response and latency. In Proceedings of
46. Yamada M, Aoyama T, Mori S, Nishiguchi S, 37th Society for Neuroscience Annual Meeting,
Okamoto K, Ito T, Muto S, Ishihara T, San Diego, CA, November 3–7, 2007
Yoshitomi H, Ito H (2012) Objective assess- 58. LeMoyne R., Coroian C., Mastroianni T.,
ment of abnormal gait in patients with rheuma- Cozza M., Grundfest W. (2010) Quantification
toid arthritis using a smartphone. Rheumatol of reflex response through an iPhone wireless
Int 32:3869–3874 accelerometer application. In Proceedings of
47. Mellone S, Tacconi C, Chiari L (2012) Validity 40th Society for Neuroscience Annual Meeting,
of a smartphone-based instrumented timed up San Diego, CA, November 13–17, 2010
and go. Gait Posture 36:163–165 59. LeMoyne R., Mastroianni T. (2011) Reflex
48. Palmerini L, Mellone S, Rocchi L, Chiari L response quantification using an iPod wireless
(2011) Dimensionality reduction for the quan- accelerometer application. In Proceedings of
titative evaluation of a smartphone-based timed 41st Society for Neuroscience Annual Meeting,
up and go test. In Proceedings of the 33rd Washington, DC, November 12–16, 2011
Annual International Conference of the IEEE 60. Seeley RR, Stephens TD, Tate P (2003)
EMBS, pp 7179–7182 Anatomy and physiology. McGraw-Hill,
49. Wagner R, Ganz A (2012) PAGAS: portable New York, NY
and accurate gait analysis system. In 61. Kandel ER, Schwartz JH, Jessell TM (2000)
Proceedings of the 34th Annual International Principles of neural science. McGraw-Hill,
Conference of the IEEE EMBS, pp 280–283 New York, NY
358 Robert LeMoyne and Timothy Mastroianni

62. Diamond MC, Scheibel AB, Elson LM (1985) mated, at-home assessment of mobility among
The human brain coloring book. Harper patients with Parkinson disease, using a body-
Perennial, New York, NY worn accelerometer. Neurorehabil Neural
63. Bickley LS, Szilagyi PG (2003) Bates’ guide to Repair 25:810–818
physical examination and history taking. 74. LeMoyne R, Coroian C, Mastroianni T (2009)
Lippincott Williams and Wilkins, New York, NY Quantification of Parkinson’s disease character-
64. Nolte J, Sundsten JW (2002) The human istics using wireless accelerometers. In
brain: an introduction to its functional anat- Proceedings of the IEEE/ICME International
omy. Mosby, St. Louis, MO Conference on Complex Medical Engineering
65. Volkmann J, Moro E, Pahwa R (2006) Basic (CME), pp 1–5
algorithms for the programming of deep brain 75. Giuffrida JP, Riley DE, Maddux BN, Heldman
stimulation in Parkinson’s disease. Mov Disord DA (2009) Clinically deployable Kinesia tech-
21:S284–S289 nology for automated tremor assessment. Mov
66. Rouse AG, Stanslaski SR, Cong P, Jensen RM, Disord 24:723–730
Afshar P, Ullestad D, Gupta R, Molnar GF, 76. Cancela J, Pansera M, Arredondo MT, Estrada
Moran DW, Denison TJ (2011) A chronic gen- JJ, Pastorino M, Pastor-Sanz L, Villalar JL
eralized bi-directional brain-machine interface. (2010) A comprehensive motor symptom
J Neural Eng 8:1–36 monitoring and management system: the bra-
67. Obwegeser AA, Uitti RJ, Witte RJ, Lucas JA, dykinesia case. In Proceedings of the 32nd
Turk MF, Wharen RE Jr (2001) Quantitative Annual International Conference of the IEEE
and qualitative outcome measures after tha- EMBS, pp 1008–1011
lamic deep brain stimulation to treat disabling 77. Pastorino M, Cancela J, Arredondo MT,
tremors. Neurosurgery 48:274–281 Pansera M, Pastor-Sanz L, Villagra F, Pastor
68. Kumru H, Summerfield C, Valldeoriola F, MA, Martin JA (2011) Assessment of bradyki-
Valls-Solé J (2004) Effects of subthalamic nesia in Parkinson’s disease patients through a
nucleus stimulation on characteristics of EMG multi-parametric system. In Proceedings of the
activity underlying reaction time in Parkinson’s 33rd Annual International Conference of the
disease. Mov Disord 19:94–100 IEEE EMBS, pp 1810–1813
69. Keijsers NL, Horstink MW, Gielen SC (2006) 78. Cancela J, Pastorino M, Arredondo MT,
Ambulatory motor assessment in Parkinson’s Pansera M, Pastor-Sanz L, Villagra F, Pastor
disease. Mov Disord 21:34–44 MA, Gonzalez AP (2011) Gait assessment in
70. Keijsers NL, Horstink MW, van Hilten JJ, Hoff Parkinson’s disease patients through a network
JI, Gielen CC (2000) Detection and assess- of wearable accelerometers in unsupervised
ment of the severity of levodopa-induced dys- environments. In Proceedings of the 33rd
kinesia in patients with Parkinson’s disease by Annual International Conference of the IEEE
neural networks. Mov Disord 15:1104–1111 EMBS, pp 2233–2236
71. Gurevich TY, Shabtai H, Korczyn AD, Simon 79. www.apple.com
ES, Giladi N (2006) Effect of rivastigmine on 80. Kostikis N, Hristu-Varsakelis D, Arnaoutoglou
tremor in patients with Parkinson’s disease and M, Kotsavasiloglou C, Baloyiannis S (2011)
dementia. Mov Disord 21:1663–1666 Towards remote evaluation of movement dis-
72. Schrag A, Schelosky L, Scholz U, Poewe W orders via smartphones. In Proceedings of the
(1999) Reduction of Parkinsonian signs in 33rd Annual International Conference of the
patients with Parkinson’s disease by dopami- IEEE EMBS, pp 5240–5243
nergic versus anticholinergic single-dose chal- 81. LeMoyne R, Mastroianni T, Grundfest W
lenges. Mov Disord 14:252–255 (2013) Wireless accelerometer configuration
73. Weiss A, Sharifi S, Plotnik M, van Vugt JP, for monitoring Parkinson’s disease hand
Giladi N, Hausdorff JM (2011) Toward auto- tremor. Adv Parkinson Dis 2:62–67
 Chapter 24

Measuring Tremor with a Smartphone

Benoit Carignan, Jean-François Daneault,
and Christian Duval

Abstract
Tremor is the most common movement disorder. However; characterizing it in large populations is not easily
accomplished since current methodologies are not adapted to large-scale field studies. To overcome this chal-
lenge, a smartphone application was developed as a stand-alone platform to assess tremor. The current book
chapter details the steps taken to validate this mobile application. Data recorded with the smartphone was
analyzed online and offline as well as compared to laboratory equipment and a clinical scale. This allowed for
the identification of the tremor properties that could reliably be characterized with the smartphone as well as
the limits of the hardware. It also allowed for the identification of tasks that could be performed with the
smartphone when tremor was being assessed. Finally, we confirmed the clinical relevance of the results
provided by the smartphone application.

Key words Recording, Assessment, Monitoring, Parkinson, Essential tremor, Multiple sclerosis,
Mobile, Health, Stand-alone, Telephone

1 Introduction

1. Tremor is the most common movement disorder. It is the pre-

dominant symptom in Essential tremor [1, 2] and is a cardinal
sign of Parkinson’s disease [3]. It can also be observed in
patients with other neurological disorders such as multiple
sclerosis [4, 5] or even traumatic brain injury [6, 7]. There are
currently several methods used to quantify tremor which have
been used in different settings. For instance, the assessment of
tremor in a laboratory can be performed using accelerometers
[8, 9] or laser displacement sensors [10–15]. While these
methods provide excellent characterization of tremor, they
require specialized equipment and trained personnel. On the
other hand, clinical scales, which are used by clinicians to eval-
uate and monitor tremor in patient populations, do not require
the use of such specialized equipment. They do however,
require trained medical personnel to perform the test, and they

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_24, © Springer Science+Business Media New York 2015

359
360 Benoit Carignan et al.

lack adequate resolution to properly track variations in tremor

over time. In addition, both laboratory and clinical assessments
require patients to go to a testing facility. In recent years, por-
table tremor evaluation devices have been developed to coun-
teract this issue [16, 17]. However, these devices are not
ubiquitous, they require that the test be performed by trained
personnel, and they are expensive. One device that is becom-
ing omnipresent in the population that has all the required
hardware and software capabilities necessary for tremor assess-
ment is the smartphone. The great majority are equipped with
accelerometers, gyroscopes and/or magnetometers which pro-
vide measures of change in position and orientation. They also
possess large data logging capability, flexible programming,
high resolution interface, and the ability to transmit data
remotely. They are recognized as having the potential to
become efficient data collection and analysis platforms to assess
and monitor the health of individuals outside the clinic [18];
provided the applications have been adequately validated [19].
In the current chapter, we present the methodology used to
validate a stand-alone application for tremor assessment using
a Blackberry® smartphone against a laboratory accelerometer
and a clinical scale [20].

2 Materials

1. Blackberry® Storm™ 9530 smartphone (Research In Motion

LTD, Waterloo, Canada) (see Note 1).
2. Blackberry® Java SDK.
(a) A multi-language workspace; we used Eclipse (see Note 2).
(b) An integrated development environment; we used
NetBeans IDE (see Note 3).
3. Blackberry® smartphone simulator for the Blackberry® Storm™
9530.
4. Laboratory tremor recording system, which includes a uniaxial
accelerometer and signal conditioner (482B11; PCB
Piezotronics, USA) (see Note 4).

3 Methods

3.1 Getting Access 1. Next are general guidelines to get access to the smartphone’s
to the Smartphone’s hardware and software. Specific codes and functions differ
Hardware between platforms such as Blackberry®, iPhone®, Android™,
and Software or Windows Phone® as well as between software versions such
as Blackberry® 7 and Blackberry® 10 which would make it
futile for us to provide the code to access the hardware and
 Measuring Tremor with a Smartphone 361

software components. Instead, we propose that the following

instructions will provide a guide for the implementation of
future mobile applications for tremor assessment on several
different platforms. The appropriate code to implement the
instructions on a given platform and software version can eas-
ily be found in the developer’s programming documentation
(see Note 5).
2. Accessing the smartphone’s hardware and software requires
programming in Java (see Note 6) using the multi-language
workspace and integrated development environment.
(a) First, there needs to be a decision on whether the acceler-
ometer will be queried solely when the application is in the
foreground or whether it can also be queried when the
application is in the background. We opted for querying
the accelerometer only when it was in the foreground as it
uses less battery and the goal of our application was not to
perform long-term tremor recordings.
For this, we needed to import the classes related to the
accelerometer channel and accelerometer sensor. We then
needed to open a channel to retrieve the accelerometer
data. Then, we specified that the accelerometer data was
gathered continuously (see Note 7).
(b) Since we also decided that the accelerometer data was to
be saved only at the end of each trial, we needed to store
the data in a buffer during each test.
This part of the code was inserted when we imported the
classes related to the accelerometer channel and acceler-
ometer sensor. We configured the channel to the acceler-
ometer to specify the number of acceleration readings to
be stored in the buffer. Each element in the buffer con-
tained acceleration readings for the x, y, and z axes and
data on when the reading took place. Then, to retrieve the
accelerometer data from the buffer, we needed to query
the buffer for accelerometer data. Accelerometer data was
queried from the buffer for each axis.
(c) Since the timing of the recordings was also essential for the
analysis that would be performed, we also needed to query
the time associated with each accelerometer reading.
This code was also inserted when we imported the classes
related to the accelerometer channel and accelerometer
sensor. The time data was queried as part of the acceler-
ometer data from the buffer. Finally, all of the data recov-
ered from the buffer was written to a .csv file on the SD
card by importing the appropriate classes and providing
the required path.
(d) We will not get into the details on how to access the screen
in order to create the user interface. The sheer number of
362 Benoit Carignan et al.

possibilities, depending on how the interface is designed,

would make this impossible to fit within this book chapter.
However, note that the orientation of the screen can be
managed, drawings or pictures can be inserted and made
into interactive buttons, aspect ratios can be modified, etc.
The procedure for each of these aspects of user interface
design are described in the Blackberry® Development Guide
or the developer’s package associated with other platforms.
(e) We opted to use the vibration alarm of the Blackberry® so as
to alert patients that a given trial was finished. This also
allowed us to synchronize our laboratory accelerometer time
series with the smartphone time series for analyses (see below).
For this we needed to import the classes related to the
alert and vibration. We then specified the time when the
vibration was to start and when the vibration was to stop
within each trial.

3.2 Selecting Tasks 1. Tremor can be observed during rest, holding a limb against
gravity (postural), during movement (kinetic or action), and
during precision tasks (intention). If an application is aimed at
patients with Parkinson’s disease, the evaluation of tremor may
be directed towards a resting position as it is the predominant
type of tremor observed in this disease. On the other hand,
patients with Essential tremor mainly exhibit postural and
kinetic tremor. As such, the tasks selected must have clinical
significance and the application must be validated for each task
for which it may be used to assess tremor (see Note 8). Since
our application was not targeted at a specific patient popula-
tion but rather at any pathological tremor, we opted to have
patients perform rest, postural, kinetic, and intention tasks. Of
course, this meant that we had to validate our application for
all the proposed tasks (see below).

3.3 Creating 1. During the first steps of validation of the smartphone, the user
the Interface interface should be as simple as possible to facilitate and expe-
dite testing. Then, depending on the results, testing procedures
may be changed to add complexity and bring the application
closer to a finished product. Here is a flowchart (Fig. 1) of a
simple interface that could be used for validation.
(a) Participant identification was entered in the smartphone
application using the tactile keyboard on the screen. We
opted for a numerical identification system to preserve par-
ticipant anonymity and coded for a three-digit number as
we knew we were testing less than 1,000 participants. We
also entered the trial number next to the patient identifica-
tion number. We coded for a 2 digit number as we knew
each participant would perform less than 100 trials. As
such, each trial was identified by the participant identifica-
 Measuring Tremor with a Smartphone 363

Fig. 1 Flowchart of a simple smartphone interface to record tremor

tion number and trial number (i.e. 019-04). It is impera-

tive that the numbers entered on the smartphone and on
the experiment records match in order to perform the
proper analyses and correctly interpret the results. A good
way to avoid mistakes would be to display the patient iden-
tification number and trial number on the smartphone’s
screen for the entire experiment (see Note 9).
(b) The length of recording for every trial should be in line
with the objectives of the study. For instance, we opted to
record trials of 10 s; which minimized fatigue due to the
large number of trials each patient had to perform but was
long enough to provide important information about the
characteristics of the tremor. However, the first 2 s of each
trial were removed during the analysis in order to remove
movement artifacts that could be present when partici-
pants moved into position. Longer trials may be performed
depending on the objective of the study and it would also
be possible to change the length of the recording between
trials with the proper code. The end of each recording was
signaled by the smartphone’s vibration alert to both the
patient and the experimenter.
(c) Data analysis will be described in detail below. Briefly,
results of data analyses were saved on the smartphone’s SD
card after each trial, with the participant’s identification
and trial numbers. Acceleration time series and related
power spectrums of the signals were saved for future
analyses and validation purposes (see Note 10).
364 Benoit Carignan et al.

(d) For a few points that need to be examined when finalizing

the application; some of which were incorporated in our
application while others are possible future developments,
see Notes 11–14.

3.4 Tremor 1. Next are some specific points that we want to emphasize when
Assessment using laboratory equipment to validate a smartphone applica-
with Laboratory tion since an extensive body of literature already exists on how
Equipment to use this type of equipment for tremor assessment.
(a) The laboratory accelerometer needs to be fixed to the smart-
phone, preferably to the back where it will not interfere with
the interface. Attention must be paid to align the axes of both
accelerometers in order to facilitate the interpretation of the
results. Since we used a uniaxial accelerometer, we aligned the
axis of the laboratory accelerometer with the axis of the
smartphone’s accelerometer that we thought would experi-
ence the tremor with the highest amplitude.
(b) The gain of the laboratory accelerometer should be set in
accordance with the type of tremor being recorded. Since we
recorded tremor ranging from less than 1 mm to more than
several centimeters, we adjusted the gain for each participant
and recorded it to take it into consideration during analysis.
(c) The length of the recording should be longer than the
smartphone recording time. Indeed, the laboratory record-
ing should be started before the smartphone recording
and should end after it in order to capture the entire smart-
phone trial. We started the recording of the laboratory
accelerometer approximately 2 s before pressing on the
start button of the smartphone application and the record-
ing ended approximately 2–3 s after the smartphone appli-
cation had stopped recording. As such, the length of the
recording of the smartphone application was 10 s while it
was 15 s for the laboratory accelerometer.
(d) Files containing data of each trial were identified in accor-
dance with files recorded on the smartphone SD card; using
the participant identification and trial numbers. As such,
we could easily match the trials that were recorded
simultaneously.

3.5 Time Series 1. Since both systems cannot be started exactly at the same time,
Analysis ( see Note 15) the time series needed to be synchronized. We opted to use the
vibration produced by the smartphone at the end of each trial
to synchronize the recording systems. The vibration frequency
produced by the Blackberry® Storm™ 9530 was about 140 Hz.
Then, after removing the signal below 100 Hz (which largely
removed human movement), it was easy to locate the begin-
ning of the vibration on the laboratory accelerometer. Once
 Measuring Tremor with a Smartphone 365

the end of the trial was located, the beginning was found using
the length of the recording. Note that the sampling rate of the
laboratory accelerometer needs to be set adequately in order to
record the vibrations produced by the smartphone.
2. Spatial analysis.
(a) The following analyses were performed on the time series
recorded by the accelerometer of both recording systems.
The mean of the signal was subtracted from each time series
to remove the earth’s acceleration component (see Note 16).
The triaxial accelerometer of the smartphone provided
three different time series (x, y, and z). These time series
were analyzed separately. They could have also been ana-
lyzed together by calculating the resultant of the three axes.
(b) Tremor amplitude: A root mean square was performed on
the time series. First, each data point of the time series was
squared. Then, the mean of all squared values was calcu-
lated. Finally, the square root of this mean was calculated.
(c) Tremor regularity: First, the signal needed to be normalized,
i.e., each value of the time series was divided by the standard
deviation of the entire time series (note that the signal aver-
age was previously set to zero). Second, the time series was
separated into 1 s epoch. The amplitude of each epoch was
calculated using a root mean square (see above). Finally, the
standard deviation of the amplitude of each epoch was con-
sidered as the measure of regularity of the signal.
3. Spectral analysis.
(a) The following analyses were performed on the power spec-
trum of the signal recorded by the accelerometer of both
systems. The power spectrum was obtained using a fast-
Fourier transform method as it is the adequate method for
this type of signal [21]. Note that values under 1 Hz and
over 20 Hz were set to zero in order to remove non-
tremor related movement such as drift [11].
(b) Tremor power distribution: This represents the percentage
of power contained within a selected frequency band. We
chose the 3–7 Hz frequency band due to its clinical rele-
vance (i.e., it is where most of the power lies when examin-
ing pathological tremors such as in Parkinson’s disease and
Essential tremor). Power distribution was calculated by
dividing the area under the curve within the selected band
by the area under the entire curve.
(c) Tremor median power frequency (MPF): This corresponds
to the frequency where half of total power (area under the
curve) was located on each side. It was calculated by initi-
ating an incremental loop where the area under the curve
before a given index was compared to the area under the
curve above.
366 Benoit Carignan et al.

(d) Tremor harmonic index: This indicates the presence of a

sharp peak in the power spectrum. First, the rectangle con-
taining the power spectrum was calculated by multiplying
the maximum value of the power spectrum (peak) by the
width of the power spectrum window. Second, the total
area under the curve of the power spectrum was calcu-
lated. Then, the harmonic index was calculated by dividing
the area of the rectangle and the area under the curve.
(e) Tremor power peak: This represents the frequency where
the power is the highest.
(f ) Tremor power dispersion from MPF: This corresponds to
the width of the band that contains 68 % of total power.
This band was centered at the MPF. As for the MPF, the
power dispersion was calculated using an incremental loop.
(g) Tremor power dispersion at the peak frequency: This is
similar to the power dispersion from the MPF, except that
the band was centered at the tremor power peak.

3.6 Validation 1. The first step of the validation was to assess the adequacy of the
code for the online analyses specifically written for the smart-
phone. To do this, tremor characteristics calculated online with
the smartphone were compared with results obtained using
traditional analyses performed offline on the signal recorded by
the smartphone. Hundreds of trials of simulated tremor were
performed with a lot of variability between them (amplitude
and frequency) and in all tasks that were to be validated. One
of the experimenter produced the simulated tremor as we felt
that this would make the validation more ecological than hav-
ing a device produce different types of oscillations. The statisti-
cal approach taken to compare the results was to perform a
Pearson correlation, Bland–Altman plot and concordance cor-
relation coefficient. After this, characteristics reaching the
threshold of acceptance (set a priori) were used in the next step
of validation. Correlation between results were expected to be
very high (r > 0.95, [20]) since any error would come only
from mathematical or programming issues and not from reso-
lution or protocol issues. Characteristics that did not reach the
threshold of acceptance were ignored in the following steps.
Of course, it may have also been possible to improve their
respective programming. Nonetheless, the results obtained
during this step made us put aside tremor power peak as it did
not meet our threshold [20].
2. The second step of the validation process was to assess the capac-
ity of the smartphone to reproduce results obtained with the
already validated laboratory devices. Therefore, results com-
puted online with the smartphone were compared with those
arising from the laboratory equipment. Figure 2 shows examples
of tremor of different amplitude recorded with the smartphone
Fig. 2 Example of tremor traces recorded with the smartphone and the accelerometer with their corresponding
power spectrum. Top pane: example of a moderate amplitude tremor. Middle pane: example of a high ampli-
tude tremor. Bottom pane: example of a low amplitude tremor (physiological tremor). (a) Example of a tremor
trace recorded with the smartphone, (b) example of the tremor trace from the same trial as in (a) but recorded
with the accelerometer, (c) power spectrum of the tremor trace recorded with the smartphone which was
calculated with the algorithms implemented within the smartphone, (d) power spectrum of the tremor trace
recorded with the smartphone which was calculated offline using our laboratory software, (e) power spectrum
of the tremor trace recorded with the accelerometer which was calculated offline using our laboratory soft-
ware. Figure as originally published in Daneault, J. F., B. Carignan, C.É. Codère, A.F. Sadikot, and C. Duval
(2012). “Using a smart phone as a standalone platform for detection and monitoring of pathological tremors.”
Front Hum Neurosci 6: 357. doi: 10.3389/fnhum.2012.00357
368 Benoit Carignan et al.

and laboratory accelerometer with their respective power spec-

trum. Concurrent validity was used since both recording
occurred at the same time. As in the first step of validation, many
trials of simulated tremor were performed with a lot of variabil-
ity. Then, the same statistical analyses were performed to select
characteristics that reached the set threshold of acceptance. To
see the table with complete results of the validation see Daneault
et al. [20]. Characteristics that did not reach the threshold of
acceptance were considered as not validated and were put aside
in any future experimentation. This step of validation also
allowed us to assess the limits of the smartphone’s hardware.
These limits were then taken into account in future interpreta-
tions of the results. For instance, we noticed that the resolution
of the smartphone’s accelerometer was insufficient to character-
ize tremor below 1 mm of amplitude [20].
3. The third step of the validation process was to assess the clini-
cal relevance of the results obtained with the smartphone.
This step is optional, but should be done if the final purpose
of the smartphone application is to assess the clinical status of
patients. For this step, patients with pathological tremor
should be tested instead of simulated tremor. Figure 3 shows

Fig. 3 Example of pathological tremor traces recorded with the smartphone

application
 Measuring Tremor with a Smartphone 369

Fig. 4 Comparison of tremor amplitude recorded with the smartphone according

to the clinical tremor score each trial was given. Asterisk (*) indicates a signifi-
cant difference from the previous group. P values were 0.004, 0.002, 0.001,
0.001, and 0.017 for the paired comparisons 0–1, 1–2, 2–3, 3–4, and 4–5,
respectively. The associated power for those tests was 0.788, 0.869, 0.973,
1.000, and 0.638, respectively. Figure as originally published in Daneault, J. F.,
B. Carignan, C.É. Codère, A.F. Sadikot, and C. Duval (2012). “Using a smart phone
as a standalone platform for detection and monitoring of pathological tremors.”
Front Hum Neurosci 6: 357. doi: 10.3389/fnhum.2012.00357

examples of different types of tremor (low, medium, and high

amplitude) recorded with the smartphone. In this step, only
tremor amplitude was validated, since it is the most commonly
used tremor characteristic in clinical scales (see Note 17).
Concurrent validation was performed against a clinical scale.
We opted to develop and validate our own clinical scale having
high measurement precision and linear increments [20], but
any clinical scale could be used. To properly validate the
smartphone results, there must be sufficient variability between
patient tremor amplitude to cover the entire clinical spectrum.
A correlation between clinical score and tremor amplitude
obtained from the smartphone was performed. Then, results
obtained with the smartphone were pooled in regard to their
associated clinical score. Next, t-tests (with Bonferroni–Holm
adjustments) between each adjacent group were performed
(see Fig. 4). This indicates that the smartphone can provide a
reliable measure of tremor of different amplitudes that can be
of clinical relevance to physicians.
370 Benoit Carignan et al.

4 Notes

1. Choosing the appropriate smartphone can be very intricate.

We chose to use the Blackberry® Storm™ 9530 because we had
one available for testing in the laboratory and we had someone
able to program in Java. Here are several issues that may influ-
ence the choice of platform. First, platforms use different pro-
gramming languages and someone in the laboratory needs to
be able to program using the chosen language in order to
develop the application. Second, a phone using the chosen
platform needs to be available in the laboratory in order to test
the application; simulators can only do so much. Finally, the
market share of each platform should also be taken into con-
sideration. For example, to reach the maximum number of
users, you may want to choose Android™, since 80 % of all
smartphones currently sold in the world run under this plat-
form. If specific hardware characteristics (i.e., higher resolu-
tion) are required, then specific phones may be targeted instead
of platforms. Another point to consider is how the software
manages data recording. For instance, we observed that when
we recorded tremor using an Android™ smartphone, the
recording was not continuous (i.e., no data was being recorded
when no movement was detected by the accelerometer).
Because we did not have the time to find solutions to circum-
vent this issue, this led us to abandon the use of this platform
for tremor recording. This should also be taken into account in
the choice of platform and analysis of the data.
2. Eclipse is a multi-language workspace that comprises a base
workspace and an extensible plug-in system for customizing
the environment. It can be used to develop applications in Java
but also in other programming languages such as C, C++,
JavaScript, etc. The Eclipse SDK includes the Eclipse Java
development tools, offering an integrated development envi-
ronment with a Java compiler and a full model of the Java
source files. It is also possible to use Microsoft Visual Studio as
a workspace instead of Eclipse.
3. NetBeans IDE is an open-source integrated development envi-
ronment for developing primarily with Java, but also with
other languages such as PHP, C, C++, and HTML5. Other
integrated development environment could be used such as
Code Blocks, IntelliJ IDEA, JBuilder, JDeveloper, etc.
Deciding on which software to use is entirely up to the devel-
opment team as most software share the same functionalities.
4. Our system included a computer, a data acquisition card, cus-
tom recording platform using Data Acquisition System
Laboratory (DASYLab, National Instruments Ireland Resources
Limited; measX GmbH & Co. KG, Germany) and custom
 Measuring Tremor with a Smartphone 371

analysis algorithms using S-Plus software (Mathsoft, Seattle,

WA, USA). Any other system could have been used, as long as
the computer had enough memory and enough processing
speed, the acquisition card had adequate resolution, the accel-
erometer was sensitive enough for tremor detection and the
algorithms for tremor analysis were proven. Furthermore, other
laboratory devices could have been used for tremor detection,
such as triaxial accelerometers, laser displacement systems,
gyroscopes, etc.
5. The detailed procedures (code) can be found in the develop-
er’s guide associated with the given platform.
6. The Java programming language is used on many platforms.
However, other languages can also be used such as C, C++,
Objective-C, Object-Pascal, etc. Understanding the require-
ments of the applications under development is essential in
order to identify which programming language may be required.
7. It is important to note that while the data was saved continu-
ously, we also set the sampling frequency to 60 Hz; the maxi-
mum allowable sampling frequency of our Blackberry® device.
The sampling frequency is important when considering the
Nyquist sampling theorem where a function with a limited fre-
quency band can be perfectly reconstructed from a sequence of
samples only if the frequency limit is not greater than half the
sampling rate. In other words, since our power spectrum win-
dow examines frequencies between 1 and 20 Hz, our sampling
frequency needed to be at least 40 Hz.
8. In a recent study [20], we validated the smartphone to assess
tremor amplitude during four different tasks: rest, postural,
kinetic and intension. Amplitudes measured with the smart-
phone were well correlated with clinical measurements for the
rest, postural and intension conditions, but not for the kinetic
condition. This was due to the fact that the high amplitude of
the voluntary movement interfered with tremor detection.
As such, we used another tremor characteristic (tremor power
distribution, see analysis section for details) as an indicator of
tremor amplitude. This highlights the fact that the application
must be validated for all tasks that may be required of patients.
For instance, if the application is targeted for patients with
Parkinson’s disease, then it could be validated only during a
resting task. However, if at some point patients with Essential
tremor will be tested or postural tremor of patients with
Parkinson’s disease will also be tested, then the application
must also be validated in these conditions.
9. The nature of protocols using smartphones usually involve test-
ing of many patients in a short period of time. Properly identify-
ing trials and introducing failsafe procedures that prevent the
overwriting of saved data can prevent massive headaches.
372 Benoit Carignan et al.

For instance, since we had not coded failsafe procedures, we

input the same patient identification number twice for two dif-
ferent patients; this led to the loss of important data associated
with the first patient.
10. Another issue that needed to be taken into account was how
the platform managed saved data files. For instance, we
observed that on the Blackberry® Storm™ 9530, the home
directory of the SD card is limited to 161 files. So after a few
participants performed many trials, the data and results were
no longer being saved on the SD card. So, once again, impor-
tant data was lost. This could have been prevented by changing
the path where data were being saved or by testing beforehand
the capacity of the home directory.
11. The ergonomic aspect of the application should take into
account the targeted population. For example, when the target
population is patient with tremor, one might consider that
they are usually elderly and they lack fine manual dexterity. As
such, large buttons would be preferable for the use of the
application and large fonts would also improve the usability of
the application. For instance, since patients with severe patho-
logical tremor used our application, the start button was
brightly colored circle as large as the screen with the word
“start” written within it in large font contrasting with the color
of the button. As a result, our participants never had any issue
starting the test.
12. The application could be designed as basic start/repeat/stop
buttons with instructions being provided on the side or it
could include pictograms or videos of the desired conditions
participants are expected to perform. Since our application was
being validated in the presence of experimenters that could
provide instructions to the participants, we opted for the basic
design. However, if participants were to be tested at home or
without the presence of an experimenter, it would be beneficial
to provide as much information as possible to the participant
in order for them to perform the test correctly.
13. A reminder/alarm could also be imbedded within the applica-
tion for long-term tremor assessment. Since we tested partici-
pants at one point in time, a reminder or alarm was not
necessary. However, if the objective was to perform long-term
monitoring of tremor, incorporating alarms to remind the par-
ticipants to perform the tests at specific intervals could be
coded within the application.
14. Instead of being solely saved on the SD card, the results could also
be automatically transmitted to a third party. While we considered
incorporating this feature within our application, it raised some
ethical issues. Specifically, this would mean sending medical infor-
mation over an unsecured network by email or text message.
 Measuring Tremor with a Smartphone 373

Though this feature would be greatly beneficial for patient

follow-up, ways to secure the network, the patient informa-
tion, and server or device on which the data will be received
needs to be implemented. This is beyond the scope of the cur-
rent chapter but should be taken into consideration when con-
ceptualizing the application.
15. Note that tremor characteristics presented in the current work
are those used in a previous paper [20]. Any tremor analyses
could be performed with proper programming depending on
the objective of the application and study.
16. Unwanted component of a signal (such as earth’s acceleration)
can also be removed using different filtering methods, such as
a high-pass filter at 1 Hz.
17. While the goal of this step was to assess pathological tremor
amplitude, we found that in one condition (kinetic task) tremor
amplitude measured with the smartphone was not an adequate
characteristic to predict clinical tremor amplitude. In fact, for
this particular condition, the high amplitude of the voluntary
movement masked the relatively lower tremor amplitude sig-
nal. We circumvented this issue by using the power distribu-
tion as a measure of tremor amplitude. Simply put, we focused
on oscillations located within the 3–7 Hz frequency band
where the majority of pathological tremor oscillations lie,
which minimized the importance of the voluntary movement
as those oscillations are usually below 3 Hz.

References
1. Bain P, Brin M, Deuschl G, Elble R, Jankovic J, 7. Iwadate Y, Saeki N, Namba H, Odaki M, Oka
Findley L, Koller WC, Pahwa R (2000) Criteria N, Yamaura A (1989) Post-traumatic intention
for the diagnosis of essential tremor. Neurology tremor–clinical features and CT findings.
54(11 Suppl 4):S7 Neurosurg Rev 12(Suppl 1):500–507
2. Louis ED (2000) Essential tremor. Arch 8. Morrison S, Mills P, Barrett R (2006)
Neurol 57(10):1522–1524 Differences in multiple segment tremor dynam-
3. Jankovic J (2008) Parkinson’s disease: clinical ics between young and elderly persons. J
features and diagnosis. J Neurol Neurosurg Gerontol A Biol Sci Med Sci 61(9):982–990
Psychiatry 79(4):368–376 9. Morrison S, Newell KM (2000) Postural and
4. Alusi SH, Worthington J, Glickman S, Bain PG resting tremor in the upper limb. Clin
(2001) A study of tremor in multiple sclerosis. Neurophysiol 111(4):651–663
Brain 124(Pt 4):720–730 10. Carignan B, Daneault JF, Duval C (2009)
5. Pittock SJ, McClelland RL, Mayr WT, The amplitude of physiological tremor can
Rodriguez M, Matsumoto JY (2004) be voluntarily modulated. Exp Brain Res
Prevalence of tremor in multiple sclerosis 194(2):309–316
and associated disability in the Olmsted 11. Carignan B, Daneault JF, Duval C (2010)
County population. Mov Disord 19(12): Quantifying the importance of high frequency
1482–1485 components on the amplitude of physiological
6. Andrew J, Fowler CJ, Harrison MJ (1982) tremor. Exp Brain Res 202(2):299–306
Tremor after head injury and its treatment by 12. Carignan B, Daneault JF, Duval C (2011) The
stereotaxic surgery. J Neurol Neurosurg organization of upper limb physiological
Psychiatry 45(9):815–819 tremor. Eur J Appl Physiol 112:1269
374 Benoit Carignan et al.

13. Daneault JF, Carignan B, Duval C (2010) New actigraph for long-term tremor record-
Bilateral effect of a unilateral voluntary modula- ing. Mov Disord 21(8):1136–1143
tion of physiological tremor. Clin Neurophysiol 18. (2010) Mobile computing platforms such as
121(5):734–743 the iPhone are beginning to make inroads into
14. Daneault JF, Carignan B, Duval C (2011) the laboratory-serious prospect or fairy tale?
Finger tremor can be voluntarily reduced Nat Methods 7(2): 87
during a tracking task. Brain Res 1370: 19. Wolf JA, Moreau JF, Akilov O, Patton T, English
164–174 JC, Ho J, Ferris LK (2013) Diagnostic inaccu-
15. Daneault JF, Carignan B, Rahimi F, Sadikot racy of smartphone applications for melanoma
AF, Duval C (2013) Postural tremors. In: detection. JAMA Dermatol 149(4):422–426
Grimaldi G, Manto M (eds) Mechanisms and 20. Daneault JF, Carignan B, Codere CE, Sadikot
emerging therapies in tremor disorders. AF, Duval C (2012) Using a smart phone as a
Springer, New York, NY, pp 133–150 standalone platform for detection and moni-
16. Yang MH, Sheu YH, Shih YH, Young MS toring of pathological tremors. Front Hum
(2003) Portable tremor monitor system for Neurosci 6:357
real-time full-wave monitoring and analysis. 21. McNames J (2013) Signal processing. In:
Rev Sci Instrum 74(3):1303–1309 Grimaldi G, Manto M (eds) Mechanisms and
17. Van Someren EJ, Pticek MD, Speelman JD, emerging therapies in tremor disorders.
Schuurman PR, Esselink R, Swaab DF (2006) Springer, New York, NY, pp 371–389
 Chapter 25

The Use of Single-Electrode Wireless EEG

in Biobehavioral Investigations
Dmitri V. Poltavski

Abstract
The purpose of this chapter is to introduce novice and intermediate EEG researchers to a convenient and
user-friendly EEG system from NeuroSky, Inc. In our recent study we were interested in changes in the
frontal cortical EEG activity of healthy adults as a function of accommodative stress during performance
of a sustained attention task. We used a commercially available low-cost wireless EEG device from NeuroSky
(MindSet), which has a single active Fp1 dry electrode capable of recording research-grade EEG coupled
with powerful noise-filtering and data software support. The convenience and ease-of-use of MindSet is
further enhanced with validated eSense meters of Attention and Meditation. In this chapter we also pro-
vide additional data analytic support for EEG power spectrum using SPSS syntax commonly used in many
biobehavioral sciences.

Key words EEG, NeuroSky, MindSet, NeuroView, SPSS, Attention, Meditation

1 Introduction

The measurement of brain electrical activity using the electroen-

cephalogram (EEG) provides a noninvasive method to directly
measure brain function and make inferences about regional corti-
cal activity. In the past 50 years it has been widely used as a diag-
nostic tool (e.g., epilepsy, schizophrenia, autism) and a powerful
research technique with a broad spectrum of applications in the
biobehavioral sciences ranging from studies of basic cognitive pro-
cesses to emotional function, dysfunction, and development. Most
researchers now agree that the ongoing EEG is derived from sum-
mated postsynaptic potentials of cortical neurons [1].
There are five types of waves in an EEG defined by their fre-
quency ranges: the frequency of the α-waves fall between 8 and
12 Hz with amplitudes between 30 and 50 μV and is typically
recorded from a conscious person at rest in a quiet environment
with the best signal from the parietal and the occipital regions of
the brain. The β-waves have frequencies between 12 and 30 Hz,

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_25, © Springer Science+Business Media New York 2015

375
376 Dmitri V. Poltavski

and amplitudes between 5 and 20 μV; they usually appear in the

frontal region of the brain while a person is conscious and alert,
and are especially prominent when the person is thinking or under-
going sensory stimulation. The θ-waves have frequencies between
4 and 7 Hz and normally have amplitudes less than 30 μV. They
are mainly released from the parietal and the temporal regions of
the brain when a person is emotionally stressed, unconscious, or
when the person’s body is in a state of deep relaxation. The δ-waves
have frequencies between 0.5 and 3 Hz, and amplitudes between
100 and 200 μV. A person will normally release δ-waves during
deep sleep, when unconscious, deeply anesthetized, or if experi-
encing hypoxia. The δ-waves are almost never recorded when a
person is conscious. The γ-waves have frequencies between 30 and
60 Hz with amplitudes ranging from 5 to 10 μV. Recent studies
have found links between the γ-waves and selective attention,
human cognition, and perceptive activities [2]. Figure 1 summarizes
various EEG waveform components.

Aplha waves

0.0 0.2 0.4 0.6 0.8 1.0

Beta waves

0.0 0.2 0.4 0.6 0.8 1.0

Gama waves

0.0 0.2 0.4 0.6 0.8 1.0

Delta waves

0.0 0.2 0.4 0.6 0.8 1.0

Theta waves

0.0 0.2 0.4 0.6 0.8 1.0

Fig. 1 EEG waveform components

The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 377

Dorsolateral Prefrontal
Cortex

Orbitofrontal

Fig. 2 The prefrontal cortex

Historically, EEG required complex, immovable high cost

equipment, vast expertise of a researcher or a medical professional
and presented a challenging problem of distinguishing brain signal
from the noise that comes from ambient electricity, muscle move-
ment, etc. In addition, traditional EEG measurements have been
recorded with Ag/AgCl electrodes, which required the use of a
conductive gel, thus necessitating long preparation time and caus-
ing discomfort to the user. For better performance, multichannel
EEG are often used (e.g., up to 256 channels), which takes longer
to prepare and adds more discomfort.
Since the frontal lobe (particularly the dorsolateral prefrontal
cortex (DLPFC), see Fig. 2) is involved in the control of human
emotions [3], behavioral inhibition [4], state of mind [5], and
attentional focus [6], the activity in this region of the brain may be
studied with EEG to answer a broad range of research questions.
With fast-paced technological advances it is now possible to record
brain wave signals conveniently from the location of interest using
state-of-the-art wireless technology, noise-filtering chips, and dry
electrode arrangement. The purpose of the present chapter is to
introduce novice and intermediate EEG researchers to a conve-
nient and user-friendly EEG system and associated software from
NeuroSky, Inc.
In our recent study [7] we were interested in changes in
accommodative lag [of the eye lens] and the frontal cortical EEG
activity of healthy adults as a function of accommodative stress
(making it harder for the participants to maintain visual stimulus in
focus while wearing −2.00 D lenses) during performance of a
sustained attention task (Conners’ Continuous Performance task).
378 Dmitri V. Poltavski

The rationale for the study was the finding that many of the adverse
symptoms expressed by individuals with accommodative and ver-
gence problems of the eyes have also been reported for individuals
diagnosed with attention problems [8, 9]. Granet et al. [9] went
further to suggest that medications used to treat ADHD may actu-
ally aggravate certain oculomotor conditions such as convergence
insufficiency without improving symptoms of inattentiveness. The
function of stimulant medication according to the neuropharma-
cological model of ADHD is to decrease the “noise” in the PFC
due to insufficient dopamine (DA) availability by increasing extra-
cellular DA in this region [10], which in behavioral studies is asso-
ciated with improved attentiveness [11] and in EEG studies with
frontal attenuation of power in EEG bands linked to inattention
such as theta and alpha [12]. In our study we hypothesized that
the disruption of visual processing alone may be sufficient to
account for symptoms of inattentiveness in the absence of changes
in the frontal cortical activity. For our study we used a commer-
cially available low-cost wireless EEG device from NeuroSky, which
is a single, dry-electrode EEG system (MindSet) with built-in noise
filtering and data software support.

2 Materials

2.1 Mindset/ The NeuroSky’s Mindset headset (with a built-in ThinkGear

ThinkGear module) incorporates a dry electrode system with a single active
pea-sized electrode (10 mm diameter) that is placed in the left
forehead area approximately 2 cm above the left eyebrow (see
Fig. 3). This roughly corresponds to area Fp1 (DLPFC) using the
International 10–20 System of electrode placement (see Fig. 4).
The reference electrode is integrated into the earpiece of the head-
set and measures electrical potential from two points on the left
earlobe. The reference is used to subtract the common ambient
noise through a process known as common mode rejection. The
earlobe is a location that experiences the same ambient noise as the
NeuroSky forehead sensor but with minimal neural activity.
The electrical potential is supplied directly to the built-in chip-
set for analog filtering with band pass at 512 Hz AD conversion and
sampling rate. Notch filters eliminate electrical noise from the grid,
which varies from 50 to 60 Hz, depending on worldwide geogra-
phy. Embedded filtering protocols eliminate known noise frequen-
cies such as muscle, pulse, and electrical devices, allowing recording
of signals with 96 % accuracy of research grade EEG [13].
Analog data is automatically converted into digital format and
analyzed by Fast Fourier Transform (FFT) in the headset circuit
board. FFT produces power values for each 1-s epoch and each
frequency bin that are transmitted via Bluetooth to the Mindset
Neuroview software that comes with the Mindset and is compati-
ble with both Windows and Mac OS operating systems.
Fig. 3 NeuroSky Mindset

Fig. 4 Twenty-one electrode locations of the International 10–20 system for EEG
(electroencephalography) recording
380 Dmitri V. Poltavski

The ThinkGear components deliver their digital data via

Bluetooth as an asynchronous serial stream of bytes in a packet
format consisting of a packet header, a packet payload and a pay-
load checksum. Such packets can be delivered to any device that
can receive a serial stream of bytes such as a PC or another micro-
processor. The Header of a Packet consists of 3 bytes: two synchro-
nization [SYNC] bytes followed by a [PLENGTH] (Payload
length) byte that indicates the length of the data payload that fol-
lows. The payload checksum byte must be used to first verify the
integrity of the Packet’s Data payload, after which (if the calculated
payload checksum and received [CHKSUM] values match), the
receiver will proceed to parse the Data Payload according to pre-
specified protocols. The Packet format is thought to be robust and
flexible: combined, the Header and Checksum provide data stream
synchronization and data integrity checks, while the format of the
Data Payload ensures that new data fields can be added to (or exist-
ing data fields removed from) the Packet in the future without
breaking any Packet parsers in any existing applications/devices.
In 2009 NeuroSky conducted benchmark tests of their dry
EEG by comparing EEG signals measured by the dry sensor sys-
tem with signals from the Biopac system, a well-known wet elec-
trode EEG system widely used in medical and research applications
[13]. EEG was recorded for various conditions such as with the
subject relaxing and in a meditative state, alert and in an attentive
state, and during eyeblink artifacts. The results of the comparisons
showed that the overall correlation coefficient for the power spec-
trum between 1 and 30 Hz was 0.715 during the resting state and
the eye blink artifacts. During the alert “attention” state the overall
correlation coefficient was even higher reaching 0.94.
The researchers concluded the power density distribution for
the NeuroSky dry electrode system showed the same pattern as the
Biopac wet electrode system except in low frequency bands where
power density of Biopac was higher than that of NeuroSky system.
NeuroSky explains this discrepancy on the basis of increased low
frequency fluctuation noise in Biopac data resulting from the use
of 3-ft-long wires between the pre-amplifier and electrodes.
NeuroSky system, on the other hand, employed 10-in. wires that
cannot move during EEG measurement, which provides more
noise resistance and possibly conferring greater ecological validity
to NeuroSky system, as it can be used across a wider range of
research contexts.

2.2 Data Analytic In our study we used SPSS 19.0 (also known as IBM SPSS since
Software 2009) statistical package that is among the most widely used pro-
grams for statistical analysis in social and behavioral sciences. More
recent SPSS base versions (latest version 22.0) and add-on mod-
ules provide a fairly versatile array of tools for inferential statistical
analyses. In addition to statistical analysis, data management of
 The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 381

high complexity (case selection, data reconfiguration and creation

of derived variables) is also quite accessible even to novice SPSS
users. In our study we used SPSS both for descriptive and inferen-
tial statistics (e.g., paired-sample t-tests and multiple regression) as
well as for extensive data management purposes utilizing its com-
mand syntax. For example, while creation of calculated variables
based on simple rules is possible through the use of pull-down
menus in SPSS, recoding of variables and creation of calculated
variables according to more complex rules and mathematical
manipulations is only possible using SPSS syntax. The syntax files
used in our study for data reconfiguration and variable derivation
are provided in the end of the chapter.

3 Methods

3.1 Settings Support software for the MindSet/ThinkGear comes as part of the
NeuroSky MindSet Research Tools (MRT) in two varieties:
1 Included with the MindSet is the NeuroView software, which
makes it easy to connect, graph, view, and record MindSet data
in real-time. NeuroView is designed to be appropriate for nov-
ice to intermediate EEG researchers wishing to view and record
EEG data in real-time. The recorded data can be easily exported
to other third-party applications (e.g., Excel, SPSS) for down-
stream data analysis and processing as it is saved in the comma-
separated value (csv) file format.
2 MRT also includes the more advanced NeuroSkyLab MATLAB
module, which adds the ability to define custom MATLAB
scripts and functions for customized processing and analysis of
MindSet data. It is targeted at the more advanced EEG
researcher who is familiar with the MATLAB environment.
For those comfortable with MATLAB scripting, NeuroSkyLab
provides much more powerful capabilities than NeuroView in
terms of customization, real-time data viewing, and real-time
analysis.
In our 2012 study [7] we used the MindView software and
performed all subsequent scripting and analysis in SPSS 19.0. This
was a preferred method of data processing as we had multiple physi-
ological and behavioral recording channels. The use of SPSS allowed
us to aggregate all subject data in one file for multivariate analyses.
Once the Neuroview is opened, the Bluetooth connection
with the ThinkGear headset is established and the headset is placed
on the subject, the status bar at the bottom of the Neuroview win-
dow will display a value for the signal quality. A Poor Signal Quality
value of 0 indicates a relatively clean signal, while higher values
indicate progressively poorer signals. A value of 200 has a special
meaning, namely, that the electrodes of the ThinkGear module/
382 Dmitri V. Poltavski

headset are not contacting the skin of the head at all. Poor signal
may be caused by a number of different things. In order of severity,
they are:
(a) Sensor, ground, or reference electrodes not being on a per-
son’s head (i.e., when nobody is wearing the ThinkGear).
(b) Poor contact of the sensor, ground, or reference electrodes to
a person’s skin (i.e., hair in the way, or headset which does not
properly fit a person’s head, or headset not properly placed on
the head).
(c) Excessive motion of the wearer (i.e., moving head or body
excessively, jostling the headset).
(d) Excessive environmental electrostatic noise (some environ-
ments have strong electric signals or static electricity buildup
in the person wearing the sensor).
(e) Excessive non-EEG biometric noise (i.e., EMG, EKG/ECG,
EOG, etc.)
To minimize interference with the signal it is thus recom-
mended to, first of all, ensure a good contact of dry electrodes with
the person’s earlobe and frontal (forehead) area by

(a) Wiping out contact areas of the electrodes and the subject’s
head with an alcohol wipe.
(b) Removing all hair and metal objects (e.g., ear rings) from the
target site.
(c) Making sure that the adjustable head-band is properly sized: snug
over the head without being too loose or too tight, and is posi-
tioned over the central line of the head (C3, Cz, C4; see Fig. 4).
It is also recommended to allow the subject to sit quietly and
relatively motionless before a good EEG signal is obtained. Thirdly,
acquisition of the initial EEG signal of good quality (i.e., value of
zero on the meter) may be improved by turning off all other elec-
tronic equipment not necessary for signal acquisition.
The optimal setting for recording is obtained when the “poor
signal” gauge at the bottom of the screen indicates “0” and the
e-sense meters show activity (when the signal quality is “200,” the
e-sense meters do not move. At the same time the raw EEG graph
may be showing fluctuations due to a mixture of extraneous noise
and EEG activity).

3.2 Recording Once acceptable signal quality is obtained (see Note 1) and EEG
recording is started (by pressing on the green “Start” button),
NeuroView has the capability of displaying ongoing EEG activity
in four different modes (by selecting each mode through the
“View” command): as raw EEG, power spectrum (μV2), and in
terms of eSense™ measures of attention and meditation (see Fig. 5
for an example).
 The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 383

Fig. 5 NeuroView live recording window

For the two different types of eSenses (i.e., Attention,

Meditation), the meter value is reported on a relative scale of 1–100.
On this scale, a value between 40 and 60 at any given moment in
time is considered “neutral,” and is similar in notion to “baselines”
that are established in conventional EEG measurement techniques.
A value from 60 to 80 is considered “slightly elevated,” and may be
interpreted as levels being possibly higher than normal (levels of
Attention or Meditation that may be higher than normal for a given
person). Values from 80 to 100 are considered “elevated,” meaning
they are strongly indicative of heightened levels of that eSense.
On the other hand a value between 20 and 40 indicates
“reduced” levels of the eSense, while a value between 1 and 20
indicates “strongly lowered” levels of the eSense. An eSense meter
value of 0 indicates the ThinkGear is unable to calculate an eSense
level with a reasonable amount of reliability possibly due to exces-
sive noise and consequently poor signal quality.
The ATTENTION eSense meter is meant to indicate the
intensity of a user’s level of mental “focus” or “attention,” such as
that which occurs during intense concentration and directed (but
stable) mental activity. Distractions, wandering thoughts, lack of
focus, or anxiety, according to the NeuroView developers may
lower the Attention meter levels.
384 Dmitri V. Poltavski

The MEDITATION eSense has been designed to indicate the

level of a user’s mental “calmness” or “relaxation” and is related to
reduced activity by the active mental processes in the brain (e.g., as
in closing in closing one’s eyes). Again the developers suggest that
distractions, wandering thoughts, anxiety, agitation, and sensory
stimuli may lower the Meditation meter values.
In our study we did not utilize the eSense meter data as the
algorithms for determining attention and meditation levels are
proprietary and were not revealed to our research group. The
developers, however, note that “some parts of the eSense algo-
rithm are dynamically learning, and at times employ some ‘slow-
adaptive’ algorithms to adjust to natural fluctuations and trends of
each user, accounting for and compensating for the fact that EEG
in the human brain is subject to normal ranges of variance and
fluctuation.” The utilization of this self-adapting methodology is
further cited by the developers as an advantage enabling research-
ers to operate on a wide range of individuals under wide range of
personal and environmental conditions while still preserving accu-
racy and reliability.
There is some empirical support to the above claim. The atten-
tion and meditation metrics have been validated by Yasui [14] who
used the system’s EEG output to discriminate between REM/
non-REM sleep, car driving, using a cell phone while driving a car
as well as students’ engagement in classroom activities and relax-
ation. Redolledo-Mendez et al. [15] tested the attention eSense
metric in a virtual environment of SecondLife using Avatar-driven
post exercise attentional assessment. The researchers reported a
significant 0.39 correlation coefficient between the eSense
Attention meter scores and attentional scores obtained during
post-exercise assessment. Crowley et al. [16] also reported a
78.04 % concordance rate between the eSense Meditation scale
categorization of one’s mental state as “stressed” and the
participant’s self-reported level of stress experienced during com-
pletion of the Tower of Hanoi test.
When EEG recording is finished (by clicking on the “Stop”
button; see Note 2), by default Neuroview suggests saving several
types of data: time-stamped signal quality; time-stamped raw EEG,
time-stamped signal quality and Attention meter data, time-
stamped signal quality and Meditation meter data, a combined
time-stamped signal quality, attention and meditation data, and
time-stamped power spectrum EEG data. For the purposes of our
study we unclicked all file type boxes except for power spectrum,
since discrete power bands were of primary importance for our
research questions.

3.3 Analysis Once each subject’s CSV power spectrum file was imported into
SPSS, mean EEG band data was derived for each subject using the
“Descriptives” command in SPSS applied to the power spectrum
over the entire duration of each session (baseline and stress in our
The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 385

study). Mean values for each EEG power band were then entered
as columns into a combined subjects’ file, where each row repre-
sented one subject. The raw power spectrum file contains fre-
quency bins with increments of 0.25 Hz. Thus before we calculated
mean values for each subject in each power band of interest (alpha,
beta and theta), we combined frequency bins to represent a specific
window of interest using the following SPSS syntax (see Note 3):
*Computing power for frequencies of the THETA wave EEG band.
COMPUTE theta_4_7HZ=(@4 + @4.25 + @4.5 + @4.75 + @5 +
@5.25 + @5.5 + @5.75 + @6 + @6.25 + @6.5 + @6.75 + @7 +
@7.25 + @7.5 + @7.75) / 16.
VARIABLE LABELS theta_4_7HZ 'mean power within the theta
EEG band'.
EXECUTE.
*Computing power for frequencies of the Alpha wave EEG band.
COMPUTE alpha_8_12HZ=(@8 + @8.25 + @8.5 + @8.75 + @9 +
@9.25 + @9.5 + @9.75 + @10 + @10.25 + @10.5 + @10.75 + @11
+ @11.25 + @11.5 + @11.75) / 16.
VARIABLE LABELS alpha_8_12HZ 'mean power within the
alpha EEG band band'.
EXECUTE.
*Computing power for frequencies of the Beta wave EEG band.
COMPUTE beta_12_30HZ=(@12 + @12.25 + @12.5 + @12.75 +
@13 + @13.25 + @13.5 + @13.75 +@14 + @14.25 + @14.5 +
@14.75 + @15 + @15.25 + @15.5 + @15.75 + @16 + @16.25 +
@16.5 + @16.75 + @17 + @17.25 + @17.5 + @17.75 + @18 +
@18.25 + @18.5 + @18.75 + @19 + @19.25 + @19.5 + @19.75 +
@20 + @20.25 + @20.5 + @20.75 +@21 + @21.25 + @21.5 +
@21.75 + @22 + @22.25 + @22.5 + @22.75+@23 + @23.25 +
@23.5 + @23.75 + @24 + @24.25 + @24.5 + @24.75 + @25 +
@25.25 + @25.5 + @25.75 + @26 + @26.25 + @26.5 + @26.75 +
@27 + @27.25 + @27.5 + @27.75 + @28 + @28.25 + @28.5 +
@28.75+ @29 + @29.25 + @29.5 + @29.75) / 72.
VARIABLE LABELS beta_12_30HZ 'mean power within the
beta EEG band'.
EXECUTE.
*Computing amplitude for frequencies of the Gamma wave EEG
band.
COMPUTE gamma_30_60HZ=(@30+ @31 + @32 + @33 + @34
+ @35 + @36 + @37 + @38 + @39 + @40 + @41 + @42 + @43 + @44
+ @45 + @46 + @47 + @48 + @49 + @50 + @51 @52 + @53 + @54
+ @55 + @56 + @57 + @58 + @59+ @60 + @30.25+ @31.25 +
@32.25 + @33.25 + @34.25 + @35.25 + @36.25 + @37.25 [email protected]
+ @39.25 + @40.25 + @41.25 + @42.25 + @43.25 + @44.25 +
@45.25 + @46.25 + @47.25 + @48.25 + @49.25 + @50.25 +
386 Dmitri V. Poltavski

@51.25 + @52.25 + @53.25 + @54.25 + @55.25 + @56.25 +

@57.25 + @58.25 + @59.25+ @60.25 + @30.5+ @31.5 + @32.5 +
@33.5 + @34.5 + @35.5 + @36.5 + @37.5 [email protected] + @39.5 + @40.5
+ @41.5 + @42.5 + @43.5 + @44.5 + @45.5 + @46.5 + @47.5 +
@48.5 + @49.5 + @50.5 + @51.5 + @52.5 + @53.5 + @54.5 +
@55.5 + @56.5 + @57.5 + @58.5 + @59.5+ @60.5 + @30.75+
@31.75 + @32.75 + @33.75 + @34.75 + @35.75 + @36.75 +
@37.75 [email protected] + @39.75 + @40.75 + @41.75 + @42.75 + @43.75
+ @44.75 + @45.75 + @46.75 + @47.75 + @48.75 + @49.75 +
@50.75 + @51.75 + @52.75 + @53.7 + @54.75 + @55.75 + @56.75
+ @57.75 + @58.75 + @59.75+ @60.75) / 124.
VARIABLE LABELS gamma_30_60HZ 'amplitude of the gamma
EEG band width'.
EXECUTE.
The fact that raw data contains power values within 0.25 Hz
windows is actually an advantage rather than a drawback. It pro-
vides flexibility in determining power windows of interest. Different
research questions may focus on specific narrow frequency ranges
previously identified as important for the particular research area.
For example, beta wave can be further subdivided into Low Beta
(12–15 Hz), Midrange Beta (16–20 Hz), and High Beta (21–
30 Hz). Gola et al. [17] recently showed that increased midrange
beta power (17–19 Hz) around the occipital region during a sim-
ple visual attention task differentiated trials ending with correct
response from those without a response.
Because distributions of power tend to be skewed, most inves-
tigators transform the data in order to normalize its distribution.
The most common transform used is the natural log transforma-
tion [18]. In our study the power distribution (assessed using mea-
sures of skewness, kurtosis and graphing) appeared skewed which
warranted an ln transformation. We used the following SPSS syn-
tax to transform our power data.
COMPUTE LN_theta_test=LN(theta_test).
VARIABLE LABELS LN_theta_test 'LN transformed power of
the theta wave (4-7HZ) in the stress condition'.
EXECUTE.
COMPUTE LN_alpha_test=LN(alpha_test).
VARIABLE LABELS LN_alpha_test 'LN transformed power of
the alpha (8-12HZ) wave in the stress condition'.
EXECUTE.
COMPUTE LN_beta_test=LN(beta_test).
VARIABLE LABELS LN_beta_test 'LN transformed power of the
beta wave (12-30Hz) in the stress condition'.
EXECUTE.
COMPUTE LN_gamma_test=LN(gamma_test).
VARIABLE LABELS LN_gamma_test 'LN transformed power of
the gamma wave (30-60Hz) in the stress condition'.
EXECUTE.
 The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 387

LN power theta (4−7HZ)

5.00 LN power of alpha (8−
12HZ)
LN power of beta wave
(12−20HZ)
Ln Theta (M = 2.77; 95%CI: 2.05 - 3.50)
4.00

Ln alpha (M = 2.21; 95%CI: 1.48 - 2.93)

3.00
Ln power mV2

2.00

1.00

Ln Beta (M = 1.53; 95%CI: 0.70 - 2.36)

0.00

−1.00

−5.74 −5.23 −4.60 −4.60 −4.41 −4.18 −3.55 −3.37 −3.28 −2.38
accommodative lag (D) in the stress condition

Fig. 6 Relationship between frontal EEG bands and accommodative lag during completion of Conners CPT
in the stress condition

3.4 Results Consistent with our research hypothesis, the results of our study
showed that performance on a computerized test of sustained
attention (Conner’s CPT) can be compromised by adding a −2.00
D accommodative stimulus to the normal −2.50 D accommodative-
vergence stimulus demand. This increased −4.50D accommodative
demand resulted in a significantly larger accommodative lag and
significantly poorer performance on the CPT (slower reaction
time, greater standard error of hit reaction time, greater response
variability, poorer stimulus detectability, greater number of perse-
verations, and higher overall probability of clinical classification) in
the absence of any appreciable change in frontal lobe electrophysi-
ological activity. There was similarly no significant correlation
between the power of individual EEG bands and accommodative
lag. Figure 6 shows a fairly flat pattern of fluctuations of the three
bands across the range of accommodative lag responses. What was
particularly interesting was the finding that accommodative lag
alone could account for a significant proportion of cases (about
40 %) with a higher probability of clinical classification on the
Conner’s CPT (higher likelihood of ADHD symptoms) even after
controlling for individual band EEG activity. The results have
direct implications for management of drug-resistant forms of
ADHD by suggesting importance of bottom-up processes in sus-
tained attention.
388 Dmitri V. Poltavski

4 Notes

1. It was our observation that with men a good quality signal is

obtained within seconds of the MindSet placement on the par-
ticipant’s head. With women we often experienced delays of up
to 10 min before the signal quality was acceptable for record-
ing. In every instance these delays were not caused by our fail-
ure to follow the NeuroSky-recommended application protocol.
For example, we would use alcohol wipes to pretreat electrode
application areas on the participant’s forehead and ear lobe,
wipe the electrodes of the MindSet and make sure there is no
interference from the participant’s hair, and that all extraneous
metal objects (i.e., ear-rings, studs, hair clips) are removed from
the ears and the head. We cannot at present explain these delays.
We do not know whether it has something to do with the cali-
bration procedure and/or the self-learning algorithm of the
ThinkGear module of the headset, but we changed our research
protocol to accommodate these delays (in about 95 % of the
cases the signal quality ultimately improved to acceptable “0”
levels). For example, we started placing the headset before the
participant started filling out baseline and demographic ques-
tionnaires, and before we conducted an optometric exam (total
duration approximately 10–15 min).
2. It is possible to save key-stoke events as the file is being
recorded. This may be useful for events for which ms-precision
latency is not critical (for instance to indicate the beginning
and the end of periods when subjects begin and finish a specific
cognitive task within a battery of tests or would attempt to
produce specific states of mind). To use this feature simply
press any key any time while recording is enabled, and a verti-
cal bar dissecting your EEG waveform will be placed on the
screen. When the entire recording is finished, each discrete
event will be automatically saved to a separate data file.
3. It is a good idea to visually inspect your power spectrum file
before you apply the SPSS syntax to it. In some instances, some
of the 0.25 bin data was missing. In these cases we used the
following syntax:
*Computing power for frequencies of the Theta wave EEG band.
COMPUTE theta_4_7HZ=(@4 + @4.5 + @5 + @5.5 + @6 +
@6.5 + @7 + @7.5) /8.
VARIABLE LABELS theta_4_7HZ ‘mean power within the
theta EEG band'.
EXECUTE.
*Computing power for frequencies of the Alpha wave EEG band.
COMPUTE alpha_8_12HZ=(@8 + @8.5 + @9 + @9.5 + @10
+ @10.5 + @11 + @11.5) / 8.
 The Use of Single-Electrode Wireless EEG in Biobehavioral Investigations 389

VARIABLE LABELS alpha_8_12HZ ‘mean power within the

alpha EEG band''.
EXECUTE.
*Computing power for frequencies of the Beta wave EEG band.
COMPUTE beta_12_20HZ=(@12 + @12.5 + @13 + @13.5 +
@14 + @14.5 + @15 + @15.5 + @16 + @16.5 + @17 + @17.5 +
@18 + @18.5 + @19 + @19.5 + @20 + @20.5 + @21+ @21.5 +@21
+ @21.5 + @23 + @23.5 + @24 + @24.5 + @25 + @25.5 + @26 +
@26.5 + @27 + @27.5 + @28 + @28.5 + @29 + @29.5)/ 36.
VARIABLE LABELS beta_12_30HZ ‘mean power within the
beta EEG band'.
EXECUTE.
*Computing amplitude for frequencies of the Gamma wave EEG
band.
COMPUTE gamma_30_60HZ=( @30+ @31 + @32 + @33 +
@34 + @35 + @36 + @37 +
@38 + @39 + @40 + @41 + @42 + @43 + @44 + @45 + @46 +
@47 + @48 + @49 +
@50 + @51 + @52 + @53 + @54 + @55 + @56 + @57 + @58 +
@59+ @60 + @30.5+ @31.5 + @32.5 + @33.5 + @34.5 + @35.5
+ @36.5 + @37.5 [email protected] + @39.5 + @40.5 + @41.5 + @42.5 +
@43.5 + @44.5 + @45.5 + @46.5 + @47.5 + @48.5 + @49.5
[email protected] + @51.5 + @52.5 + @53.5 + @54.5 + @55.5 + @56.5 +
@57.5 + @58.5 + @59.5+ @60.5) / 62.
VARIABLE LABELS gamma_30_60HZ 'mean power within
the gamma EEG band'.
EXECUTE.

References

1. Davidson RJ, Jackson D, Larson CL (2000) 5. Cao B, Li Y, Li H (2012) Electrophysiological

Human electroencephalogrphy. In: Cacioppo difference between mental state decoding and
JT, Tassinary LG, Berntson GG (eds) Handbook mental state reasoning. Brain Res 1464:53–60
of psychophysiology, 2nd edn. Cambridge 6. Luck SJ, Ford JM, Sarter M et al (2012)
University Press, New York, NY, pp 27–52 CNTRICS final biomarker selection: control of
2. Uhlhaas PJ et al (2011) A new look at gamma? attention. Schizophr Bull 38(1):53–61
High –(>60Hz) γ-band activity in cortical net- 7. Poltavski DV, Biberdorf D, Petros T (2012)
works: function, mechanisms and impairment. Accommodative response and cortical activity
Prog Biophys Mol Biol 105(1–2):14–28 during sustained attention. Vision Res 63:1–8
3. Jaworska N, Berrigan L, Fisher D et al (2012) 8. Borsting E, Rouse M, Chu R (2005) Measuring
A pilot study of electrocortical activity in dys- ADHD behaviors in children with symptom-
functional anger: decreased frontocortical acti- atic accommodative dysfunction or conver-
vation, impaired attention control, and gence insufficiency: a preliminary study.
diminished behavioral inhibition. Aggress Optometry 76:588–592. doi:10.1016/j.
Behav 38(6):469–480 optm.2005.07.007
4. Kramer UM, Solbakk AK, Funderud I et al 9. Granet DB, Gomi CF, Ventura R, Miller-Scholte
(2013) The role of the lateral prefrontal cor- A (2005) The relationship between convergence
tex in inhibitory motor control. Cortex insufficiency and ADHD. Strabismus 13:163–
49(3):837–849 168. doi:10.1080/09273970500455436
390 Dmitri V. Poltavski

10. Cools R, Aarts E, Mehta MA (2011) life applications. J Physiol Anthropol

Paradoxical effects of drugs on cognitive func- 28(3):145–150. doi:10.2114/jpa2.28.145
tion: the neuropsychopharmacology of the 15. Rebolledo-Mendez G, Dunwell I, Martínez-
dopamine and other neurotransmitter sys- Mirón EA, Vargas-Cerdán D et al (2009)
tems. In: Kapur N (ed) The paradoxical brain. Assessing NeuroSky’s usability to detect atten-
Cambridge University Press, Cambridge, tion levels in an assessment exercise. Paper pre-
pp 397–417 sented at the 13th International Conference on
11. Sheridan MA, Hinshaw S, D’Esposito M Human-Compute Interaction, San Diego, CA
(2010) Stimulant medication and prefrontal 16. Crowley K, Sliney A, Pitt I et al (2010) Evaluating
functional connectivity during working a brain-computer interface to categorise human
memory in ADHD: a preliminary report. J emotional response. Paper presented at the 10th
Atten Disord 14(1):69–78. doi:10.1177/ IEEE International Conference on Advanced
1087054709347444 Learning Technologies, Sousse, Tunisia
12. Loo SL, Hale TS, Macion J, Hanada G, 17. Gola M, Magniski M, Szumska I et al (2013)
McGough JJ, McCracken JT, Smalley SL EEG band activity is related to attention and
(2009) Cortical activity patterns in ADHD attentional deficits in the visual performance of
during arousal, activation and sustained atten- elderly subjects. Int J Psychophysiol http://
tion. Neuropsychologia 47:2114–2119 dx.doi.org/10.1016/j.ijpsycho.2013.05.007
13. NeuroSky, Inc (2009) Brain wave signal (EEG) 18. Davidson RJ, Chapman JP, Chapman LJ et al
of NeuroSky, Inc. http://www.neurosky.com/ (1990) Asymmetrical brain electrical activity
AcademicPapers.aspx. Accessed 12 Sep 2013 discriminates between psychometrically
14. Yasui Y (2009) A brainwave signal measure- matched verbal and spatial cognitive tasks.
ment and data processing technique for daily Psychophysiology 27:528–543
 Chapter 26

Smartphone Based Monitoring System for Long-Term

Sleep Assessment
Alexandre Domingues

Abstract
The diagnosis of sleep disorders, highly prevalent in Western countries, typically involves sophisticated
­procedures and equipment that are highly intrusive to the patient. The high processing capabilities and
storage capacity of current portable devices, together with a big range of available sensors, many of them
with wireless capabilities, create new opportunities and change the paradigms in sleep studies. In this
work, a smartphone based sleep monitoring system is presented along with the details of the hardware,
software and algorithm implementation. The aim of this system is to provide a way for subjects, with no
pre-­diagnosed sleep disorders, to monitor their sleep habits, and on the initial screening of abnormal
sleep patterns.

Key words Smartphone, Sleep parameters, Home monitors, ECG, Tachogram, Accelerometer, Sleep
Diary, Dream Diary

1  Introduction

Sleep disorders (SD) form a class of medical problems generally

characterized by changes of physiological or behavioral sleep pat-
terns. Their impact on both young and adult populations is well
documented and can be related with a wide range of short and
long-term consequences for the health of the subjects, including
anxiety, memory and cognitive impairments, high blood pressure,
obesity, and psychiatric problems, among others.
The golden standard for the diagnosis of SD is the
Polysomnography (PSG) [1], which is by far the most reliable and
accurate method. The Hypnogram, derived from PSG data, is a
graphical representation of sleep stages as a function of time.
Several parameters can be computed from the Hypnogram to
quantify and characterize sleep, such as the Sleep Efficiency (SE),
Sleep Onset Latency, REM sleep percentage (REMp), Non-REM
sleep percentage (NREMp), and REM latency.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_26, © Springer Science+Business Media New York 2015

391
392 Alexandre Domingues

However, PSG involves complex acquisition hardware and

long setup procedures. It is uncomfortable to the subject and is
usually done in clinical facilities. These highly constrained condi-
tions ­prevent its use in a non-intrusive way in normal daily life and
limit the duration of the typical exam, which is usually performed
over 1 or 2 nights.
These constraints, together with the generalized tendency to
adopt low-cost diagnosis solutions, have increased the demand for
reliable, accurate, and portable devices for sleep monitoring.
In the past, medical telemetry systems were mostly limited to
ambulatory telemetry in hospitals but the advent of smartphones
and tablets, equipped with large storage capacity and high process-
ing capabilities, together with a wide range of physiological data
sensors have allowed physiological and behavioral data to be
acquired outside clinical environments, often across several days.
The concepts of Body Area network and Body Sensor network
are now a reality and their application to healthcare is now being
fully explored by both research and commercial organizations.
Useful sources of data, for portable sleep monitoring purposes,
include the classical physiological sources, such as ECG, blood pres-
sure, and temperature, but also the behavioral patterns of the sub-
ject during the circadian cycle. These patterns can be assessed using
sleep and dream diaries and activity sensors, i.e., accelerometers.
In this work, a practical implementation of a smartphone based
sleep monitoring system is described. The system is composed of
the hardware, required for signal acquisition, and the supporting
software; (1) Sleep Monitor, (2) Sleep Diary, and (3) Dream Diary
and the algorithm responsible for the computation of three sleep
parameters (Sleep Efficiency, REM sleep percentage, and Non-­
REM sleep percentage) from physiological (ECG) and behavioral
data (accelerometer).

2  Materials

The following items were used on the implementation of the

described system.
1. BITalino: A Multimodal Platform for Physiological Computing
[2] (see Note 1).
2. A Bluetooth enabled smartphone/tablet running Android
operating system. The minimum recommended version of
Android is 2.2.
3. A personal computer with the Android Development Tools
(ADT) and any supported Integrated Development
Environment (IDE), e.g., Eclipse (see Note 2).
 Smartphone Based Monitoring System for Long-Term Sleep Assessment 393

3  Methods

The implementation of the smartphone based sleep monitoring

system is divided in hardware preparation and implementation of
the software and required algorithms.

3.1  Hardware The system relies on BITalino, a highly customizable physiological

Preparation signal acquisition platform. The device, shown in Fig. 1, was
adapted for the considered application. The following sections of
the board were included:
1. Micro-controller unit.
2. Bluetooth unit.
3. Power unit.
4. Light-emitting diode.
5. Accelerometer.
6. Electrocardiography.
The six sections were packed together and connected to three
pre-gelled electrodes that glue directly to the skin, resulting in the
configuration shown in Fig. 2. Since ECG morphology is not

Anatomy of a Bitalino
Micro-Controller Unit
up to 1000Hz sampling rate
6 analog inputs (4@10-bit + 2@6-bit)
4 digital inputs, and 4 digital outputs
Electromyography (EMG) Electrocardiography ECG
measures the electrical activity tracks the electrical changes
produced by your muscles associated with your heartbeat

Electrodermal Activity (EDA) Accelerometry (ACC)

senses your sympathetic nervous know your physical activity and
system reactions movement related information

Light-Emitting Diode (LED)

make your BITalino shine

Light (LUX)
know the changes in ambient
light or other light sources
Bluetooth
a widespread standard
for short-range wireless connectivity Power
so that you can operate
your creations

Fig. 1 BITalino is a customizable biosignal acquisition system, based on the Arduino hardware, currently being
used by several researchers due to its simplicity and reliability. The standard board contains several amplifying
and signal accommodation components, allowing the acquisition of: Electromyography, Electrodermal Dermal
Activity, Light Intensity, Electrocardiography, and Accelerometry. The board also included an LED, a program-
mable Micro-controller Unit, Bluetooth Unit, and a Power Stage. The different components can be included in
the system according to the specific requirements of the intended application
394 Alexandre Domingues

Fig. 2 The Sleep Monitor uses six sections of the BITalino board: The Micro-­
controller Unit, the Bluetooth, the Power Unit, the LED, the Accelerometer, and the
Electrocardiography Unit. The six sections were connected and configured in a
compact format

important for this application, precise placement of the device is

not fundamental. Figure 3 shows the recommended approximate
locations of the three electrodes (see Note 3).

3.2  Sleep Parameter The sleep parameter estimation method, displayed in Fig. 4, is
Estimation Software composed by the preprocessing and feature extraction procedures,
and Algorithm followed by the classification stage, designed to reject the ambigu-
ous features, and regularization.
The multimodal dataset includes the ECG, from which the RR
signal is computed, and Accelerometer (ACT) data. The prepro-
cessing operations are required to reduce the movement artifacts,
normalize the data across different patients, and prepare it for fea-
ture extraction.
ECG filtering and QRS complex detection is performed
according to the methods described in [4], the tachogram is then
constructed from the detected R peaks, and downsampled to 2 Hz.
Magnitude normalization and DC component removal are applied
to both RR and ACT data according to:

a (n) − µ (n)
ă ( n ) =
σ (n)

Smartphone Based Monitoring System for Long-Term Sleep Assessment 395

Fig. 3 Precise placement of the three electrodes is not fundamental but should
be in the area shown in the figure

where a(n) is the original sample, μ(n) and σ(n) are the mean and
standard deviation of the data within the 5 min window centered
at the nth sample and ă ( n ) is the normalized sample.
After preprocessing, each dataset is divided in contiguous
epochs of T = 30  s.
Let wj = [RRj; ACTj] represent a T dimensional window, con-
taining the multimodal data from the jth epoch, where j ∈ [1, …, M]
and M the total number of epochs.
The extracted features and the extraction procedures are the
following:
1. RR features: The RR frequency domain features are computed
in the LF = [0.015 … 0.15]Hz and HF = [0.15 … 0.4]Hz
bands [3]. In order to extract these features from each RRj an
eight-order autoregressive model (AR) is fitted to the extended
window RRj* = [RRj − 3, …, RRj] (see Note 4) and a set of opti-
mal coefficients aRR and a residual, ERR, are obtained. The
power spectrum is computed from the estimated AR coeffi-
cients and the following features are extracted:
PMj: Magnitude of the high frequency pole of the filter impul-
sive response filter (IIR) described by the coefficients aRR.
PPj: Phase of the high frequency pole.
ERRj: Residual of the AR model fitted to RRj*.
TPj: Total power (LF + HF).
HFj: Power in the HF range.
LFj: Power on the LF range.
LF/HFj: Power ratio between the two frequency bands.
MHRj: Mean heart rate on the considered RRj.
396 Alexandre Domingues

Fig. 4 After the acquisition of RR and ACT data (5-min segment shown on top of
the image) the sleep parameter estimation algorithm consists of (1) Preprocessing,
where the noise is filtered and the signal normalized, (2) Feature extraction,
where the information regarding the states of interest is extracted, (3)
Classification by two binary classifiers, having a rejection option for outliers and/
or ambiguous samples, and (4) Regularization allowing a precise computation of
the parameters by taking into account the number of rejected samples and the
accuracy of the classifiers. The acquired signals consist of: RR: R-to-R interval
tachogram, computed from the ECG and ACT—Accelerometer data, a measure-
ment of the user’s activity

2. ACT features: The features extracted from ACT are computed

from each window ACTj* = [ACTj − 3, …, ACTj + 3] (see Note 5)
centered on the jth epoch.
The following features are extracted:
EAR j
AR—Coefficients (a{1,..,4}j) and residue ( ) of a four-order AR
model fitted to ACTj* .
RMM—Weights (w{1,..,3}j), parameters (r{1,..,3}j) and the Kullback–
Leibler (KLj) divergence of the Rayleigh Mixture Model
(RMM) [5] distribution fitted to ACTj*.
∑h ( k ) x ( k )
2
Magj—The energy of x(k) = ACTj* given by where
h(k) is a Hanning window. k

In order to minimize the inter-patient variability, a normalization

operation was performed. Let fij denote the vector containing
 Smartphone Based Monitoring System for Long-Term Sleep Assessment 397

all the samples from feature i and subject j, the normalization

is performed according to
1
f ij ( n ) = fij ( n ) − µij
−
σ ij
1+ e
where f ij ( n ) is the nth normalized sample and μij and σij the
mean value and standard deviation of fij respectively. This nor-
malization step ensures that all features fall in the range [0, …, 1]
The estimation of the Sleep Parameters is based on two binary
classifiers that independently classify all the samples into (i) Sleep/
Wakefulness—SW classifier and (ii) REM/Non-REM—RN classifier.
In machine learning and statistics, classification is generally
characterized as way to identify a class or label to which a given
observation or sample belongs.
In this chapter, the term classification is used in the context of
supervised learning. This is, a group of labelled data is used to infer
a discriminative rule which then allows new, unlabeled data, to be
classified into one of the considered classes.
Each classifier is designed to take into account a rejection fac-
tor (RF), rejecting a specified percentage of samples, whose classi-
fication is ambiguous.
In large biomedical datasets, such as the one considered, the
systematic rejection of unreliable segments and/or samples has
been shown to increase the accuracy of the classification proce-
dures without compromising the overall result [5]. The rejection
works by computing the true or estimate posterior probability of
the most probable class for each sample and rejecting those which
are below the specified percentage. The rejection factor was defined
as 10 % (see Note 6).
Therefore each classifier maps each sample into one of three
classes: SW ∈ {sl, wk, r} and RN ∈ {rs, ns, r} where sl, wk, rs, ns, and
r refer to Sleep, Wakefulness, REM sleep, Non-REM sleep, and
Rejected sample, respectively.
During the training step (see Note 7), the two classifiers are
trained only with data from the two considered classes (Sleep;
Wakefulness for SW classifier and REM; Non-REM for RN classi-
fier). However, during the test, they map samples belonging to
three classes (Wakefulness; REM; Non-REM). Any sample from a
class not predicted by the classifier will either be miss-classified or
rejected.
The Support Vector Classifier (SVC) with a quadratic kernel
(see Note 8) was chosen for the classification tasks. Figure 5 shows
an example classification problem, for illustration purposes each
sample is limited to two features. The 2D scatter plot shows the
two features and the SVC, trained with these two features, together
with the rejection areas. The presented data corresponds to real
data, acquired with the described system, the presented features
398 Alexandre Domingues

Fig. 5 The figure shows the 2D scatter plot of two features (Mean Heart Rate and LF/HF ratio) and the Support
Vector Classifier, trained with these two features, together with the rejection areas. The figure illustrates a
binary classification problem, where each sample is classified as Sleep, Wakefulness, or Rejected. The shown
data corresponds to real data, acquired with the described system, but for illustration purposes the description
of each sample is limited to two features: Mean Heart Rate and LF/HF ratio. Every sample falling in the area
above the Wakefulness rejection boundary will be classified as Wakefulness; every sample falling below the
Sleep rejection boundary will be classified as Sleep. Samples falling between the two boundaries will be
rejected. The “+” and “*” markers shown the true class of each sample. All the samples inside the rejection
areas are considered ambiguous from a classification point of view and thus discarded. The plotted data cor-
responds to real data, acquired with the described system

correspond to the mean heart rate and the LF/HF ratio during
Sleep and Wakefulness states.
An accurate computation of the sleep parameters implies a reli-
able estimation of the number of epochs (samples) during each sleep
state (Wakefulness/Sleep; REM/Non-REM). Since the output of
the classifiers is affected by an error and since a certain number of
samples are rejected during this classification, a regularization is
required to properly estimate the number of samples on each state.
The final step in the computation of the sleep parameters is
thus the regularization of the classifier output. The regularization
operation corrects the estimated number of Sleep, Wakefulness,
REM and Non-REM epochs by taking into account the p ­ ercentage
of mis-classified samples and an estimation of the number of sam-
ples that were rejected by the classifiers on each class.
Consider a binary classifier C, with a reject option that maps
each sample into one of three labels l ∈ {p, n, r}where p, n and r
denote Positive, Negative and Reject.
 Smartphone Based Monitoring System for Long-Term Sleep Assessment 399

The confusion matrix is represented as

Tp Fn Rp 
A= 
 Fp Tn Rn 

with Tp, Fn, Fp, Tn, Rp and Rn the True Positives, False Negatives,
False Positives, True Negatives, Rejected Positives and Rejected
Negatives respectively.
The positive (θp,i) and negative (θn,i) correction factors and the
fraction of rejected samples per class (ωp,i and ωn,i) are computed
for each training dataset as
Tpi + Fpi
θ p,i =
Tpi + Fn i

Fn + Tn i
θ n ,i = i
Fpi + Tn i

Rpi
ωp,i =
Rpi + Rn i

Rn i
ωn ,i =
Rpi + Rn i

With i ∈ [1, …, M] and M the number of training sets.
Let N(.) represent a counting operator, the number of pre-
dicted samples in each class can be corrected taking into account
the performance of the classifier as
N ( p)
N ( p˘ ) =
θp

And estimating the number of rejected samples from each class as

N ( rp ) = ωp N ( r )

The expressions used to calculate the Sleep parameters finally
become:
N ( s˘ )
SE =
N ( s ) + N ( w)

NREM p =
( )
 + N (r )
N ns ns

( N ( ns ) + N ( rs ) + N ( r ) ) × SE
400 Alexandre Domingues

REM p =
( )
 + N (r )
N rs ns

( N ( ns ) + N ( rs ) + N ( r ) ) × SE
with SE computed from the output of the SW classifier and NREMp
and REMp from the RN classifier. The three parameters are pre-
sented to the user in percentage.
The described system and sleep parameter estimation method
was validated with a group of 25 healthy volunteers. Each volun-
teer performed a standard PSG, supervised by a trained technician.
The Hypnogram was used as the ground truth, to properly identify
the correct state on each epoch. The sleep parameter estimation
algorithm was tested using a “Leve-One-Patient-Out” policy, i.e.,
the computation of the regularization parameters and the training
of the classifiers was performed considering all the datasets, except
one, which was then used to test the algorithm. This process was
repeated for all the subjects. This means that the classification and
computation of sleep parameters is always performed on data pre-
viously unseen by the algorithm.
The described method achieves an average estimation error of
approximately 4, 10, and 5 % in the estimation of the SE, REM,
and NREM percentages, respectively, when compared with the
values obtained from PSG data.
The estimation error is computed as
true − estimated
error = ×100
true
Where true is the true value of the parameter and estimated the
value computed by the algorithm.
The structure of the sleep monitoring software is shown in
Fig. 6. A new record is started on each night; the system will start
acquiring the ECG and Accelerometer data, and display them in
real-time together with an estimate of the power spectrum of the
RR signal (Heart Rate Variability). After each acquisition, the data
is processed and the sleep parameters computed and stored.

3.3  Sleep The scheme of the Android implementation of the Sleep and
and Dream Diary Dream Diaries are shown in Figs. 7 and 8 The Sleep Diary has a
simple structure, with a main menu from which Events can be cre-
ated, displayed and deleted. The Insert Event menu allows the cre-
ation of a new event, from a list of standard types. Each event and
respective information is stored in a sql database, the implementa-
tion of this database is supported by Android through SQLite, a
software library that implements a self-contained, serverless, zero-­
configuration, SQL database engine.
The Dream Diary follows the same structure of the Sleep Diary
although with more complex options. The first option “Record
 Smartphone Based Monitoring System for Long-Term Sleep Assessment 401

Fig. 6 Structure of the sleep monitor software. A simple menu allows to create a New Record or retrieve
a previous record. After each record the data is processed and stored in a local database

Fig. 7 Structure of Sleep Electronic Diary (SeD). All the entries are stored in a database, common with the sleep
monitor and DeD
402 Alexandre Domingues

Fig. 8 Structure of Dream Electronic Diary (DeD). All the entries are stored in a database, common with the
sleep monitor and SeD

Dream” starts a new voice recording that is saved together with

date information. The Dream can be then be transcribed into text,
using any available speech-to-text service. Simple data analysis can
be applied to the transcribed dream (e.g., word occurrence count)
to look for the most relevant terms, which often have clinical
relevance.
The remaining options of the menu allow the user to listen,
read or delete a given dream.

4  Notes

1. The described system requires a reliable device to acquire the

physiological signals that is able to transmit them in real time
and, preferably, that is able to store the data on its internal
memory. The availability of a public API is also fundamental to
properly control the data that is acquired. The chosen device,
BITalino, has all the characteristics mentioned above and
allows a future extension of the platform by including other
physiological signals. Another tested and recommended com-
mercial device is the Bioharness, by Zephyr Technology.
2. It is recommended to develop the signal processing algorithm
(i.e., sleep parameter estimation) on a different platform, such
as Matlab or R, in order to easily debug and test the algorithm,
before writing the Java version. It is also possible to develop
 Smartphone Based Monitoring System for Long-Term Sleep Assessment 403

the algorithm in other language, such as C and C++, test it on

Matlab and import it directly to the Android project. The
details of this procedure are outside the scope of this chapter.
3. Since precise electrode placement is not rigid, the polarity of
the acquired ECG signal must be taken care in the preprocess-
ing procedures. Many devices, e.g., BITalino, do this correc-
tion by default.
4. A 2-min window was chosen since it is the most common size
referred in the literature. This window introduces a smoothing
effect in the feature extraction, for specific applications, e.g.,
when higher temporal resolution is required, smaller windows
can be used, taking into account that lower frequencies might
be cut-off.
5. A window larger than the epoch size (T = 30 s) is required for
most feature extraction steps. The 3.5 min window, defined for
the accelerometer signal was found to be optimal, leading to
the highest accuracy rates.
6. Several rejection factors were tested and 10 % resulted in a
compromise between high classifier accuracy (which increases
in the rejection factor) and a low number of samples rejected
(low rejection factors result in less discarded samples).
7. The classification task described for this system is a problem of
supervised learning. This implies that the reader must have sev-
eral sets of data, obtained in controlled conditions (i.e.,
Polysomnography exams, with hypnograms elaborated by
trained technicians) in order to train the classifiers. An unsuper-
vised learning approach was also considered but quickly aban-
doned due to the high complexity of the classification problem.
8. Several classification libraries are freely available to download:
PRTools, for Matlab has a quick learning curve and includes all
the major classifiers; Weka is a popular collection of machine
learning algorithms for data mining tasks, available for Java
and MLC++ is a library of C++ classes for supervised machine
learning.

References
1. Kushida CA et al (2005) Practice parameters 4. Seabra JC et al (2011) Rayleigh mixture model
for the indications for polysomnography and for plaque characterization in intravascular
related procedures: an update for 2005. Sleep ultrasound. IEEE Trans Biomed Eng 58(5):
28(4):499–521 1314–1324
2. Guerreiro J et al (2013) BITalino: a multimodal 5. Lewicke A, Sazonov E, Corwin MJ, Neuman M,
platform for physiological computing. Schuckers S (2008) Sleep versus wake classifica-
Proceedings of the international conference on tion from heart rate variability using computa-
informatics in control, automation and robotics tional intelligence: consideration of rejection in
(ICINCO), Reykjavik, Iceland classification models. IEEE Trans Biomed Eng
3. Clifford GD (2002) Signal processing methods 55(1):108–118
for heart rate variability analysis. Ph.D. disserta-
tion, St Cross College
 Chapter 27

Intracranial Ventricular Catheter Placement

with a Smartphone Assisted Instrument
Ulrich-W. Thomale

Abstract
Mobile technology has recently been introduced for blood pressure measurements or glucose level con-
trols. In surgical disciplines the use of smartphone applications is mostly restricted as training tools or
knowledge resources.
Simple surgical procedures which are performed often in certain disciplines may be performed with
limited accuracy since routine and overwork of medical staff lead to less awareness to possible mistakes. In
these cases simple and effective means are necessary to achieve better patient safety. In this context, a surgi-
cal instrument for ventricular catheter placement in neurosurgical patients was designed which is assisted
by measurements undertaken in a smartphone software application specifically visualizing the use of this
instrument and achieving better accuracy for catheter positioning. On theoretical ground, the angulation
of the catheter trajectory towards the surface of the skull in a coronal reconstructed CT or MR image is
determined as the simplified but the most relevant individual parameter for correct ventricular catheter
placement. Transfer of a CT/MRI image onto the smartphone can be performed via mail as anonymous
file. Using this image, the trajectory measurement can be performed individually in a few steps by calibra-
tion of the image size, definition of the frontal entry point, and virtual placement of the instrument on the
surface of the skull. Then the angulation can be adjusted and measured to place the catheter’s trajectory
towards the ipsilateral ventricle and the catheter length is determined. The parameters are now given by
the app and visualized on the image in order to be applied to the surgical site of the patient.
The tool represents a widely available and cost-effective solution as navigation technique which is
simple to apply in order to achieve better accuracy in ventricular catheter placement for higher safety in a
large cohort of neurosurgical patients.

Key words Hydrocephalus, Ventricular catheter, Neurosurgery, Smartphone application, Neuro-

navigation

1 Introduction

Smartphones and tablets are information technology resources

widely available around the world. The introduction of application
software tools has enhanced practical issues in daily life. Mobile
availability of information resources such as encyclopedia, news
agencies, social media, and organizing and communication tools,

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_27, © Springer Science+Business Media New York 2015

405
406 Ulrich-W. Thomale

e-mail service, telephony, social media and cloud services, and

planning tools such as traveling agencies, navigation map resources
as well as service engines including banking tools, purchase appli-
cations has changed our daily behavior dramatically. Mobile tech-
nology has recently been introduced in medicine for different
purposes such as knowledge resources, cardiac monitoring [1],
and blood glucose level detection [2]. For surgical disciplines the
integration of mobile technology is still limited [3] and mostly
restricted to knowledge or training resources [4].
Computer technology is used in surgical applications since the
beginning of the 1980s. Herewith virtual planning of surgical pro-
cedures on the basis of anatomical imaging was followed by the
direct transfer of a virtual plan to surgical performance. Among
other disciplines Neurosurgery was one of the most important
medical fields, who introduced computer assistance as the so-called
neuronavigation systems in order to increase patient safety. This was
necessary predominantly to target deep seated lesions in the brain,
to enhance minimal invasive approaches and avoid surgical related
morbidity, while approaching the lesion through unaffected brain
tissue [5]. The software technology enabled the spatial reconstruc-
tion of volume image datasets along all necessary sections in order
to virtually visualize the optimal entry, trajectory and target. With
optical camera systems or electromagnetic fields the registration of
the patient anatomy in the operation room could be co-registered
to the image volume data sets. That enabled the surgeon to visual-
ize any adjacent anatomical structures before exposing them by
pointing a spatially tracked instrument on the surface of the patient´s
skin by computer-assisted visualizing of the pointers position on the
anatomical image data sets and its related structures underlying the
skins surface. For the neurosurgeon this technology became routine
in daily practice during the past decades. It is important to notice
that the preparation of such a system for surgery lasts a time period
of 20–40 min depending on the experience of the surgeon and the
operational structure of the system. Thus the system is used not in
all procedures in neurosurgery but only in cases in which the user
defines the necessity to enhance anatomical guidance through such
a system. Some procedures might profit from such a system but the
relation of preparation time and time of surgery during a simple
procedure may not be reasonable.
We have recently introduced a medical product of a neurosurgi-
cal instrument together with an iPhone/iPad application in order
to enhance the reliability of intracranial ventricular catheter posi-
tioning via a frontal entry point for external drainage, cerebrospinal
fluid shunts, and Rickham/Ommaya reservoirs [6]. The surgical
procedure of placing such a catheter is one of the most frequent
procedures in neurosurgery and is generally classified by neurosur-
geons to be a simple procedure. It is taught to young residents dur-
ing their first year of training. However, the rate of ventricular
 Ventricular Catheter Placed with Smartphone Assistance 407

catheter malpositioning has often been neglected. Looking at the

recent published data this malplacement rate was detected between
12.5 and 44 % [7]. Neuronavigation systems are used for this pro-
cedure in limited cases with narrow ventricles for long standing
implants with high technical and cost intensive efforts. In a survey
among neurosurgeons it was stated that a neuronavigation system
would not be used for this indication at all in about one-third. Less
than two-thirds would use such a system if the additional time effort
would not prolong the procedure by 5 to 10 min, while only 7 %
would use it when the time effort exceeds 10 min [8]. In this con-
text a mobile technology approach for guiding the placement of
ventricular catheters appears to be reasonable in order to use a
widely available technology of smartphones with time-efficient soft-
ware in order to gain better patient safety on a broad basis.
The theoretical ground for the proposed technique is to sim-
plify the parameters to the relevant measures for ventricular cath-
eter placement. It was identified that the angulation of the catheter
trajectory towards the surface of the skull in the coronal plane is
the relevant individual parameter for correct placement of ventric-
ular catheter. Due to the shape the ventricular system being parallel
in its surface to the frontal skull in longitudinal (sagittal) section
does simplify this measure with an angulation in this direction
being always 90°. Thereby, the individual parameter to be detected
remains the angulation of the trajectory towards the skull surface
in the frontal (coronal) section since the paramedian surface of the
skull has an individual decline (Fig. 1). This idea could be used in
order to design an instrument in which a tube guides the ventricu-

Fig. 1 Schematic drawing of the ventricular anatomy in relation to the brain and skull surface depicted with a
ventricular catheter entered via a frontal (precoronal) paramedian bur hole (a). The possible trajectories are
given in simplified frontal/coronal (b) and longitudinal/sagittal (c) view. The individual measurement of the
trajectory’s angulation in the coronal plane is necessary for correct placement of the ventricular catheter. In
the sagittal plane the insertion angle towards the skull surface is always 90° using a frontal entry point
408 Ulrich-W. Thomale

lar catheter rectangular in the longitudinal (sagittal) orientation

and with an individual angle adjustable in frontal (coronal) orienta-
tion (Fig. 2). Measurements are necessary on individual patient
images like CTs or MRIs in order to apply this instrument cor-
rectly during surgery. In our first experiences we realized that this
measurement is difficult to perform in regular DICOM viewing
software (digital imaging and communication in medicine, ISO
12052) and thus the hypothesis was not easy to sell to the neuro-
surgical community. Thus it was planned to apply a software tool
to perform the necessary measurement for the specifically defined
instrument. This was achieved in a smartphone software tool also
known as “app” (Fig. 3) and represents some major advantages for
its performance: (1) The availability of the mobile devices for the
surgery is not restricted to expensive planning workstations, (2)
the planning procedure after data transfer can be performed inde-
pendent to any location, basically on the way to the OR, and (3)
The simple and efficient procedure of planning with the software
leading to a virtual visualizing the surgical instrument on the patient’s
anatomy will motivate the user to safely apply this technique in a

Fig. 2 The ventricular catheter guide with the guiding tube, and the base which
consists of a semicircular rod with angulation marks and a base. The angulation
can only be adjusted in one plane. The guide must be aligned such that the fixed
angulation of 90° will be applied in sagittal orientation while the individual angu-
lation will be applied in coronal plane
 Ventricular Catheter Placed with Smartphone Assistance 409

Fig. 3 Smartphone application software is available in the Apple App Store for iPhone and iPad

high number of patients. Other available software applications

being able for angulation measurements (e.g., Angle Meter PRO,
Angle Meter, bubble level), are not sufficient to fulfill this task, due
to different reasons: (1) Medical product regulations need to be
adhered; (2) Procedure dependent visualization of the instrument
as well as different safety advices need to be implanted; and (3)
Training tools for surgeons being unacquainted to this technique
should be integrated. Thus, a surgical instrument was directly
linked to specifically designed smartphone/tablet software applica-
tion and received the CE license as medical product in Europe,
which should be introduced in this manuscript.
The smartphone assisted placement of ventricular catheters is a
mobile IT device the surgeon may carry with him daily and be freely
available without sophisticated efforts. This is especially important
if a surgical technique is meant to be simple but has some risk of
failure and only a few measures are necessary in order to increase
patient safety. Since these measurements are not necessarily offered
easily by regular viewing software on DICOM standard (digital
imaging and communication in medicine, ISO 12052) it can be
well provided by specially designed and user friendly mobile appli-
cation software run on mobile devices like iPhone or iPad.
410 Ulrich-W. Thomale

2 Materials

1. Image acquisition: Patient’s cranial imaging acquired by com-

puter tomography (CT) or magnetic resonance imaging (MRI)
are sliced in coronal orientation, parallel to the body axis
2. Image data: An anonymized, coronal image section is needed at
the level anterior to the third ventricle, where the anterior horns
of the lateral ventricles are well depicted (see Note 1). This sec-
tion should be in close proximity to the chosen frontal entry
point and the target inside the frontal horn should well be iden-
tifiable (see Note 2). Hereby, no additional reconstruction of
the image is necessary in MRI (Fig. 4a). For CT scanning a spi-
ral thin sliced cranial CT is warranted, which needs coronal
reconstruction by radiologist since this is mostly not part of the
routine procedure. It has to be noted that some inaccuracy must
be calculated when measurements are performed in a strict coro-
nal image section since the trajectory in this image will be a
projection from the more oblique alignment of the catheter tra-
jectory. This however is mostly sufficient for a moderate ven-
tricular enlargement. Alternatively, in markedly narrow ventricles
an individual reconstruction of the image dataset along a pre-
planned trajectory for the ventricular catheter must be used, in
order to measure the trajectories’ angulation towards the skull
surface at the correct entry point (Fig. 4b).

Fig. 4 MRI images prepared for transfer to the software app. (a) Coronal image with moderate ventricular
enlargement. (b) Reconstructed image along the catheter trajectory as coronal like section. The trajectory of
the catheter is depicted as solid line towards the ventricle. The bitemporal diameter is measured before trans-
fer in order to warrant image size calibration
 Ventricular Catheter Placed with Smartphone Assistance 411

3. Mobile device: The iPhone, iPad, or iPad mini (Apple Inc.,

USA) is used to perform the measurements within the respec-
tive image using the software application (Thomale Guide
App, Miethke-Aesculap, Germany). The software is currently
designed for iOS 6.1 or higher. The user is responsible for the
proper application of the software as well as the handling of
personal data of the patients (Fig. 3).
4. Guiding instrument: A guiding instrument, the so-called
Thomale Guide (Miethke-Aesculap, Germany) consists of a
selection of titanium tubes with varying inner diameters being
equivalent to the outer diameters of the available ventricular
catheters (Fig. 2). One of the tubes is attached to a semicircular
rod which is fixed on the base of the instrument and may be
moved along the rod with a slider and can be fixed in various
angle positions with a screw. This angulation represents the
coronal angle of the catheter trajectory towards the surface of
the skull where the entry point is located and the instrument is
placed on. The base of the instrument is an open ring where the
semicircular rod is mounted to. A linear mark on the base which
is placed rectangular to the semicircular rod defines the longi-
tudinal alignment for proper placement of the skull. An open
window within the base may be used on the lateral side in order
to view the catheter entering the intracranial space without
touching the dura mater and to enter with a forceps to hold the
catheter in place in its final position.

3 Methods

1. Image data transfer: The coronal oriented image section (coro-

nal or coronal like along the trajectory as CT or MRI) must be
transferred to the mobile device (iPhone/iPad, see Note 3).
Firstly, an image of the section must be generated as .jpg, .tif,
.bmp, or .png file and saved on the computer. The transfer may
then be performed by two possibilities. Either it is attached to
an e-mail or is sent to the account which is accessible on the
mobile device. The image will then be integrated on the cam-
era roll and may be imported into the Thomale Guide App.
The transfer to the mobile device can also be done by connec-
tion to the computer with an USB cable. The file manager can
detect one folder on the mobile device which is the camera
role. By drag and drop or copy paste the image file can directly
be integrated on the mobile devices’ camera roll to be imported
into the App (Fig. 5)
2. Application software: Within the application software
(Thomale Guide App) a main menu gives an overview about
the different functions and resources. Herein, different infor-
412 Ulrich-W. Thomale

Fig. 5 Workflow for image data transfer to the mobile device, which can be established by mail or with an USB
cable. Image may be given as .png, .jpeg, .bmp, or .tif format

mation can be requested. Beside general warning notices and

company information the most valuable section is the user
manual. Here the instruction for the surgical instrument and
the software application tool can be downloaded. In addition
a video of a representative surgical procedure for a guided
catheter placement may be viewed.
In order to start the application for individual measure-
ments a picture must be opened from the camera roll. The
image must then be adjusted to an overlay in size and orienta-
tion followed by calibration of the image which is related to
the bitemporal diameter of the section. In the next step the
entry point will be defined with paramedian localization in
alignment of the medial limb of the visualized frontal horn
towards the surface of the skull. Then, the points of contact of
the guiding instrument with a size of 2 cm in diameter will be
placed on each side 1 cm apart from the entry point precisely
on the surface of the skull. This will define the virtual position
of the instrument being placed on the bone over the entry
point. The baseline trajectory will be given in 90° angulation
towards the skull surface. Now a sliding bar enables the user to
pivot the trajectory from the entry point to its optimal position
targeting the ipsilateral frontal horn in between the septum
pellucidum and the nucleus caudatus. The angle deviation
from 90° will be given at the bottom of the image. As addi-
tional measurement any length from the entry point may be
measured. That is performed for the distance to midline (entry
point position) as well as for the length of the ventricular cath-
eter. The latter one will be determined by setting the catheter´s
tip location inside the ventricle on the selected trajectory. This
result will also be given next to the calculated angle (Fig. 6).
The image calculation can now be saved in an archive where all
patient´s results are restored for intraoperative performance.
 Ventricular Catheter Placed with Smartphone Assistance 413

Fig. 6 Measurements as performed in the iPhone app in two representative cases. (a) Angulation of the trajec-
tory (red dotted line) towards the skull surface is 5.8° as deviation from 90° (light blue dotted line) with a tilt
of the guiding tube towards the midline of the patient. Catheter length is 71 mm (green dotted line). (b) In the
trajectory’s inline view reconstruction image the angulation is 6.8° tilt towards midline. The catheter length is
determined with 64 mm (Color figure online)

Basically, the App includes the following functions: (1)

Measurements for catheter placement by integrating an image
from the camera roll; (2) A camera tool in order to take a pic-
ture from medical imaging. Officially, only supposed to be
used for demonstration purposes, but may be helpful in emer-
gency situations or in technically less supplied world regions
were digital image data transfer functionality is not available;
(3) An archive option to safe and reopen all performed mea-
surements; (4) Training and user manual resources for reassur-
ing correct usage.
3. Surgical steps: For surgery the entry point is predefined at the
patient´s anatomy (see Note 4). The coordinates are related to
the midline and the nasion (the point of nasal impression
between the eyes, where the top of the nose meets the fore-
head). Thereby the distance to nasion will be at 11.5 cm (may be
less in children) while the distance to midline will be deter-
mined by the measurements in the software as distance to
414 Ulrich-W. Thomale

Fig. 7 Coordinates for the entry point are given as distance from nasion and
distance from midline. While the latter one can be determined in the software
app (usually 2–3 cm), the distance to nasion may be at 11.5 cm

midline and is usually 2–3 cm (Fig. 7, see Note 5). At this

point the skin incision will be performed to expose sufficient
area of skull in order to place the base of the guide with a
diameter of 2 cm on the bone (see Note 6). The bur hole will
be drilled and the dura mater exposed. An incision will open
the dura and a pin point coagulation of the exposed brain
cortex is necessary to acquire the entry point for the catheter.
Now the guide will be adjusted to the respective angulation of
the catheter trajectory towards the surface of the skull. The
guide will be placed on the skull over the bur hole with orien-
tation of the marks on the guide’s base being parallel to mid-
line and applying the individual adjusted angle along the
coronal plane (see Note 7). The catheter will now be inserted
in the guiding tube and inserted in the brain tissues entry
point avoiding any contact to bone or dural edges (Fig. 8a, see
Note 8). Through the lateral “window” of the guide the
length markings on the catheter can be detected while insert-
ing the catheter to the ventricle (see Note 9). As soon as the
precalculated length of the catheter is reached the catheter
stylet will be removed and the CSF flow will be verified. An
anatomical forceps which is inserted also through the guides
“window” may fix the catheter in place while then removing
the guide from the catheter (Fig. 8b). All additional surgical
steps necessary for catheter connection to a shunt or to exter-
nalize it may then be performed as usual.
 Ventricular Catheter Placed with Smartphone Assistance 415

Fig. 8 Intraoperative application of the guiding instrument. (a) The catheter is

inserted with the guide through the bur hole by observing the free pass into the
brain tissue avoiding any contact to the dural edges. (b) In the lateral “window”
of the guide a forceps may be inserted in order to fixate the catheter in place
after placement and removing the guiding instrument

4 Notes

1. The patient data must be transferred to a mobile device with-

out individual patient’s information of name or date of birth.
The user is responsible not to carry any personal information
of any patient on his smartphone.
2. For calculation of the insertion parameters only coronal cCT
or cMRI images must be imported and the user must be aware
to use those images with the correct orientation either long the
body axis in close proximity to the entry point or as recon-
structed coronal sections along the trajectory.
3. The photo capture feature may be used in selected cases but is
only recommended for demonstration purposes, since any tilt
between the camera and the taken image may cause inaccuracy
in further measurements.
416 Ulrich-W. Thomale

4. The technique is intended to be applied only for a frontal entry

point but not an occipital or parietal one.
5. The correct side of the patient where the measurements are
performed being left or right must be equivalent to the side of
surgery since asymmetry in anatomy may cause inaccuracy of
the applied measures.
6. The surface of the skull must well be prepared and freed from
periosteum in order to achieve a plane positioning of the guide
solely on the bone and not being tilted by any surrounding soft
tissue.
7. The catheter guide must be placed on the skull in correct align-
ment parallel to the midline in order apply the adjusted angula-
tion to the coronal plane. Hereby, it is important to avoid the
inversion of the guide’s position on the skull. That may happen
when the surgeon does recognize the angulation side of the
catheter but not of the device being the opposite since it pivots
around the entry point.
8. During surgery the free pass of the catheter into the brain tis-
sue must be guaranteed since any contact to adjacent tissue
around the bur hole such as the bone or the dura layer may
cause deviation of the catheters trajectory.
9. The “window” of the guide should be accessible by the sur-
geon in order to view the correct insertion of the catheter and
to fixate the catheter after correct placement by a forceps while
retracting the stylet and the guide.
The correct placement of ventricular catheters has been more
and more addressed in the literature in the past years. Different
studies have stressed that malplacement of ventricular catheters
especially in lifelong implants of cerebrospinal fluid diverting shunt
system is correlating with the implant survival and necessity for
surgical revisions [9]. In a surgical sophisticated discipline such as
neurosurgery, an apparently simple procedure of placing a ventric-
ular catheter is may be not sufficiently respected, since malplace-
ment rates are reported to be up to 44 % in recent literature and
the widely used free hand technique was defined as risk factor for
improper placement [7]. For transient therapeutic options with
external ventricular drainage, incorrect placement of the catheter
may influence sufficient cerebrospinal fluid drainage or incorrect
measurement of intracranial pressure. Technical advances are often
cost-intensive and time-consuming to be set up for a relative short
lasting surgical procedure. Hence, the proposed technique of a
guiding tool assisted by a mobile health software application may
further enhance quality of treatment and patient’s safety for an
every-day use in neurosurgery.
 Ventricular Catheter Placed with Smartphone Assistance 417

References
1. Carter T, O’Neill S, Johns N et al (2013) 6. Thomale UW, Knitter T, Schaumann A et al
Contemporary vascular smartphone medical (2013) Smartphone-assisted guide for the place-
applications. Ann Vasc Surg 27(6):804–809 ment of ventricular catheters. Childs Nerv Syst
2. Brooke MJ, Thompson BM (2013) Food and 29(1):131–139
Drug Administration regulation of diabetes- 7. Wilson TJ, Stetler WR Jr, Al-Holou WN et al
related mHealth technologies. J Diabetes Sci (2013) Comparison of the accuracy of ven-
Technol 7(2):296–301 tricular catheter placement using freehand
3. Naftel RP, Safiano NA, Falola MI et al (2013) placement, ultrasonic guidance, and stereotac-
Technology preferences among caregivers of tic neuronavigation. J Neurosurg 119(1):
children with hydrocephalus. J Neurosurg 66–70
Pediatr 11(1):26–36 8. O’Neill BR, Velez DA, Braxton EE et al (2008)
4. Jenny JY (2013) Measurement of the knee flex- A survey of ventriculostomy and intracranial
ion angle with a Smartphone-application is pre- pressure monitor placement practices. Surg
cise and accurate. J Arthroplasty 28(5):784–787 Neurol 70(3):268–273
5. Orringer DA, Golby A, Jolesz F (2012) 9. Hayhurst C, Beems T, Jenkinson MD et al (2010)
Neuronavigation in the surgical management of Effect of electromagnetic-navigated shunt place-
brain tumors: current and future trends. Expert ment on failure rates: a prospective multicenter
Rev Med Devices 9(5):491–500 study. J Neurosurg 113(6):1273–1278
 Part III

mHealth Cancer Imaging Technologies

 Chapter 28

High-Resolution Microendoscope for the Detection

of Cervical Neoplasia
Benjamin D. Grant, Richard A. Schwarz, Timothy Quang,
Kathleen M. Schmeler, and Rebecca Richards-Kortum

Abstract
Cervical cancer causes 275,000 deaths each year with 85 % of these deaths occurring in the developing
world. One of the primary reasons for the concentration of deaths in developing countries is a lack of effec-
tive screening methods suited for the infrastructure of these countries. In order to address this need, we
have developed a high-resolution microendoscope (HRME). The HRME is a fiber-based fluorescence
microscope with subcellular resolution. Using the vital stain proflavine, we are able to image cell nuclei
in vivo and evaluate metrics such as nuclear-to-cytoplasmic ratio, critical to identifying precancerous epi-
thelial regions. In this chapter, we detail the materials and methods necessary to build this system from
commercially available parts.

Key words High resolution imaging, Cancer detection, Low-resource settings

1 Introduction

Cervical cancer is a leading cause of cancer and cancer-related

deaths among women worldwide, with over 500,000 new cases
and 275,000 deaths occurring annually worldwide [1]. More than
85 % of the cases of deaths occur in low and middle-income coun-
tries, largely due to a lack of effective screening and secondary cer-
vical cancer prevention programs. In the USA and other
high-income countries, the incidence and mortality has decreased
by approximately 70 % over the past 40 years [2]. This decline is
largely due to the introduction in 1941 of the Papanicolaou (Pap)
smear, which has led to a systemic effort to detect early cervical
cancer and precancerous lesions [3].
Current approaches in high-income countries include screen-
ing with Pap and/or HPV testing. Patients with abnormal results
undergo colposcopy with biopsies of abnormal appearing areas. If
clinically significant precursor lesions are identified, ablative
(cryotherapy) or excisional procedures (loop electrosurgical exci-

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_28, © Springer Science+Business Media New York 2015

421
422 Benjamin D. Grant et al.

sion procedure (LEEP)) are performed. Although these algorithms

are effective, they are expensive and require high-level infrastruc-
ture and well-trained personnel. In addition, they require three
separate patient visits with communication of test results between
visits. There is therefore a significant need for alternative solutions,
particularly in low-resource settings. One such approach is visual
inspection with acetic acid (VIA), in which acetic acid is applied to
the cervix. If there is whitening of the epithelium in response to
the acetic acid, indicating a precancerous lesion, immediate treat-
ment with cryotherapy or LEEP is performed (See & Treat). The
sensitivity of VIA to detect cervical dysplasia and cancer is similar
to standard colposcopy, but with a markedly lower specificity [4].
This low specificity translates into false positive results, resulting in
the overtreatment of many benign conditions, increasing the cost
of prevention programs and causing unnecessary concern for
patients. Therefore, there is a significant need for diagnostic meth-
ods that add specificity and better identify patients requiring inter-
vention, particularly in low and middle-income countries where
colposcopically directed biopsies and histopathologic review are
often not available.
The high-resolution microendoscope (HRME) has been devel-
oped over the past decade to be used to image epithelial tissue
in vivo [5–9]. This device offers subcellular resolution of 4.4 μm
and can be utilized to examine areas of tissue that are visually suspi-
cious for precancer in order to assess whether the epithelial cells
exhibit morphological alterations. These alterations include
increased nuclear area and increased nuclear-to-cytoplasmic ratio,
hallmarks of precancer. In the context of cervical cancer surveil-
lance, this device allows for in vivo, subcellular evaluation of suspi-
cious areas identified during visual inspection by acetic acid (VIA).
This offers some important potential advantages to improve see-
and-treat programs in the developing world, where pathologists are
scarce [10] and patient follow-up is poor [11]. The HRME pro-
vides real-time visualization of the nuclei, providing the clinician
with valuable information, thus aiding in making a diagnosis. It
thus could provide the opportunity for immediate treatment with
increased specificity and comparable sensitivity to VIA [9, 12].
The HRME system, shown schematically in Fig. 1a, is a fiber-
optic fluorescence microscope. It can be built using commercially
available optical components including lenses, filters, mirrors, and
optomechanical accessories. The entire system, shown in Fig. 1b, is
approximately the length and width of a laptop with a height of 5 in.
With proper choice of illumination source, detector and filters the
system can be modified to work over a broad range of wavelengths.
One common in vivo application relies on the use of proflavine, a
fluorescent topical antiseptic that stains cell nuclei with a peak excita-
tion wavelength of 445 nm and peak emission wavelength of
515 nm. Proflavine is not FDA approved for in vivo use as a topical
contrast agent and a proper research protocol is required for its use.
 High-Resolution Microendoscope 423

Fig. 1 Diagram of the HRME imaging system

Regardless of the chosen fluorescent contrast agent, the system relies

on the use of a coherent multimodal fiber-optic bundle that both
supplies the excitation light to and returns the emission light from
the sample. Different sized fibers can be used; however, inter-fiber
spacing and fiber bundle diameter affect the system’s resolution and
field of view. Depending on the application the field-of-view can be
increased at the expense of system resolution or resolution can be
increased with a corresponding decrease in field-of-view. The mate-
rials and methods provided here are specific to the detection of pro-
flavine, but with the changes detailed in Note 3 the system can be
used for a wide array of fluorescent contrast agents.

2 Materials

2.1 Optical Lenses 1. 10× Plan Achromat Objective: 0.25 NA, 10.6 mm working
and Mirrors distance, infinity corrected (Thorlabs RMS10X).
2. Tube Lens: 1-in. diameter spherical achromatic doublet,
150 mm focal length, 400–700 nm wavelength anti-reflective
coating (Thorlabs AC254-150-A).
3. Condenser Lens: 1-in. diameter aspherical plano-convex con-
denser lens, 17.0 mm focal length, 400–700 nm wavelength
antireflective coating (Newport KPA031-C).
4. Protected Aluminum Mirror: 1-in. diameter, round, protected
aluminum mirror, λ/10 flatness, >90 % reflectivity from 400 to
700 nm (Thorlabs PF10-03-G01).

2.2 Optical Filters 1. Dichroic Mirror: 1-in. diameter, 45° angle of incidence,
485 nm cut-off (Chroma 485dclp).
2. Excitation Filter: 1-in. diameter, thin-film bandpass filter, 452
center wavelength with 45 nm bandwidth (Semrock FF01-
454/45-25).
424 Benjamin D. Grant et al.

3. Emission Filter: 1-in. diameter thin-film bandpass filter, 550

center wavelength with 88 nm bandwidth (Semrock
FF01-550/88-25).

2.3 Optomechanics 1. Cage Cube: 30 mm cage cube compatible with Thorlabs’ cage
system (Thorlabs C6W).
2. Blank Cage Cube Cover Plate: cover plate for 30 mm cage
cube (Thorlabs B1C).
3. Fixed Cage Cube Platform: rotating round platform for plac-
ing optics in 30 mm cage cube system (Thorlabs B3C).
4. Cage Cube Optics Mount: optics mount for 1-in. optics com-
patible with the cage cube platform (Thorlabs B5C).
5. End Cap: 1-in. diameter end cap (Thorlabs SM1CP2).
6. Cage Plate: 30 mm square cage plate for 1-in. optics (Thorlabs
CP02).
7. Cage Rods: 6 mm diameter steel rods for Thorlabs 30 mm
cage system, 1.5″ (4× Thorlabs ER1.5), 3.0″ (4× Thorlabs
ER3), 6.0″ (4× Thorlabs ER6).
8. Right-Angle Kinematic Mirror Holder: holds 1″ diameter mir-
ror to redirect light in 30 mm cage systems (KCB1).
9. Lens Tube, 1″: threaded lens tube for 1-in. optics, 1 in. in
length (Thorlabs SM1L10).
10. Lens Tube, 3″: threaded lens tube for 1-in. optics, 3 in. in
length (Thorlabs SM1L30).
11. Focusing Z-Translator: translation mount for focusing in 1 μm
increments (Thorlabs SM1Z).
12. Camera Adapter: SM1 to C-mount adapter to allow c-mount
capable camera to interface with Thorlabs parts (Thorlabs
SM1A9).
13. RM1 Adapter: SM1 to RMS adapter to connect commercial
objective to system (Thorlabs SM1A3).
14. SM1 Coupler: externally threaded coupler for attaching two
internally threaded 30 mm Thorlabs components (Thorlabs
SM1T2 2×).
15. SMA Receptacle: attaches SMA connector to Thorlabs 30 mm
system (Thorlabs SM1SMA).
16. Retaining Rings: holds optical components in lens tubes
(Thorlabs SM1RR).

2.4 Illumination 1. Royal Blue LED: 900 mW royal blue LED with heat sink,
centered at 455 nm (Thorlabs M455L3).
2. High power LED fDriver: 12.0 V constant current driver with
variable current selection from 0 to 1,200 mA (Thorlabs,
LEDD1B).
 High-Resolution Microendoscope 425

2.5 Additional 1. C-mount CCD Camera: minimum 8-bit monochrome cam-

Imaging Components era, capable of up to 100 ms exposure time, minimum 10
frames per second (e.g., Point Grey Grasshopper
GRAS-14S5M).
2. Laptop computer: Any computer capable of smoothly running
selected camera.
3. Coherent fiber bundle terminated with an SMA connector:
minimum 5 ft length (e.g., Myriad Fiber FIGH-30-850N).

2.6 Proflavine 1. 1× PBS (1 L).

Solution 2. Proflavine hemisulfate salt hydrate (Sigma-Aldrich P2508).

2.7 Proflavine 1. Dissolve appropriate weight of proflavine in 1× PBS solution

Solution (0.01 % w/v) to make desired volume of .01 % w/v proflavine solution using
sterile glassware and instruments.
2. Filter using a 0.22 μm filtration system.

3 Methods

3.1 Optics Assembly 1. Figure 2 illustrates the assembly of the tube lens, excitation
filter and cage cube. Start by placing the excitation filter [4]
into the 3 in. lens tube [5]. The filter should be placed as far
down into the tube as possible, so that it is coincident with the
beginning of the external threads. The filter should be facing
the male end of the tube [5] where it interfaces with the cage
cube [6] because the light will be entering the filter from this
direction. Next, secure the filter with a 1-in. retaining ring [3].
Now the tube lens [2] should be placed in the lens tube [5]
completely, such that it is flush with the retaining ring [3].
Ensure that the more convex side of the lens faces the male end
[5] of the lens tube, as this is where collimated light will enter
the lens. Secure the lens with a second retaining ring. Finally,
screw the lens tube into the cage cube [6].
2. The assembly of the kinematic mirror mount, the camera and
the lens tube is shown in Fig. 3a. Place the aluminum mirror
[1] in the right-angle kinematic mirror mount [2]. Create four
7.5″ cage rods by screwing each 1.5″ rod [3] into a 6″ rod [4].
Next, screw the 7.5″ cage rod into the four holes on one face
of the kinematic mirror mount [2]. Slide the cage cube [5]
onto the cage rods so that the 3″ lens tube is facing the kine-
matic mirror mount. Next, attach the camera [8] to the kine-
matic mirror mount [2] by using the SM1 coupler [6] and the
C-mount to SM1 camera adapter [7]. At this point, turn the
camera on and the focus the system on an object sufficiently far
away to achieve infinity focus (in order to properly interface
with the infinity-corrected objective). To focus the system,
426 Benjamin D. Grant et al.

Fig. 2 Assembly of primary imaging optics: [1] retaining ring, [2] tube lens, [3]
retaining ring [4] emission filter, [5] lens tube, [6] cage cube

move the cage cube along the cage rods until the image appears
focused on the camera image. Using a 150 mm focal length
lens tube with the Grasshopper camera results in a slight air
gap between the lens tube and the kinematic mirror mount
demonstrated in Fig. 3b.
3. Install the cage cube optics mount [1] on the cage cube plat-
form [2] using the provided bolt. Next, secure the dichroic
mirror [3] into the optics mount [1]. Make sure the dichroic
mirror is facing out of the optics mount as shown in Fig. 4a.
4. Install the cage cube platform [2] onto the cage cube [4].
Rotate the cage cube platform so that the dichroic mirror is
facing at a 45° angle between holes five and six as depicted in
Fig. 4b. Secure the cage cube platform at this angle for the
time being with provided screws. See Note 1 for more
information.
5. Figure 4c demonstrates the installation of the cage cube cover
plate and end caps. Place the blank cage cube cover plate [7]
on the top of the cage cube [4] and cover the rear hole with
the end cap [8]. These help reduce stray light.
6. Figure 5 depicts the installation of the commercial objective
and the SMA adapter to the system. The SMA adapter allows
an SMA-terminated fiber to be integrated directly into the sys-
tem. The fiber screws directly into the SMA adapter and the
SMA adapter screws into the Z-translating stage. Install the
RMS to SM1 adapter [1] on the 10× commercial objective [2].
With the adapter in place, install the objective into the cage
cube [5]. Next, place the SM1 threaded SMA connector [4]
into the Z-translating stage [3]. Attach the Z-stage translator
 High-Resolution Microendoscope 427

Fig. 3 Attaching the camera and kinematic mirror mount. (a) Exploded view showing the relative locations of
[1] the aluminum mirror, [2] the kinematic mirror mount, [3] the 1.5″ cage rods, [4] the 6″ cage rods, [5] the
cage cube, [6] the SM1 coupler, [7] the c-mount to SM1 coupler, and [8] the camera. (b) The top view of the
system assembled to this point showing the gap between the kinematic mirror mount and the lens tube

onto the main system by sliding it onto the cage rods. Do not
tighten onto the cage rods until the system is moderately
focused. To focus, install the fiber bundle using the SMA con-
nections. With the camera on, face the distal end of the fiber
directly towards a light. Move the Z-translator manually along
the cage rods until the fiber bundle is relatively focused, then
tighten the Z-translator on the cage rods. Again, while facing
428 Benjamin D. Grant et al.

Fig. 4 Assembly of the dichroic mirror in the cage cube system. (a) Exploded view of [3] the dichroic mirror, [2]
the cage cube platform and [1] the cage cube optic mount. (b) Installation of [2] the cage cube platform onto
[4] the cage cube, illustrating [1] the cage cube optic mount and [3] the dichroic mirror at a 45° angle to open-
ings [5] and [6]. (c) Addition of [7] the blank cover plate and [8] the end cap to block stray light from entering
[4] the cage cube

the fiber towards a light source, adjust the Z-translator using

the fine adjustment knob to achieve optimal focus. When the
fiber bundle is focused correctly individual fiber-optic cores are
readily visible.
7. Connect the LED [1] to the cage plate [3] using the SM1
coupler [2] as shown in Fig. 6a. Figure 6b illustrates the instal-
lation of the excitation filter and collimation optics. Place the
excitation filter [5] all the way into the 1.5 in. lens tube so that
it is coincident with the male end of the lens tube [4]. Place the
filter facing the female end of the lens tube where the LED
light will be entering. Secure it in place with a retaining ring
[6]. Next, place the condenser lens [7] in the lens tube, adja-
cent to the retaining ring [6]. Secure it with a final retaining
ring [8]. Finally, screw the 1.5 in. lens tube [4] into the cage
cube assembly [9]. Screw the four 3-in. cage rods [10] into the
 High-Resolution Microendoscope 429

Fig. 5 An exploded view showing the installation of [2] the objective lens into [5]
the cage cube using an [1] RMS to SM1 adapter and the installation of [3] the
Z-translating stage and [4] the SMA fiber mount

same face of the cage cube assembly [9]. The installation of the
LED assembly to the system is shown in Fig. 6c. Connect the
LED by guiding the LED-cage plate [3] assembly onto the
four 3-in. cage rods [10]. Adjust the position of the LED to
maximize the light intensity exiting the distal end of the fiber
bundle. This can be achieved by aiming the distal end of the
fiber at a power meter while adjusting the LED position.
Secure the LED in this position. Figure 7 shows the entire sys-
tem both with (A) and without (B) major optomechanics.

3.2 Cervical Imaging 1. Patient should be enrolled in an appropriate research protocol

Procedure and his/her informed consent should be documented.
2. Sites to be examined by the HRME should be selected by a
gynecologist using standard visual inspection with acetic acid
and/or Lugol’s Iodine solution, with or without the aid of a
colposcope.
3. The surface of the cervix should be cleaned using a cotton
swab to remove any debris and mucous.
4. 0.01 % proflavine is then applied to the cervix using either a
sterile cotton swab or a spray bottle.
5. Imaging can then be performed by placing the fiber-optic
probe in direct contact with the tissue in the previously selected
areas of interest and slowly scanning lightly across the surface
of the cervix. See Note 2 for information on acquiring optimal
images.
430 Benjamin D. Grant et al.

Fig. 6 Illumination installation. (a) Installation of [1] the LED using an [2] SM1
coupler and [3] the cage plate (b) Exploded view of illumination optics. The [5]
excitation filter is secured into [4] the 1.5″ lens tube using [6] a retaining ring.
Next, [7] a condenser lens is secured using [8] an additional retaining ring
directly behind the filter. [4] The 1.5″ lens tube, now containing the excitation
optics, is screwed into [9] the cage cube system. Finally, [10] the 3-in. cage rods
are installed in each corner of the cage cube (c) The [1–3] LED cage plate
assembly is connected to the system using [10] the cage rods

3.3 Sample Results The scarceness of cytopathologists in developing countries neces-

and Applications sitates an alternative to traditional Pap smear methods [11]. The
in Mobile Health HRME is portable, battery powered and does only require mini-
mal training. Furthermore, image analysis can provide real-time
objective calculations of image parameters that have shown strong
correlation with pathological diagnosis [9, 12]. Suspicious regions
of the cervix can be visually identified by a trained medical provider
with high sensitivity by VIA. Nurses can be trained to be proficient
at VIA in as little as 5 days [13]. VIA followed by immediate
treatment by cryotherapy is the current recommendation by the
WHO in developing countries [13]. However, due to the relatively
 High-Resolution Microendoscope 431

Fig. 7 Overall Diagram of the HRME. (a) Overall diagram of completed system
with major components labeled: [1] CCD camera, [2] kinematic mirror mount, [3]
3" lens tube, [4] cage cube with cover plate, [5] 10× objective lens, [6]
Z-translating stage, [7] SMA connector, [8] LED. (b) The HRME without major
optomechanics, with major components labeled: [1] CCD Camera, [2] mirror, [3]
emission light path, [4] tube lens, [5] emission filter, [6] dichroic mirror, [7] exci-
tation filter, [8] condenser Lens, [9] excitation light path, [10] LED

low specificity of VIA alone, many patients are unnecessarily

treated. The HRME offers the possibility of maintaining the sensi-
tivity and ease of VIA while adding specificity.
Numerous small pilot studies have been conducted to evaluate
the HRME’s ability to detect precancerous cervical lesions in vivo
[9, 12]. One such study took place at the Princess Marina Hospital
in Botswana. Patients undergoing routine colposcopic examination
following an abnormal Pap smear were eligible for this study. After
the application of acetic acid, the HRME was placed on regions
demonstrating acetowhitening as determine by VIA. Additionally,
432 Benjamin D. Grant et al.

the HRME was used to image one control area that appeared nor-
mal by visual inspection. Biopsies were taken from the correspond-
ing areas of the cervix for pathologic diagnosis [12]. Figure 8
provides an example of the HRME’s ability to discriminate between
benign and precancerous acetowhitening. Figure 8a, d are wide-
field images of the cervix after application of acetic acid [12]. The
white arrows in both images indicate suspicious areas based on ace-
towhitening. Figure 8b is an HRME image corresponding to the
suspicious region in Fig. 8a. The HRME image shows normal cel-
lular features: the underlying nuclei are small, evenly spaced and
show low eccentricity. A biopsy of this region was found to be nor-
mal by histopathology and a sample image is provided in
C. Conversely, Fig. 8e, the HRME image from the suspicious
region in D, shows precancerous cellular features. The nuclei are
large, crowded, and irregular, indicative of high grade dysplasia.
Histopathological analysis of the biopsy from this area confirmed
that the underlying disease was high grade cervical interepithelial
neoplasia (CIN 3) [12].

Fig. 8 Comparison of colposcopic images, HRME images, and histologic diagnosis. Both (a) and (d) show col-
poscopic images of cervices with regions undergoing acetowhitening (indicated by white arrows). When the
suspicious lesion in (a) is imaged with the HRME (b), the HRME shows that the nuclei are small, round and
well-spaced. However, when the suspicious lesion (d) is imaged with the HRME (e) the nuclei are ragged,
irregular, and crowded. Histopathologic diagnosis of tissue biopsied from the regions of interest in (a) and (d)
confirm that (a) is a non-neoplastic lesion while that of (d) is high grade dysplasia (CIN3) [12]
 High-Resolution Microendoscope 433

Fig. 9 The average N/C ratio versus histopathologic diagnosis for a pilot study
in Botswana [12]

Due to both the increase in size and density of nuclei in

precancerous lesions, nuclear to cytoplasmic ratio has shown to be
a strong, objective metric to discriminate between normal and
high grade precancerous lesions in small pilot studies [9, 12].
Figure 9 shows the nuclear-to-cytoplasmic ratios versus histopath-
ologic diagnosis for the 44 biopsy sites imaged in the Botswana
study [12]. While larger, multicenter trials must be conducted to
verify these results, preliminary data suggests using the HRME in
conjunction with a VIA-trained nurse could allow treatment of
precancerous lesions in low-level hospital settings with higher
specificity than VIA alone. Due to the portable nature of the
HRME and its ability to run on battery power, it could be used in
settings where traditional methods could not. Additionally, smaller
versions of the HRME that require only a tablet or a cell-phone are
currently under development.

4 Notes

1. Dichroic Mirror Alignment: One of the most difficult steps of

the assembly procedure is properly aligning the dichroic mir-
ror. This alignment can be optimized after the device is com-
pletely assembled by rotating the cage platform and measuring
the power coming out of the fiber. The highest power is
achieved when the dichroic mirror is optimally aligned and it
should be firmly secured at this location.
2. Debris on the Fiber: Another potential pitfall is debris on the
fiber. This will sometimes occur during in vivo imaging and can
be solved by wiping the distal end of the fiber with an alcohol
434 Benjamin D. Grant et al.

swab or a piece of lens paper. Finally, achieving optimal image

quality depends on keeping the fiber in direct contact with the
tissue, which becomes easier with practice.
3. Optimizing for alternative contrast agents: One of the greatest
merits of this system is its versatility. It can be configured to
image other fluorescent markers by selecting the appropriate
LED, filters, and dichroic mirror. Other cameras, fiber bun-
dles, tube lenses and objectives can be substituted for those
listed here. When choosing a camera, choose a lens tube and
objective that provide at least two pixels per individual fiber in
the fiber bundle. This will ensure the resolution is limited only
by fiber-to-fiber spacing. Similar lenses, LEDs, and optome-
chanics can be purchased from other suppliers as well.

References
1. Jemal A, Bray F, Center MM et al (2011) 8. Pierce MC, Vila PM, Polydorides AD et al
Global cancer statistics. CA Cancer J Clin (2011) Low-cost endomicroscopy in the
61:69–90 esophagus and colon. Am J Gastroenterol
2. Kitchener HC, Castle PE, Cox JT (2006) 106(9):1722–1724
Chapter 7: achievenements and limitations of 9. Pierce MC, Guan Y, Quinn MK et al (2012) A
cervical cytology screening. Vaccine 24S3:63–70 pilot study of low-cost, high-resolution
3. Papanicolaou GN, Traut HF (1941) The microendoscopy as a tool for identifying
diagnostic value of vaginal smears in carci- women with cervical precancer. Cancer Prev
noma of the uterus. Am J Obstet Gynecol Res 5(11):1273–1279
42:193–206 10. Hitchcock CL (2011) The future of telepa-
4. Sankaranarayanan R, Esmy PO, Rajkumar R thology for the developing world. Arch Pathol
et al (2007) Effect of visual screening on cervi- Lab Med 135(2):211–214
cal cancer incidence and mortality in Tamil 11. Suba EJ, Murphy SK, Donelly AD et al (2006)
Nadu, India: a luster-randomised trial. Lancet Systems analysis of real-world obstacles to suc-
370(9585):398–406 cessful cervical cancer prevention in developing
5. Muldoon TJ, Pierce MC, Nida DL et al (2007) countries. Am J Public Health 96(3):480–487
Subcellular-resolution molecular imaging 12. Quinn MK, Bubi TC, Pierce MC et al (2012)
within living tissue by fiber microendoscopy. High-resolution microendoscopy for the
Opt Express 15(25):16413–16423 detection of cervical neoplasia in low-resource
6. Shin D, Pierce MC, Gillenwater AM et al settings. PLoS One 7(9):e44924
(2010) A fiber-optic fluorescence microscope 13. World Health Organization (2012) Prevention
using a consumer-grade digital camera for of cervical cancer through screening using
in vivo cellular imaging. PLoS One 5(6): visual inspection with acetic acid (VIA) and
11218 treatment with cryotherapy. A demonstration
7. Muldoon TJ, Anandasabapathy S, Maru D et al project in six African countries: Malawi,
(2008) High-resolution imaging in Barrett’s Madagascar, Nigeria, Uganda, The United
esophagus: a novel, low-cost endoscopic micro- Republic of Tanzania, and Zambia. WHO
scope. Gastrointest Endosc 68(4):737–744 Document Production Services
 Chapter 29

Skin Lesions Image Analysis Utilizing Smartphones

and Cloud Platforms
Charalampos Doukas, Paris Stagkopoulos, and Ilias Maglogiannis

Abstract
This chapter presents the state of the art on mobile teledermoscopy applications, utilizing smartphones
able to store digital images of skin areas depicting regions of interest (lesions) and perform self-assessment
or communicate the captured images with expert physicians. Mobile teledermoscopy systems consist of a
mobile application that can acquire and identify moles in skin images and classify them according their
severity and Cloud infrastructure exploiting computational and storage resources. The chapter presents
some indicative mobile applications for skin lesions assessment and describes a proposed system developed
by our team that can perform skin lesion evaluation both on the phone and on the Cloud, depending on
the network availability.

Key words Image analysis, Skin lesions, Skin cancer, Melanoma, Mobile dermoscopy, Mobile com-
puting, Teledermatology, Cloud infrastructures, Android, iOS, Smartphones

1  Introduction

Skin cancer is among the most frequent types of cancer and one of
the most malignant tumors. Its incidence has increased faster than
that of almost all other cancers and the annual rates have increased on
the order of 3–7 % in fair-skinned populations in recent decades [1].
Currently, between 2 and 3 million non-melanoma skin cancers and
132,000 melanoma skin cancers occur globally each year. Skin cancer
is the most common form of cancer in the USA. More than 3.5
million skin cancers in over two million people are diagnosed annu-
ally One in every three cancers diagnosed is a skin cancer and, accord-
ing to Skin Cancer Foundation Statistics, one in every five Americans
will develop skin cancer in their lifetime [2]. Each year there are
more new cases of skin cancer than the combined incidence of
cancers of the breast, prostate, lung, and colon [3]. Treatment of
non-melanoma skin cancers increased by nearly 77 % between 1992
and 2006 [4]. Over the past three decades, more people have had
skin cancer than all other cancers combined [3].

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_29, © Springer Science+Business Media New York 2015

435
436 Charalampos Doukas et al.

The cutaneous melanoma, which is the most common type of

skin cancer, is still incurable through conventional cancer treat-
ments. However, when it is diagnosed at early stages it can be
treated and cured (i.e., removed) without complications. The sta-
tistics reveal that from 1970 to 2009, the incidence of melanoma
increased by 800 % among young women and 400 % among young
men [5]. One person dies of melanoma every hour (every 57 min),
while an estimated 76,690 new cases of invasive melanoma will be
diagnosed in the USA in 2013 [6].
All the above facts and figures prove the importance of the early
diagnosis in skin cancer related diseases and especially in the mela-
noma cases. However the differentiation of early melanoma from
other pigmented skin lesions (e.g., benign neoplasms that simulate
melanoma) is not trivial even for experienced dermatologists; in
several cases primary care physicians seems to underestimate mela-
noma in its early stage [7] (see Note 1). The latter has attracted the
interest of many researchers, who have developed systems for auto-
mated detection of malignancies in skin lesions. Such systems
require the acquisition of the skin lesion image using techniques
like epiluminescence microscopy (ELM or dermoscopy), transmis-
sion electron microscopy (TEM), and image acquisition using still
or video cameras [8]. The latter systems consist of expensive hard-
ware equipment that is installed in dermatological assessment/
treatment centers and require an on-site visit from the patient.
An easier and less demanding method for skin cancer screening
and early detection could be based on the usage of modern smart-
phones, which are low cost widely used devices capable of skin
lesion image capture and communication. These so called mobile
teledermoscopy systems can be utilized as an efficient, low cost and
early warning alternative method of skin cancer diagnosis [11].
Although the mobile teledermoscopy systems are less accurate
compare to the expensive ELM or TEM devices, patients could use
them for an early characterization of lesions and estimation for fur-
ther assessment still can use them.
This chapter presents an overview of the state of the art in such
teledermoscopy systems working into any mobile environment
(Android, iOS, etc.), along with the enabling technologies. In
most of the systems a Cloud infrastructure provides the essential
data storage and processing components for pattern recognition
and effective skin cancer detection.
The skin consists of a number of layers with distinct functional
and optical properties. White light shone onto the skin penetrates
the skin layers and whilst some of it is absorbed, much is reflected
back and can be acquired by an optical sensor. The epidermis is a
large layer, which contains the melanin producing cells, the mela-
nocytes, and their product, melanin. Melanin is a pigment that
strongly absorbs light in the blue part of the visible and in the
ultraviolet (UV) spectrum, acting as a protecting filter from harm-
ful effects of UV radiation (see Fig. 1).
Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 437

Fig. 1 Normal skin lesions and main components (source: MediceNet)

Fig. 2 Illustration of (a) typical melanoma, (b) dysplastic nevus, and (c) non-­
dysplastic (common) nevus

Pigmented skin lesions appear as patches of darker color on

the skin. In most cases the cause is excessive melanin concentra-
tion in the skin. In benign lesions (e.g., common nevi) melanin
deposits are normally found in the epidermis (see Fig. 2). In malig-
nant lesions (i.e., melanoma), the melanocytes reproduce melanin
at a high, abnormal rate (see Fig. 3).
Dysplastic nevi are skin lesions that have high risk of becoming
melanomas since the temporal deformation of melanin is a major
indication of melanoma [1].
In the conventional procedure, the following diagnosis
methods are mainly used [12]: (1) ABCD rule of dermoscopy
(2) Pattern Analysis; (3) Menzies method; (4) 7-Point Checklist;
and (5) Texture Analysis.
438 Charalampos Doukas et al.

Fig. 3 Illustration of Melanocytes and Melanoma on skin (source: MediceNet)

The ABCD rule investigates the asymmetry (A), border (B),

color (C), and differential structures (D) of the lesion and defines
the basis for a diagnosis by a dermatologist. More specifically:
(A) Asymmetry: The lesion is bisected by two axes that are posi-
tioned to produce the lowest asymmetry possible, in terms of bor-
ders, colors, and dermoscopic structures. The asymmetry is
examined with respect to a point, one or more axes. The asymme-
try index is computed by first finding the principal axes of inertia of
the tumor shape in the image and it is obtained by overlapping the
two halves of the tumor along the principal axes of inertia and
dividing the non-overlapping area differences of the two halves by
the total area of the tumor.
(B) Border: Border based features describing the shape of the lesion
are then computed. In order to extract border information, image
segmentation is performed. It is considered to be a very critical
step in the whole process of skin lesion identification and involves
the extraction of the region of interest (ROI), which is the lesion
and its separation from the healthy skin. Most usual methods are
based on thresholding, region growing and color transformation
(e.g., principal components transform, CIELAB color space, and
spherical coordinates) and declarative knowledge (melanocytic
lesion images segmentation enforcing by spatial relations based
declarative knowledge) are used for determining skin lesion fea-
tures. The latter methods are characterized as region approaches,
because they are based on different colorization among the malig-
nant regions and the main border. Another category of segmenta-
tion techniques are contour approaches using classical edge
detectors (e.g., Sobel, Canny) that produce a collection of edges
leaving the selection of the boundary up to the human observer.
The most popular border features are the Greatest Diameter,
the Area, the Border Irregularity, the Thinness Ratio, the Circularity
index, the variance of the distance of the border lesion points from
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 439

the centroid location, and the Symmetry Distance. The Circularity

index (CIRC) is mathematically defined by the following equation:
4 Aπ
CIRC = (1)
p2

where A is the surface of the examined area and p its perimeter.

Symmetry Distance (SD) calculates the average displacement
among a number of vertexes as the original shape is transformed in
to a symmetric shape. The symmetric shape closest to the original
shape P is called the symmetry transform (ST) of P. The SD of an
object is determined by the amount of effort required to transform
the original shape into a symmetrical shape, and can be calculated
as follows:
1 n −1 
SD = ∑
n i =0
Pi − Pi (2)

Apart from regarding the border as a contour, emphasis is also
placed on the features that quantify the transition (swiftness) from
the lesion to the skin. Such features are the minimum, maximum,
average, and variance responses of the gradient operator, applied
on the intensity image along the lesion border.
(C) Color: Color properties inside the lesion are examined and the
number of colors present is determined. They may include: Light
Brown, Dark Brown, Black, Red (red vascular areas are scored),
White (if whiter than the surrounding skin), Slate-blue. In addi-
tion, color texture might be used for determining the nature of
melanocytic skin lesions. Typical color images consist of the three-
color channels RGB (red, green, and blue). The color features are
based on measurements on these color channels or other color
channels such as CMY (Cyan, Magenta, Yellow), HSV (Hue,
Saturation, Value), YUV (Y-luminance, U-V chrominance compo-
nents), or various combinations of them, linear or not. Additional
color features are the Spherical coordinates LAB average and vari-
ance responses for pixels within the lesion are calculated as

L = R2 + G 2 + B2 (3)
B 
Angle A = cos −1   (4)
L 
 R 
Angle B = cos −1   (5)
 L sin ( Angle A ) 

Color variegation may be calculated by measuring minimum, max-

imum, average, and standard deviations of the selected channel
values, color intensity and by measuring chromatic differences
inside the lesion.
440 Charalampos Doukas et al.

Fig. 4 Asymmetry border color features; (a) Asymmetry Test, (b) Border Test, (c) Color variegation (source:
http://www.dermoncology.com)

(D) Differential structures: The number of structural components

present is determined (see Fig. 4), i.e., Pigment Network, Dots
(scored if three (3) or more are present), Globules (scored if two
(2) or more are present).
The Pattern Analysis method seeks to identify specific patterns,
which may be global (Reticular, Globular, Cobblestone,
Homogeneous, Starburst, Parallel, Multicomponent, Nonspecific)
or local (Pigment network, Dots/globules/moles, Streaks, Blue-­
whitish veil, Regression structures, Hypopigmentation, Blotches,
Vascular structures).
The Menzies method looks for negative features (Symmetry of
pattern, Presence of a single color) and positive (Blue-white veil,
Multiple brown dots, Pseudopods, Radial streaming, Scar-like
depigmentation, Peripheral black dots/globules, Multiple (5–6)
colors, Multiple blue/gray dots, Broadened network).
The 7-point checklist [13, 14] refers to seven criteria that
assess both the chromatic characteristics and to the shape and/or
texture of the lesion. These criteria are Atypical pigment network,
Blue-whitish veil, Atypical vascular pattern, Irregular streaks,
Irregular dots/globules, Irregular blotches, and Regression struc-
tures. Each one is considered to affect the final assessment with a
different weight. The dermoscopic image of a melanocytic skin
lesion is analyzed in order to evidence the presence of these stan-
dard criteria; finally, a score is calculated from this analysis, and, if
a total score of three (3) or more is given, the lesion is classified as
malignant, otherwise classified as nevus.
Finally Texture Analysis is the attempt to quantify texture
notions such as “fine,” “rough,” and “irregular” and to identify,
measure, and utilize the differences between them. Textural features
and texture analysis methods can be loosely divided into two catego-
ries: statistical and structural. Statistical methods define texture in
terms of local gray-level statistics that are constant or slowly varying
over a textured region. Different textures can be discriminated by
comparing the statistics computed over different sub-regions.
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 441

Some of the most common textural features are as follows:

Neighboring gray-level dependence matrix (NGLDM) and lattice
aperture waveform set (LAWS) are two textural approaches used for
analyzing and detection the pigmented network on skin lesions.
Dissimilarity, d, is a measure related to contrast using linear increase
of weights as one moves away from the Gray Level Co-occurrence
Matrix (GLCM) diagonal. Dissimilarity is calculated as follows:
N −1
d= ∑P i, j i−j (6)
i , j =0

where i is the row number, j is the column number, N is the total
number of rows and columns of the GLCM matrix, and
Vi , j
Pi , j = N −1
(7)
∑V
i , j =0
i, j

is the normalization equation in which Vi,j is the Digital Number
(DN) value of the cell i,j in the image window (i.e., the current
grayscale pixel value).
Angular Second Moment, (ASM), which is a measure related to
orderliness, where Pi,j is used as a weight to itself :
N −1
ASM = ∑ iP 2
i, j (8)
i , j =0

GLCM Mean, μi, which differs from the familiar mean equa-
tion in the sense that it denotes the frequency of the occurrence of
one pixel value in combination with a certain neighbor pixel value
and is given by
N −1
µi = ∑i (P ) i, j (9)
i , j =0

GLCM Standard Deviation, σi, which gives a measure of the
dispersion of the values around the mean
N −1

∑ P (i − µ )
2
σi = i, j i (10)
i , j =0

In ref. [8] we have presented a detailed comparison between
the most common methods and research works for analyzing skin
lesion images. Annotation of skin images is also particularly
important for medical decision support systems since medical
­
experts have difficulties in understanding the criteria of decision in
many existing systems. A semantic taxonomy and hierarchy of skin
lesion images using ontological approached is also presented in
ref. [21] (Fig. 5).
442 Charalampos Doukas et al.

Fig. 5 Differential structures; (a) Pigmented network, (b) Dots, (c) Brown globules, (d) Branched streaks
(source: http://www.dermoncology.com)

According to a survey we performed there are several commercial

(or free) mobile applications for skin lesion assessment available for
Android and iOS devices. In this section we present the basic char-
acteristics and features of the most significant of them.

1.1  Doctor Mole This application uses augmented reality technology to scan the
Application skin lesions in real time and extracts instant risk feedback on
Asymmetry, Border, Color, Diameter, and Risk. The photos which
are taken by user are stored in the mobile phone for future refer-
ence and examination of the lesion progress through time.
In addition, it is designed to help in the assessment of moles.
This allows the user to easily take a photo of the desired target
mole accurately with the correct lighting and distance. All the cal-
culations are performed on the phone so as results are provided
instantly. At any point user has the ability to review all photos,
results and make comparisons between mole images captured at
different times.
Below, we present a few screenshots of this application (Fig. 6).

1.2  Mole Detective Mole Detective takes pictures of moles on user’s skin and analyzes
the symptoms of melanoma in order to increase the chance of
detecting skin cancer in early stages. The survival rate of melanoma
is a dismal 15 % at stage four. However, when caught early, the sur-
vival rate is 95 %. Users have the ability to indicate the location of
the mole their body and take pictures of it. Then, users can analyze
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 443

Fig. 6 (a) Compare saved photos. (b) Results of mole. (c) Preview of the mole. (d) Display risk of mole (source:
https://play.google.com/store/apps/details?id=com.revsoft.doctormole)
444 Charalampos Doukas et al.

the specific dermatological mole using the dermatologist ABCD

method that provides a risk factor based on the symptoms of the
mole that may or may not be shown (see Fig. 7).

1.3  MelApp Mel-App is an image-based risk assessment mobile application that

assists in the early detection of melanoma. Generally, the process-
ing has two steps:
●● Taking a picture of the skin lesions of concern with the phone’s
camera, centering the mole using the annotation tool and
enlarging it with the zoom feature and/or resizing the green
circle.
●● Pinpointing the mole size and its evolution by sliding the cor-
responding indicator bar and tap on “Check Risk.” Within sec-
onds Mel-App provides a risk analysis of the uploaded picture
for being a melanoma or a benign region.
It uses highly sophisticated patent protected state-of-the-art
mathematical algorithms and image-based pattern recognition
technology to analyze the uploaded image. These pictures also can
be stored on Mel-App and saved according to date, label or risk.
A basic way to assess lesions or moles is by using Mel-App’s ABCDE
feature to manually adjust the Asymmetry, Border Irregularity,
Color Density, Diameter, and Evolution of your image prior to
tapping on “Check Risk” (Fig. 8).

1.4  OnlineDermClinic The OnlineDermClinic application uploads all the dermatological

photos of skin problem with additional features of each photo
which are completed from user.
This application consists of three different steps in which the
user adds important information. As a result, the OnlineDermClinic
creates a medical event from this information in databases which is
stored in the OnlineDermClinic.com site.
Specifically, these steps include the choice of doctor based by
area and biography, upload of dermatological image, and the
return and display of results.
If user has a rash, skin growth, changing mole, itchy skin prob-
lem, or any condition of the skin, hair, or nails, then get rapid medi-
cal attention using the photo consult tools by OnlineDermClinic.
com. The OnlineDermClinic App is the real tool and extension of
the site. Having a virtual appointment allows patients much-needed
access to US board certified dermatologists for diagnosis and man-
agement of skin disease.
An important remarkable innovation is the DermaLearn diag-
nosis wizard, which is an algorithmic-based set of questions to
help patients and mid-level health care providers arrive at a priori-
tized list of potential diagnoses, or what physicians call a differen-
tial diagnosis. OnlineDermClinic.com also provides an extensive
written and video library of diseases and treatment options for
education and research (Fig. 9).
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 445

Fig. 7 (a) Start the app. (b) Location’s mole. (c) Result of scanned mole. (d) Take a photo of mole (source:
https://play.google.com/store/apps/details?id=com.moledetective)
446 Charalampos Doukas et al.

Fig. 8 (a) History. (b) Results of image (source: https://play.google.com/store/apps/details?id=com.melapp)

1.5  SpotMole SpotMole provides a simple way to have a quick check of skin spots
and moles. It takes as input an image of a mole which was loaded
by phone’s gallery or was taken from camera. It may detect signs of
melanoma using image processing and pattern recognition tech-
niques. Moreover, it has embedded an automatic mole analysis
using the device’s camera/gallery and algorithms.
Firstly every user should have to take a close-up and well cen-
tered snapshot of a skin spot or mole and next run an analysis.
SpotMole automatic assessment software uses standard visual anal-
ysis procedures to check spots on the skin. The features extracted
are common in dermatology and widely employed in visual inspec-
tion of the skin spots. These are Asymmetry, Border, Color,
Diameter, and Evolution of the mole (ABCDE). Furthermore, this
application provides extra settings, which each user has the chance
to change the parameters of image processing and list with the
doctors who are closer to your location (Fig. 10).

1.6  SkinLesion The application is designed to take as input a dermatological image

Detector from SD card of mobile phone or from camera or can download it
from the personal account of various Cloud Services. The image
exists as a number of series of processes such as filtering, normal-
ization, categorization, etc. The results of this processing are the
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 447

Fig. 9 (a) Select the doctor. (b) Take a photo. (c) List with any mole. (d) Describe the problem (source: https://
play.google.com/store/apps/details?id=pkg.onlinedermclinic)
448 Charalampos Doukas et al.

Fig. 10 (a) Details of melanoma. (b) Basic menu. (c) In progress of analysis. (d) Result image (source: https://
play.google.com/store/apps/details?id=com.spotmole)
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 449

return of the original image defined by the area of melanoma, the

features, accuracy rate of focused area. When users select an image,
they have the choice to change the parameters of processing, define
the location of skin problem on the human body and complete a
specific questionnaire which includes information about focused
area. At the end, if application’s users have access to the Internet
via Wi-Fi or the mobile network of the mobile can send the image to
the server that is configured on the Cloud platform of Okeanos [20].
On the contrary, if they do not have Internet access the image
processing can be done very well on the cell processor. In both
cases, the features of image are stored locally to the database of
mobile. As the result, the user can view or edit all the results by
history of app.
In conclusion, all applications that we have seen so far offer the
opportunity to each user can get a dermatological image from his
mobile (camera, storage, etc.) to process and the results to be con-
sidered as medical data. Each application uses a range of technolo-
gies as tools to achieve its purpose (Cloud Computing, Web
service, Android Architect). The user, depending on which envi-
ronment used, has the ability to be able to process the image either
on the web or on mobile processor. Processed in locally to the
mobile phone makes the running of algorithm slower than to
upload a Web service.
The following table summarizes the way each application pro-
cesses the image as well as indicated whether manual assessment by
the user is needed before analyzing the images (Table 1).

2  Material

In this Section we present a system that can perform skin lesion

evaluation and feature classification both on the phone and on the
Cloud, depending on the network availability. The mobile appli-
cation can be implemented on both Android and iOS devices.

Table 1
Overview of processing method (local or remote) of related applications

Applications/features Upload image Local processing Need for manual assessment

DoctorMole NO YES NO
Mole Detective NO YES NO
MelApp NO YES NO
OnlineDermClinic YES NO NO
SpotMole NO YES NO
SkinLesionDetector YES YES NO
450 Charalampos Doukas et al.

Fig. 11 An overview of the system’s architecture

Local classification should be performed when Internet connection

is not available. However, the Cloud platform provides better
results in terms of assessment speed and allows experts to continu-
ously build a better training model. In addition it offers online
storage capabilities. Figure 11 presents an overview of the sys-
tem’s architecture. The system is divided into two parts: the
mobile part that acquires the skin lesion image from the user and
performs the basic image processing steps and the Cloud part that
contains the classification model and performs the classification
task for a new image.

2.1  The Mobile The mobile application provides all the essential functionality for
Application acquiring the image (either from a storage media device like an
SD card or through the mobile’s camera) (see Note 2). It is also
responsible for segmenting the pigmented skin lesion and extract-
ing the essential features. It also allows users to add contextual
information (like age, inheritance factor in melanoma, exposure
to UV radiation) and information about the assessed skin lesion
(like multiplicity factor, age estimation of the lesion, etc.).
The image segmentation and feature extraction is performed
on the mobile device using a method presented in ref. [9]. The
method does not require the usage of specific image processing
libraries and thus can be implemented on every mobile platform.
Initially the skin lesion region is segmented and then texture (like
ASM and GMSM), size (like area, perimeter), asymmetry index,
and color features are generated. More information on the features
utilized can also be found in ref. [4]. The features along with the
contextual information are then encrypted using AES symmetrical
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 451

encryption with a 128-bit length key stored within the mobile

device. Once encrypted, the data can be uploaded to the Cloud
service for the characterization process. The communication with
the service is performed through a HTTP POST request using the
REST Web Service API provided by the service. REST Web
Services is a very lightweight communication protocol suitable
for Cloud application interacting with mobile devices [15]. The
response of the REST API call is the skin type estimation (mela-
noma, dysplastic nevus, or benign nevus).
For evaluation purposes we have also ported the classification
module (based on the WEKA Engine as discussed in the following
section) to the mobile application allowing the whole image pro-
cessing, feature extraction, and characterization to be performed
on the mobile device.

2.2  The Cloud Part The Cloud part consists of a Java EE (Java Enterprise Edition for
Web applications) application that provides both the management
graphical interface and the interfaces for the communication with
the sensors. As a Cloud infrastructure the Jelastic platform has
been selected. The Jelastic [12] is a Platform as a Service (PaaS)
type Cloud provider that allows users to deploy Java-based applica-
tions providing all the essential components (application server
instances, databases, load balancers, etc.) and all the appropriate
scalability. Jelastic provides full access to the application server run-
time environment, which enables the deployment of additional
Java extensions like encryption and authentication libraries.
For the specific application the Tomcat application server along
with a MySQL database has been utilized. Data decryption has
been achieved using the java cryptographic extension implement-
ing a symmetric (AES) mechanism using the same encryption key
with the mobile application. Communication with the mobile
application is performed through a REST Web Service API. Once
the data has been received from the mobile device and decrypted,
the contextual data is stored into the database for future usage.
The features are then processed by the data classification module
for characterizing the skin lesion. The module utilizes the WEKA
classification engine [14]. The classification engine uses train
models that have been previously created using the WEKA tool.
Several models can be used to achieve the best accuracy; a train
model can initially validate a new feature set for discriminating the
­corresponding image between three classes: melanoma, dysplastic
nevus, and benign nevus. In case a melanoma or a benign nevus is
estimated, a second model can be utilized that characterizes features
to melanoma or nevus providing this way best accuracy to the user.
The train models can be updated with more data or advanced clas-
sification techniques and parameters at any time without affecting
the use of the mobile application and without the need for an
update. The Cloud application has been designed to enable also
452 Charalampos Doukas et al.

the management and storage of skin lesion images on the Cloud.

Users are provided with the option to (anonymously) upload images
for further manual assessment by experts and enhancement of the
training models or as an image repository for their own usage.
Storage resources and data maintenance (e.g., backups, recovery in
case of system failure and redundancy) is provided automatically by
the Jelastic Cloud platform [16].
The next section presents and discusses an initial evaluation of
the system based on Android mobile device.
Initial Evaluation

3  Methods

3.1  Description For the initial evaluation of the proposed system’s mobile part, an
of the System Android application has been developed and deployed on an
Implementation Android Samsung Galaxy S Plus phone (Android OS v2.3,
1.4 GHz CPU, 512 MB RAM). User initially enters the contextual
information like age, UV exposure, estimated skin lesion age, etc.
(see Fig. 12). Then the skin lesion image is provided, either
through browsing the image file on a local storage media (e.g., an
SD memory card) or by capturing it using the mobile’s camera.
The lesion is segmented, features are extracted and the segmented
image presented to the user (see Fig. 12).
Taking into account the processing capabilities of modern
smartphones a fast algorithm based on a local thresholding tech-
nique was adopted for image segmentation. The window size, the
threshold value and degree of overlap between successive moving
windows were the procedure parameters. These parameters were
tuned so that skin lesions separation was performed satisfactory.
Image thresholding is performed using only the intensity value;
therefore the image is firstly converted into gray scale. Furthermore,
image pixels are smoothed using a standard Gaussian filter, whose
moving window size value is appropriately tuned, for reducing
noise. According to the proposed method, image intensity is
directly compared to the average value of this specific feature com-
puted for all pixels that reside within a wide rectangular area (win-
dow) around this pixel. If the pixel feature value is less that the
average window value minus a characteristic threshold value the
pixel is assumed to be part of the skin lesions region (the pixel is
interpreted as “dark” pixel). If the window is wide enough to con-
tain the entire image for all pixels within it, the technique is called
global thresholding. Typically, to avoid individual image intensity
differences, the feature values are normalized by the average fea-
ture value for all pixels in the image. Thus threshold is provided as
a per-cent fraction of this value. Local (or global) thresholding
partitions the image into objects that can be interpreted as “islands”
of “dark” pixels within the frame. These objects contain image
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 453

Fig. 12 Screenshots of the Android mobile application. On the left: the initial screen for entering contextual
data. On the middle: A skin lesion image segmented by the mobile application. On the right: history data of
two images

pixels extracted by the thresholding process (“dark” pixels) and are

characterized by a certain proximity property in the sense that each
pixel in the object has or has not at least one “dark” pixel in the
neighborhood, that is to say it is close to the former by a distance
of one pixel in any direction. A recursive algorithm was imple-
mented to extract each “island” object. According to this algo-
rithm, the image pixels are scanned one by one. If an un-scanned
“dark” pixel is met, the pixel is added to a new “island” object and
recursively an adjacent “dark” pixel is sought within its neighbor
(adjacent pixels in all directions). In this way all coherent “dark”
pixels of the object are located and characterized as scanned. The
algorithm complexity is polynomial and specifically linear in terms
of image dimensions (number of pixels). In this way all “islands”
are located. The image features can then be evaluated for all “dark”
pixels of the image or for each “island” object separately. The two
step algorithm for skin lesion segmentation is also illustrates as a
flowchart in Fig. 13.
The image features utilized in the proposed mobile applica-
tion are the ABCD rule based, which are discussed in Subheading 2
Materials. The extracted features are encrypted among with the
contextual information and transmitted to the Cloud service.
The response contains the lesion estimation that is presented to
the user. On the Cloud part, a Web Service application developed
as described in Subheading 3 has been deployed on the Jelastic
Service [16].
454 Charalampos Doukas et al.

Fig. 13 Adaptive thresholding and object extraction algorithms’ flowcharts

3.2  Initial Evaluation For the initial evaluation of the system a dataset consisting of over
Results 3,000 skin lesion image sets of manually classified images has been
utilized. The dataset contained about 800 images with melanoma,
600 with dysplastic nevus and the rest 1,600 images with benign nevi.
A subset of them, e.g., 80 % of the images is used as learning set and
the other 20 % of the samples are used for testing using the trained
classifier. The images in both learning and test subsets are exchanged
for all possible combinations to avoid bias in the solution.
We have evaluated a number of different classification algo-
rithms provided by the WEKA tool. Accuracy (%) in correctly clas-
sifying instances, Root Mean Square Error (RMS), True Positive
Rates (TPR), False Positive Rates (FPR), and Area under ROC
Curve (AUC) have been also utilized as evaluation metrics. The
latter metrics are considered established evaluation techniques in
machine learning and classification problems [17–19]. The first
experiment involves the evaluation of the algorithms in character-
izing melanoma versus dysplastic and non-dysplastic (i.e., benign)
skin lesions. The corresponding results are presented in Table 2.
The TPR and FPR refer to the detection of melanocytic class.
Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 455

Table 2
Classification results for the characterization of melanocytic, dysplastic,
and Non-dysplastic skin lesions

A (%) RMS TPR FPR AUC

Bayes Networks 68.70 0.4044 0.942 0.011 0.997
NBL 70.58 0.4077 1.0 0.013 0.999
MLR 75.04 0.338 0.986 0.0 1.0
SVM 77.06 0.3911 1.0 0.0 1.0
MultiLayer Perceptron 75.15 0.3536 0.957 0.0 1.0
RBF Network 72.56 0.3561 1.0 0.004 1.0
KStar 67.24 0.4463 0.246 0.0 0.969
LWL 73.20 0.3592 1.0 0.0 1.0
Classification via 74.44 0.3406 0.942 0.0 1.0
Regression
NBTree 71.98 0.373 0.681 0.02 0.986
CART 73.50 0.3548 1.0 0.0 1.0
Symbol explanation: A accuracy, RMS root mean square error, TPR true positive rates,
FPR false positive rates, AUC area under curve

As indicated by the conducted experiment SVM performs bet-

ter in characterizing skin lesions between melanocytic, dysplastic,
and non-dysplastic with an average accuracy of 77.06 %.
The second experiment involves the discrimination between mel-
anoma and benign regions. In this case several classifiers (e.g., SVM,
CART) have managed to identify properly melanocytic skin regions
against benign ones with 85–90 % accuracy. The results are quite
promising and demonstrate the feasibility of utilizing the proposed
system as a remote and early diagnosis method for skin cancer.
After having evaluated the aforementioned classifiers and selected
the most effective one (e.g., SVM) we have proceeded in evaluating
the overall system performance. The training model has been
deployed into the Cloud application and various skin lesion images
from the untrained dataset have been characterized using the mobile
application. The following table presents results ­regarding the time
(in seconds) to perform the analysis using commercial Wi-Fi and 3G
networks and the time to perform the image processing and classifi-
cation on the device. TR corresponds to the total time it takes to
process the image and upload the essential data to the Cloud service
and retrieve the classification result. TPL represents the time needed
for performing image segmentation and feature extraction on the
phone. TCL corresponds to the classification time on the phone and
TL to the processing time on the device (TL = TCL + TPL) when both
feature extraction and classification are performed locally. (Table 3)
456 Charalampos Doukas et al.

Table 3
Performance results (in seconds) for the mobile application

Image resolution/network type TR TPL TCL TL

400 × 300 (Wi-Fi) 3.57 3.1 1.328 4.43
400 × 300 (3G) 4.08 3.1 1.328 4.43
521 × 437 (Wi-Fi) 4.13 3.6 1.328 4.93
521 × 437 (3G) 4.21 3.6 1.328 4.93
640 × 960 (Wi-Fi) 5.35 4.2 1.328 5.52
640 × 960 (3G) 5.45 4.2 1.328 5.52

The total response time is not affected considerably by the net-

work type. As expected feature extraction time is affected by the
resolution of the image and is slightly slower when performed on
the device than when results are retrieved from the Cloud. In gen-
eral the performance of the system is considered acceptable for a
mobile application.
The system has also been evaluated by a small number of users
[10] in terms of usability and effectiveness. A Mean Opinion Score
(MOS) has been calculated from their responses. The usability of
the mobile application (in terms of user interface convenience and
application performance) has scored 75 %, whereas the effective-
ness has scored 80 %. Users have identified two main issues of the
mobile application that need improvement: the contextual data
entry and the difficulty in some cases to acquire skin lesion images
with proper assessment quality using the mobile camera. Both
issues can be resolved by using different mobile equipment (e.g.,
tablet PC) or better camera device resolution.

4  Notes

1. The differentiation of early melanoma from other benign skin

lesions is not a trivial task even for experienced dermatologists
[3]. On the other hand the early diagnosis of skin cancer is of
severe importance for the outcome of the therapeutic proce-
dure and the basis for reducing mortality rates. This paper has
presented our work on a mobile system that can be easily used
to perform an initial estimation of a skin lesion’s severity. The
system classifies skin images acquired from a mobile phone
into melanoma, dysplastic nevus and common (benign) nevus.
The scope is to identify potential early cases of melanoma and
urge the users to visit an experienced physician whenever a
possible dangerous lesion is detected.
 Evaluating Skin Lesions Utilizing Smart Phones and Cloud Platforms 457

Fig. 14 Artificial Illumination and Lenses System for smartphones

2. In the proposed system physical illumination form the envi-

ronment was used. However we intend in the future to develop
an additional module that will include artificial illumination
sources on the phone and advanced lenses for better focus.
The introduction of such a module, which is illustrated in
Fig.  14 is expected to improve considerably the quality and
reproducibility of the captured images.
The main advantages of the proposed system are the utilization
of a Cloud infrastructure for online storage, continuously improv-
ing the classification model and providing accurate characteriza-
tion results in various mobile platforms. In addition it is the first
system to collect contextual information from the user that can be
later used for a better assessment from an expert and the progress
assessment as well.

References

1. Marks R (2000) Epidemiology of melanoma. incidence of melanoma among young adults:

Clin Exp Dermatol 25:459–463 an epidemiological study in Olmsted County,
2. Ultraviolet radiation and the INTERSUN Minnesota. Mayo Clin Proc 87(4):328–334
Programme, source: World Health Organization. 6. American Cancer Society. Cancer facts & fig-
http://www.who.int/uv/faq/skincancer/en/. ures 2013. http://www.cancer.org/acs/groups/
Accessed 31 Aug 2013 content/@epidemiologysur veilance/docu-
3. Stern RS (2010) Prevalence of a history of skin ments/document/acspc-036845.pdf. Accessed
cancer in 2007: results of an incidence-based 31 Aug 2013
model. Arch Dermatol 146(3):279–282 7. Pariser RJ, Pariser DM (1987) Primary care
4. Rogers HW, Weinstock MA, Harris AR et al physicians errors in handling cutaneous disor-
(2010) Incidence estimate of nonmelanoma ders. J Am Acad Dermatol 17:239–245
skin cancer in the United States, 2006. Arch 8. Maglogiannis I, Doukas C (2009) Overview of
Dermatol 146(3):283–287 advanced computer vision systems for skin
5. Reed KB, Brewer JD, Lohse CM, Bringe KE, lesions characterization. IEEE Trans Inf
Pruit CN, Gibson LE (2012) Increasing Technol Biomed 13(5):721–733
458 Charalampos Doukas et al.

9. Argenziano G et al (2003) Dermoscopy of melanoma diagnosis by one click? Semin Cutan
pigmented skin lesions: results of a consensus Med Surg 28(3):203–205, ISSN 1085-5629
meeting via the Internet. J Am Acad Dermatol 15. AlShahwan F, Moessner K, Carrez F. (2011)
48(5):680–693 ‘Distributing resource intensive mobile web
10. Argenziano G, Fabbrocini G, Carli P, De services’. 2011 International Conference on
Giorgi V, Sammarco E, Delfino M (1998) Innovations in Information Technology, IIT
Epilu­minescence microscopy for the diagnosis 2011, pp. 41–46
of doubtful melanocytic skin lesions. 16. AlShahwan F, Moessner K, Carrez F (2011)
Comparison of the ABCD rule of dermatos- Distributing resource intensive mobile web
copy and a new 7-point checklist based on services. 2011 International conference on
pattern analysis. Arch Dermatol 134(12): innovations in information technology, IIT
1563–1570 2011, 25–27 Apr, Abu Dhabi, pp 41–46
11. Betta G, Di Leo G, Fabbrocini G, Paolillo A, 17. Hall M, Frank E, Holmes G, Pfahringer B,
Scalvenzi M (2005) Automated application of Reutemann R, Witten IH (2009) The WEKA
the “7-point checklist” diagnosis method for data mining software: an update. SIGKDD
skin lesions: estimation of chromatic and shape Explor Newsl 11(1):10–18, doi:10.1145/
parameters. In: Proceedings of the IEEE 1656274.1656278, http://doi.acm.org/
instrumentation and measurement technology 10.1145/1656274.1656278
conference (IMTC ‘05), May 2005, pp
18. Walter SD (2005) The partial area under the sum-
1818–1822
mary ROC curve. Stat Med 24(13):2025–2040
12. Maglogiannis I (2003) Automated segmenta-
19. Gatsonis C, Paliwal P (2006) Meta-analysis of
tion and registration of dermatological images.
diagnostic and screening test accuracy evalua-
J Math Model Algorithm 2:277–294
tions: methodologic primer, research, funda-
13. Maglogiannis I, Pavlopoulos S, Koutsouris D mentals of clinical research for radiologists. Am
(2005) An integrated computer supported J Roentgenol 187:271–281
acquisition, handling and characterization sys-
20. Okeanos Cloud Infrastructure. https://okeanos.
tem for pigmented skin lesions in dermatologi-
grnet.gr/home/
cal images. IEEE Trans Inf Technol Biomed
9(1):86–98 21. Maragoudakis M, Maglogiannis I (2011) A
14. Massone C, Brunasso AMG, Campbell TM, medical ontology for intelligent web-based skin
Peter Soyer H (2009) Mobile teledermoscopy: lesions image retrieval. Health Informatics J
17(2):140–157
 Chapter 30

Melanoma and Other Skin Lesion Detection

Using Smart Handheld Devices
George Zouridakis, Tarun Wadhawan, Ning Situ, Rui Hu, Xiaojing Yuan,
Keith Lancaster, and Courtney M. Queen

Abstract
Smartphones of the latest generation featuring advanced multicore processors, dedicated microchips for
graphics, high-resolution cameras, and innovative operating systems provide a portable platform for run-
ning sophisticated medical screening software and delivering point-of-care patient diagnostic services at a
very low cost. In this chapter, we present a smartphone digital dermoscopy application that can analyze
high-resolution images of skin lesions and provide the user with feedback about the likelihood of malig-
nancy. The same basic procedure has been adapted to evaluate other skin lesions, such as the flesh-eating
bacterial disease known as Buruli ulcer. When implemented on the iPhone, the accuracy and speed achieved
by this application are comparable to that of a desktop computer, demonstrating that smartphone applica-
tions can combine portability and low cost with high performance. Thus, smartphone-based systems can be
used as assistive devices by primary care physicians during routine office visits, and they can have a significant
impact in underserved areas and in developing countries, where health-care infrastructure is limited.

Key words Skin cancer detection, Melanoma screening, Buruli ulcer, Smartphones in health care,
Handheld devices

1  Introduction

During the past two decades, a number of noninvasive diagnostic

procedures have been developed for screening skin cancer, among
which dermoscopy [1, 2] is the one most widely used by clinicians
and researchers alike, mainly because it is significantly better
(>50 %) at detecting melanoma than mere naked eye inspection
[3–5]. Depending on the particular system used, dermoscopy
relies on lesion magnification (macro lenses provide between 6
and 100 times magnification) and special light sources (e.g., oil
immersion or optics-based cross-polarization illumination, transil-
lumination epiluminescence, and oftentimes multispectral light-
ing) to render the epidermal layers of the skin transparent and
inspect features of pigmented lesions not visible to the naked eye.

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0_30, © Springer Science+Business Media New York 2015

459
460 George Zouridakis et al.

Furthermore, digital dermoscopy systems, i.e., computerized

systems that also offer automated image analysis, provide excellent
performance for melanoma detection, with sensitivity and speci-
ficity between 73–96 % and 73–100 %, respectively [4]. An early
study [6] showed that these rates were superior to the ones
attained by general practitioner (who showed a sensitivity and
specificity of 62 % and 63 %, respectively) and comparable to the
rates of dermatologists (80 % and 60 %, respectively) and dermos-
copy experts (90 % and 59 %, respectively). Rightfully then der-
moscopy has been gaining in popularity over the years and, at
present, there are several learning resources readily available,
including books, interactive CDs, and online tutorials for didacti-
cal purposes that can provide information to the general public,
health-care trainees, and specialists (e.g., www.dermoscopy.org,
www.dermoncology.com, www.dermoscop.com).
In terms of equipment, owing to unique technological devel-
opments in the last several years, dermoscopes are currently avail-
able in a variety of forms, including stand-alone devices on a cart
and battery-operated handheld devices that can provide even
multifrequency lesion illumination for true multispectral imaging.
For instance, MelaFind (www.melafind.com) can acquire ten mul-
tispectral images from a pigmented skin lesion and, based on its
morphological characteristics, classify it as having high or low 3D
disorganization in about a minute. Similarly, the DermLite II mul-
tispectral device (www.dermlite.com), when attached to an off-the-­
shelf high-resolution digital camera or the iPhone, can provide four
multispectral images captured with white cross-polarization epilu-
minescence light for surface pigmentation, blue light for surface
color variation, yellow light for superficial vascularity, and red light
for deeper coloration and vascularity, using 32 bright LEDs—8 per
color. The images can then be analyzed in real time on the iPhone
or uploaded to a computer for off-line analysis.
Unprecedented recent technological advances in the field of
digital communications have transformed cellular phones from sim-
ple single-function telephony devices to smartphones, i.e., handheld
mini computers capable of performing complex, processor- and
memory-intensive operations. In particular, the latest generation of
devices provide the user with a wide array of communication, enter-
tainment, global guidance, and image analysis options, thanks to
new multicore processors, dedicated advanced graphics microchips,
and high-resolution cameras. The evolution of smartphone tech-
nology and the explosive growth of smartphone adoption around
the world provide a unique opportunity for delivering image-based
point-of-care medical diagnostic services at a very low cost. This is
especially important in the developing world, where access to health
care resources is often very limited.1
1
 
IEEE-EMBS Special Topics Conference on Point-of-Care Healthcare
Technologies, Bangalore, India, 2013, http://pocht.embs.org/2013/
 Skin Lesion Screening using Smartphones 461

The latest image-based screening methods can be deployed on

both iOS-based (Apple, Inc.) and Android-based (Google, Inc.)
mobile devices, two platforms that are currently dominating the
world of mobile computing. Both platforms offer powerful operat-
ing systems and both of them offer Software Development Kits
(SDKs) and a large collection of Application Programming
Interfaces (APIs) that allow programmers to easily develop native
applications.

1.1  Skin Cancer Risk Over two million people are diagnosed with some form of skin
cancer annually, including malignant melanoma, basal cell carci-
noma, and squamous cell carcinoma [7, 8]. Melanoma represents
only 5 % of all new cases; however, it accounts for the vast majority
of skin cancer deaths globally [9]. In the USA, melanoma is esti-
mated to claim 12,650 lives in 2013 [7], whereas melanoma in situ,
a very early stage of melanoma, can add tens of thousands of more
cases, as it is growing at 15 % per year [10]. For patients whose
melanoma is detected early the overall 5-year survival rate is about
98 %, but this rate falls to 62 % when the disease reaches the lymph
nodes, and to only 15 % when the disease metastasizes to distant
organs [7]. Thus, early detection is of paramount importance.
The most effective method for early detection of melanoma is
skin self-examination [11]. In fact, a substantial percentage of mel-
anomas are detected initially by individuals who examine them-
selves visually [12–14] according to the “ABCDE” rules
recommended by the American Academy of Dermatology [15]
and report to the doctor any suspicious skin changes. However,
the differentiation of early melanoma from other pigmented skin
lesions, such as benign neoplasms that mimic melanoma, is not
trivial, and even primary care physicians may underestimate a mela-
noma in its early stage [16]. This has motivated researchers to
develop automated systems for screening early melanoma that can
be deployed to the general population using the now-a-days ubiq-
uitous smartphones for image acquisition and real time analysis.

1.2  Diagnostic The first studies for automated classification of pigmented skin
Criteria lesions appeared in the literature more than two decades ago [17],
but today the majority of the analysis procedures follow a set of
objective dermoscopic criteria that classify a lesion based on the
presence or absence of certain characteristic features of melanoma.
These criteria were established after several methods were evalu-
ated by an international group of experts during the 2000
Consensus Net Meeting on Dermoscopy [1]. Here we present
only the three most popular lesion-scoring schemes, since these are
the ones we have implemented in our iPhone-based system pre-
sented in later sections. All three scoring schemes are summarized
in Table 1.
462 George Zouridakis et al.

Table 1
Common melanoma detections schemes: diagnostic criteria and scoring rules

Algorithms ABCD rule 7-point checklist Menzies’ method

Criteria (A) × 1.3: Asymmetry Major criteria (2 points): Negative features (−1 point):
(B) × 0.1: Border Atypical pigment network Symmetry
(C) × 0.5: Color Blue-white veil Single color
(D) × 0.5: Differential Atypical vascular pattern Positive features (+1 point):
structures (network, Minor criteria (1 point): Blue-white veil
structureless area, streaks, Irregular streaks Multiple brown dots
dots and globules) Irregular pigmentation Pseudopods
Total score: Irregular dots/globules Radial streaming
4.75: Benign Regression structures Scar-like depigmentation
4.8–5.45: Suspicious Total score: Peripheral black
5.45: Highly suspicious <3: Nonmelanoma dots-globules
≥3: Melanoma Multiple colors (5+)
Multiple blue/gray dots
Broadened network
Total score:
≤0: Nonmelanoma
≥1: Melanoma

The ABCD rule [18, 19] relies on a list of appropriately

weighted visual features (namely, asymmetry, border irregularity,
color irregularity, and the presence of other dermoscopic differen-
tial structures) to compute a total score of malignancy that classi-
fies a lesion as malignant, clinically doubtful, or benign.
Menzies’ method [20] employs both positive features (pres-
ence of blue-whitish veil, multiple brown dots, pseudopods, radial
streaming, depigmentation, peripheral black dots-globules, multi-
ple colors, multiple blue/gray dots, and broadened network) and
negative features (symmetry of pattern or single color) to score a
lesion. Absence of negative features and presence of one or more
positive ones would characterize a lesion as melanoma.
The 7-point checklist [21] uses seven criteria based on char-
acteristic structures of melanoma to evaluate a lesion. These crite-
ria are separated into major (worth two points each) and minor
(worth one point each), and include color and geometric patterns.
A score of three or greater would detect a melanoma with 95 %
sensitivity [22].

1.3  Image Analysis Several melanoma detection systems have been developed during
the past two decades, most of which follow a three-stage procedure
to analyze a lesion, namely, image segmentation, feature extrac-
tion, and lesion classification.
Image segmentation is a critical step as it defines the contour
and thus the symmetry of the lesion, which is a basic feature used
 Skin Lesion Screening using Smartphones 463

for lesion classification. Additionally, it separates the image into

two nonoverlapping regions, foreground (lesion) and background
(healthy skin), each with distinct color and texture features that are
also essential for lesion classification. Algorithms used for image
segmentation are generally of two types: one, region-based meth-
ods, such as adaptive thresholding [23], watershed transformation
[24], and region growing [25, 26] all of which attempt to identify
the entirety of the foreground pixels (pixels within a lesion), and
two, border-based methods, such as adaptive snakes [27] and gradi-
ent vector flow [23], which attempt to detect only the boundary
between foreground and background (the lesion border).
Feature extraction entails the computation of image descrip-
tors using both local features, in which case a lesion is divided into
blocks and features are extracted from each block separately [28–
31], as well as global features extracted from the entire lesion taken
as a whole [32–36]. These low level features typically relate to the
high level scoring schemes described in Subheading 2, and account
for the lesion’s contour, geometry, texture, and color properties.
Lesion classification, finally, relies on the development of a reli-
able classifier, which is typically constructed using a well-defined
dataset with lesions of known classification obtained independently
using the knowledge of one or more domain experts. The data are
typically divided into two sets, one for training and one for testing,
and the classifier automatically assigns a label to each image in the
training set. Then the labels are compared against the actual labels
and some performance measures are computed. This process of
supervised classification is typically repeated several times, each time
adjusting some classification parameters, until satisfactory perfor-
mance measures are obtained. During the deployment phase, how-
ever, the classifier assigns labels to lesions of unknown classification
in an unsupervised fashion.
In terms of classification procedures, several approaches have
been developed over the past several years, including linear dis-
criminant analysis [37–39], artificial neural networks [26, 33, 34,
40, 41], logistic regression [42], k-nearest neighbors [43], and
support vector machines [25]. The performance of these tech-
niques varies substantially, as does the number of images used to
obtain these measures. A recent review [44] summarizing the
results of 16 previous studies showed that, to compute perfor-
mance measures, our studies employed both the largest number of
lesions and the largest number of melanomas (we analyzed 5,363
lesions with 407 melanomas, whereas other studies analyzed as few
as 77 lesions out of which only 5 were melanomas). Similar vari-
ability was found in the overall performance results, showing sen-
sitivity between 73 and 100 %, specificity between 73.9 and 93.8 %,
and area under the curve between 83 and 93.3 %, when a receiver
operator characteristic curve was available.
464 George Zouridakis et al.

Fig. 1 Handheld cameras used by popular melanoma screening systems: (left) Melafind, (middle) SolarScan,
and (right) SIAscope V

1.4  Digital Melafind (MELA Sciences, Irvington, New York), depicted in

Dermoscopy Systems Fig. 1a (http://goo.gl/PQW62), is an FDA approved device for
melanoma detection. It uses multispectral epiluminescence to cap-
ture images at ten wavelengths, ranging from 430 to 950 nm.
Then it performs segmentation on the 430 nm (blue) image to
identify the lesion border and extracts features from all ten images,
including wavelet-based texture, lesion asymmetry, border smooth-
ness, color distribution, and lesion diameter. Subsequently, using a
greedy forward feature-selection approach, these features are
reduced to approximately 13 important ones. Texture and features
extracted from infrared images are the most significant ones for
melanoma detection. In an early study [45] using leave-one-out
learning on 246 images and a linear classifier, Melafind achieved
100 % sensitivity and 84 % specificity, while in a more recent pro-
spective multicenter study [46] with 1,831 difficult skin lesions it
showed a sensitivity of 98 % and specificity of 9.9 %, which was still
superior to that of clinicians (3.7 %).
SolarScan (Polartechnics Ltd, Sydney, Australia), depicted in
Fig.  1b (http://goo.gl/gIDtST), reads pigmented skin lesions
acquired from epiluminescence images and, through MoleMap, it
can also track changes in the morphology of a lesion over time.
SolarScan employs a semi-automated segmentation procedure and
selects a set of features relating to texture patterns, color, and
geometry. Trained on a set of 1,644 skin lesion images and tested
on an independent set of 786 images, SolarScan achieved 85 %
sensitivity and 65 % specificity [6].
SIAscope V (Astron Clinica, UK), depicted in Fig. 1c (http://
goo.gl/efgKoX), uses spectrophotometric intracutaneous analysis
for melanoma recognition. It generates eight images of a skin lesion
using light wavelength ranging from 400 to 1,000 nm. Based on
analysis of optical features and logistic regression for ­classification,
the system can achieve 83 % sensitivity and 80 % specificity [47].

1.5  Smartphone-­ In the last couple of years, a plethora of smartphone applications

Based Dermoscopy has been marketed directly to the public offering a variety of ser-
Systems vices, including information about melanoma and instructions
for skin self-examination and tracking individual lesions over time.
A subset of applications can also analyze lesion images and provide
 Skin Lesion Screening using Smartphones 465

Fig. 2 iPhone attachments for skin cancer screening: (left ) DermLite and (right ) Handyscope attachments

a probability that a lesion is malignant. Some applications allow the

user to analyze existing images, capture images using the smart-
phone camera directly, or use attachments that mount to the device
to capture dermoscopic images. The images are then analyzed on
the phone, or sent to an online service for interpretation by a
dermatologist. Most of these applications are available for both
the iOS (Apple) and Android (Google) platforms, and include
(in alphabetical order) DermScope, Handyscope, Melanoma
iABCD rule, Melanoma Visual Risk Calculator, MelApp, Mole
Detective 2, Skin Scan, SkinVision, and UMSkinCheck to men-
tion but a few. Two of the most popular iPhone attachments,
depicted in Fig. 2, are (again in alphabetical order) the DermLite
(3Gen Inc., San Juan Capistrano, CA) (http://goo.gl/EduFXB)
and the Handyscope (FotoFinder Systems GmbH, Birnbach,
Germany) (http://goo.gl/61OGlI), which offer dermoscopic
polarized or nonpolarized imaging, with or without skin contact.
There are other systems not available for downloading on mobile
app stores, but they can run on smartphones [48], smartphones
and cloud platforms [49], or simply employ general purpose cell
phone based clinical information systems for data transmission that
integrate with medical records systems [50].
For over a decade, our research efforts have focused on the
development of an automatic skin cancer detection system that
uses computerized analysis of dermoscopic images [30, 51–55]
and has resulted in two US patents [56, 57]. We initially developed
a specialized hardware system for melanoma detection [58], but
later we focused on software implementations of the system. The
hardware prototype and the software-based systems are described
in the next sections.

2  Materials

2.1  Hardware
Implementation A block diagram of a self-contained handheld system for automated
detection of melanoma is shown in Fig. 3. The hardware presented is
a prototype for the handheld skin cancer detector and is an alternative
466 George Zouridakis et al.

Fig. 3 Block diagram of a Texas Instruments TMS320DM642-based automated melanoma detection system

to the iPhone-based device. The prototype is based on the Texas

Instruments high speed (5,760 MIPS) TMS320DM642 video/
imaging fixed-point digital signal processor that features a general-
purpose and fully programmable architecture. The system is capable
of capturing, processing, and transmitting high-resolution images
(1,280 × 720 pixels) at 30 frames/s in real time. Communication
with the remote server for image processing and storage is accom-
plished via a Bluetooth module.
This system consists of several key components: a low-noise
3 Mpixel CMOS sensor (Micron MT9001) (see Note 1) converts
the incoming high-resolution images into digital signals streamed
to DM642. The high-resolution images require tremendous pro-
cessing power as well as large data and program memory. The
limited-­
size on-chip memory of TMS320DM642 necessitates
that external SDRAM (such as the Micron MT48LC4M32B2)
be added to the system. A Philips video encoder, the SAA7121
 Skin Lesion Screening using Smartphones 467

digital-to-­
analog converter, converts the digital signal back to
analog for real time image display on an LCD. A boot flash (AMD
AM29LV033C) is used for program storage. A universal asynchro-
nous receiver-transmitter (UART transceiver TI TLC16C752B_9)
is added to the system for interfacing with the Bluetooth module
(TAIYO YUDEN EYMF2CAMM). A keypad and LCD display
terminal provide the human machine interface (HMI) functions.
This prototype system is designed with a HMI board, and includes
buttons and LEDs, which enable the user to control and monitor
system function. The power management module (PWR.MGMT.)
includes various switching regulators to meet the voltage require-
ments of the different blocks in the system. An 8-bit latch extends
control I/Os, and power regulators generate 1.4 V core voltage
and 3.3 V I/O voltage for the TMS320DM642 and other chips.
The TMS320DM642 has several power-down modes to minimize
power consumption when idle. The 1.4 V core voltage can be
reduced to 1.2 V with optimized software (reduced MIPS con-
sumption) and CPU clock can be slowed down for further power
savings. A picture of the prototype for the handheld skin cancer
detector is shown in Fig. 4.

2.2  Software Early on a desktop application was developed [30, 52, 55, 59]
Implementation based on Matlab (MathWorks Inc., Natick, MA) that could auto-
matically analyze a skin lesion image and calculate the probability

Control Buttons: Menu, Up, SDRAM RS232 Transceiver to

Down, Select, … Bluetooth Connector

Video
Out

Video
Encoder

Camera DSP Power Power In

Switching

Fig. 4 Prototype of the handheld skin-cancer detection device

468 George Zouridakis et al.

of malignancy. This application has been subsequently ported to a

C/C++ library that runs on Apple iOS-based devices [55, 60, 61].
Our more recent research [59, 60, 62] has extended beyond skin
cancer to Buruli ulcer [63], which is a flesh-eating bacterial disease
that affects predominately children in sub-Saharan Africa. We now
have a second desktop application based on Matlab that imple-
ments Buruli ulcer detection [59, 62] with computing require-
ments that are sufficiently minimal to allow deployment on a
mobile device, such as the iPhone 5. Our latest endeavors concen-
trate on the development of an encompassing general software
framework to facilitate development of image analysis applications
on mobile and embedded devices [60].

3  Methods

3.1  The System Our skin cancer detection system consists of two main modules,
namely, image acquisition and image analysis. We implement image
acquisition using the DermLite II Multispectral dermoscope (www.
dermlite.com) which can provide white light cross-­ polarization
(XLM) epiluminescence, blue light for surface coloration, yellow
light for superficial vascularity, and red light for deeper coloration
and vascularity, using 32 bright white LEDs—8 per color. This der-
moscope gives the choice between free-floating “dry” skin imaging
and oil-immersion epiluminescence, at any color. The dermoscope
can be attached to a high-resolution camera, e.g., the Sony
Cybershot DSC-W300, which provides a resolution of 13.5 MP.
Under this configuration, image analysis must be implemented off
line on a laptop or tablet computer. Real time analysis is accom-
plished when the dermoscope is attached directly to an iPhone
4/5/5s that combines image acquisition and image analysis in one
device. Furthermore, if multispectral imaging is not required, a
smaller dermoscope, such as the pocket Dermlite DL1, can also be
used with the iPhone. Figure 5 shows various configuration exam-
ples, namely, the DermLite II Multispectral d ­ ermoscope (left), the
same dermoscope attached to an off-the-shelf Sony camera (mid-
dle), and the Dermlite DL1 attached to an iPhone 5s (right).

3.2  Overall Analysis The entire procedure used in our system is graphically depicted in
Procedure Fig. 6. Initially, one or more images of a lesion acquired with light
of different frequencies or under different illumination modalities,
including transillumination (TLM) and cross-polarization (XLM)
epiluminescence, are preprocessed and then segmented with four
different methods. The segmented images are fed into a scoring
stage that takes into account statistical properties of the lesion and
selects the best segmentation result out of which a refined lesion
boundary is computed and superimposed onto the original images.
Subsequently, geometrical, statistical, texture, and color features are
 Skin Lesion Screening using Smartphones 469

Fig. 5 The DermLite II Multispectral dermoscope (left) can be attached to an off-the-shelf Sony Cybershot
camera (middle). Similarly, the Dermlite DL1 can be attached to an iPhone 5s (right)

Fig. 6 Software architecture implemented by our automated system for lesion classification. Multispectral
images acquired under transillumination (TLM) or cross-polarization (XLM) epiluminescence imaging undergo
distinct stages of analysis before a probability of malignancy is computed for the lesion under evaluation

extracted from the segmented lesions using several methods, and

then the combined feature vectors are fed into a feature selection
stage that reduces dimensionality and identifies the best discrimi-
nant features. These features are used for a first partial classification
according to various clinical schemes, including but not limited to
the ABCD rules, Menzies’ method, and the 7-point checklist
explained in Subheading 2. Finally, a decision fusion stage com-
bines the partial classification labels using bagging and boosting
470 George Zouridakis et al.

Fig. 7 Software architecture optimized for the iPhone. A suspicious lesion is evaluated for malignancy sepa-
rately using the ABCD rules, Menzies’ method, and the 7-point checklist

ensemble methods, and provides a probability of malignancy for the

particular skin lesion.
The system is interactive and the user has the option to accept
or reject the automatic segmentation results. In the latter case, the
user can either choose one of the rejected segmentation results, or
manually segment the image, and the system then automatically
updates all computations and displays the new recommendation
on the screen.
The above procedure is currently implemented in Matlab and
runs on personal computers. However, due to memory and pro-
cessing power constraints on handheld devices, a lighter version of
the analysis procedure has been implemented and optimized to
run on the iPhone. We started with porting only the simplest
methods, but newer generations of the iPhone have fast multicore
processors that can handle more advanced algorithms. Thus, we
are now gradually porting also some of the more demanding algo-
rithms to the mobile platform. Each stage/module is implemented
as a single-input single-output function written in the C/C++ lan-
guage. A stable version of the reduced analysis procedure opti-
mized for the iPhone 4 is graphically depicted in Fig. 7.

3.3  Specific Typically, each image is approximately 1.5 MB in size (1,368 × 

Analysis Steps 1,712 pixels) stored in high quality ‘jpeg’ format that defines red
(R), green (G), and blue (B) color components for each pixel
 Skin Lesion Screening using Smartphones 471

individually, thus requiring a 1,368 × 1,712 × 3 data array. RGB

images are typically stored as 24-bit files, where the R, G, and B
components are 8 bits each, potentially allowing for 16 million
colors.
Each image undergoes a completely automated analysis pro-
cedure consisting of five main steps, namely, preprocessing, seg-
mentation, postprocessing, feature extraction, and classification.
Furthermore, each stage entails a series of sub-steps that are sum-
marized below.

3.3.1  Stage 1: The very first processing step involves conversion of color RGB
Preprocessing images to gray scale (Gray) using the standard formula [64] for
averaging the R, G, and B channels, Gray = 0.299 × R + 0.587 × G + 
0.114 × B.
Hair artifacts and salt-and-pepper noise are also removed, or
minimized, using median filtering [65, 66], which in general
reduces noise while it preserves image edges. We use a fast two-
dimensional algorithm [67] with two structuring elements of 5 × 5
and 11 × 11 pixels. In this algorithm, a histogram is used to store
the neighboring pixel values. When moving the center pixel over
the image, only a part of the histogram is modified, thus reducing
the computational complexity from O(n2) to O(n). The exact steps
of the algorithm in a generic pseudocode language are shown in
Fig. 8.
Histogram Equalization, needed to increase the contrast of
images with pixel intensity levels grouped in a narrow range, is
performed by spreading out the most frequent intensity values.
Given a gray scale image of size M × N pixels, we start by comput-
ing the intensity histogram H(j) and the cumulative density func-
tion, cdf(i) = ∑ j = 0iH(j), and then we use monotonic nonlinear

Fig. 8 Huang’s median filtering algorithm optimized for the iPhone

472 George Zouridakis et al.

mapping [68] to distribute the new pixel values uniformly in the

range [0, 255] from the following equation,
 cdf ( i ) − cdf min 
New Pixel Value ( i ) = round  × 255 
 ( M × N ) − cdf min 

Background Correction is accomplished using homomorphic filter-

ing [69], whereby an image (F) is modeled as a product of illumi-
nation (I) and reflectance (R) given by F(x, y, ab) = I(x, y, a) R(x, 
y, b), where a and b are pairs of the color RGB components. Initially
a mask that contains regions with low intensity (less than the 50th
percentile) and high frequency (expressed as high values of the first
derivative) is prepared, and then, for each RGB component an
FL(x, y) = log(F(x, y, ab)) is computed. The illumination component
is modeled by a fourth-order polynomial surface [70]. To fit a sur-
face to the data, the reflective Newton method for minimization is
used for each ab, after excluding regions corresponding to the
previous mask. The illumination is given by I(x, y, a) = exp(SP) and
the reflectance by R(x, y, b) = exp(FL − SP). Background correction
is optional on a handheld device and often omitted since it is very
resource consuming.

3.3.2  Stage 2: Image To identify the border that separates the foreground lesion from
Segmentation the background healthy skin, in our early implementations [70, 71]
we used four classical segmentation methods based on sigmoid
histogram remapping [72], principal component transformation
[32, 73], histogram Gaussian modeling [73], and fuzzy clustering
[74], and later we employed evolutionary strategy-based segmen-
tation. The objective function we adopted favors an ellipse that
divides the image into two homogeneous areas with minimum
variation in both regions, and it is given by F(X, Y, a, b, θ) = ∫ ω|I(x, y) 
− c1|2dxdy + ∫ Ω\ω|I(x, y) − c2|2dxdy, where I(x, y) represents the image
intensity value at coordinates (x, y), ω is the area enclosed by the
ellipse defined by (X, Y, a, b, θ), Ω is the area of pixels with non-
zero intensity value, and c1 and c2 represent the average intensity
value of the pixels inside and outside ω, respectively.
More recently, however, we developed a better method based
on active contours [75] and narrow band graph partitioning
that includes geodesic active contours and gradient vector flow
snakes that rely on gradient information. The method follows
active contour region fusion [53] in a hierarchical approach. Briefly,
it first segments the lesion into small regions based on strict con-
straints on homogeneity and edge strength using region-based
active contours, and then merges the regions together based on
centroid criteria and gradient information. To further improve
robustness and computational efficiency, we followed narrow band
graph partitioning. This approach is more robust against weak or
false edges and against lesion asymmetry. The performance of the
 Skin Lesion Screening using Smartphones 473

Fig. 9 Segmentation of an asymmetric lesion with weak border: (a) modified level set method and (b) narrow
band energy based approach (right) optimized for the iPhone

modified level-set method, shown in Fig. 9a, is compared to our

approach shown in Fig. 9b. The results obtained by the latter
approach closely match manual segmentations by clinicians [53].
Even though these techniques work great on desktop and lap-
top computers, they may be slow on handheld devices, especially of
previous generations. For this reason, we have implemented addi-
tional segmentation techniques, namely, the iterative self-­organizing
data analysis technique algorithm (ISODATA) [76, 77], the itera-
tive Fuzzy C Means (FCM) algorithm [78, 79] and the original
level-set energy-minimization active contour algorithm [80] as a
good compromise between accuracy and execution speed.
The ISODATA algorithm [77] is used to determine a mask in
a gray scale image using a threshold T, so that all pixels greater
than T are set to zero. The various steps of the algorithm in a
generic pseudocode language are shown in Fig. 10. In our imple-
mentation the maximum number of iterations is set to N = 15 and
the length of the smoothing filter is set to K = 5.
The Iterative Fuzzy C Means (FCM) [78, 79] is an
­unsupervised clustering algorithm that assigns objects into two or
more classes, so that objects in the same class have more similar
features than objects in the other classes. To apply FCM, an image
is first converted into a one-dimensional array by concatenating all
rows. FCM then minimizes the following objective function
Jm = ∑ i = 1N ∑ j = 1Cμijm||xi − ci||2, with 1 ≤  m < ∞, where m is a real
number greater than 1, N is the total number of points, C is the
number of clusters known a priori, xi is the i-th object in the
d-dimensional feature space, μij is the degree of membership of
object xi to cluster j, and cj is the centroid of the cluster. The FCM
algorithm is implemented in the steps shown in Fig. 11.
Active Contours [80, 81] are computer-generated curves that
move within images to detect region boundaries. We implement a
level-set method based on minimization of energy [82]. The basic
model is that an image u0 is formed of an inner and outer regions,
u0i and u0o, respectively, where the boundary to be detected,
474 George Zouridakis et al.

Fig. 10 Various steps to compute the threshold T needed to segment an image

using the ISODATA algorithm
 Skin Lesion Screening using Smartphones 475

Fig. 11 Various steps to segment an image using the FCM algorithm

Fig. 12 Various steps to segment an image using the Level Set algorithm

denoted by C0, is represented by u0i. The active contour model

minimizes the term

F ( c1 ,c2 ,C ) = µ ⋅ length ( C ) + λ1 u0 ( x,y ) − c1 dxdy + λ2 u0 ( x,y ) − c2 dxdy

2 2
∫
in ( C )
∫
out ( C )

where μ ≥ 0, λ1 ≥ 0, λ2 ≥ 0 are fixed parameters, C is the unknown
curve, and c1 and c2 are averages of the pixel intensity values inside
and outside the curve C, respectively. The first term regularizes the
length of C. Thus, the minimization problem is formulated as
inf
F ( c1 ,,c2 ,,C ) . The optimized level-set algorithm [72] is
c1 , c2 , C
implemented as described in Fig. 12.
As an example, Fig. 13 compares the segmentation results
obtained when the benign lesion of Fig. 13a was segmented using
Fuzzy C Means and Active Contours. The mask separating
foreground from background pixels obtained from the former
­
algorithm had an ambiguous, fuzzy border shown in Fig. 13b,
whereas the mask produced by the latter had a very clear boundary
as seen in Fig. 13c.
476 George Zouridakis et al.

Fig. 13 Example of image segmentation using the Fuzzy C Means and Active Contour algorithms

Fig. 14 Example showing the need for postprocessing the mask identified by the segmentation algorithm to
remove the area inside the lesion

3.3.3  Stage 3: After segmentation, regardless of the particular method used, of

Postprocessing particular interest is the identification of a lesion area that repre-
sents a single contiguous region without holes or islands. The ini-
tial border of the lesion is typically jagged and may contain disjoint
edges, which, among other things, complicate computation of
measures defined on the lesion, such as border length. An example
demonstrating the need for postprocessing is shown in Fig. 14,
where the segmentation mask of the malignant lesion shown in
Fig. 14a includes the small area seen in Fig. 14b inside the lesion.
However, the small area represents background pixels and must be
removed.
The first requirement of a single contiguous area can be met
using connected component labeling [83], whereas the need for a
continuous smooth border can be accomplished using a curve
smoothing procedure, as described below.

Connected Component We apply this algorithm first to the foreground pixels to remove
Labeling regions outside the lesion and then to the background pixels to
remove holes inside the lesion. More specifically, we start the pro-
cedure by scanning all the pixels in a binary image and grouping
 Skin Lesion Screening using Smartphones 477

them into different components based on similar intensity values

that are, in some way, connected with each other. We then label all
the foreground pixels in the image and, based on the area of the
components, we first change the pixel intensity values to back-
ground and then remove all components except the lesion. Next,
we label all the background pixels inside the lesion and change the
intensity values of these components to foreground. The result is a
binary image with no holes inside the lesion. Thus, the connected
component algorithm [83] requires two passes, whereby the first
pass records equivalences and assigns temporary labels, whereas
the second pass replaces each temporary label by the label of its
equivalence class. The algorithm is implemented efficiently using
the union-find data structure, which can store and find disjoint sets
fast using the operations of union (merges two sets into one) and
find (determines the set to which a particular element belongs).
The set of connected pixels with same intensity values are stored in
a tree structure in which a node represents a label that points to its
parent node. This is accomplished with a vector PARENT whose
subscripts span the set of all possible labels and values are the labels
of the parent nodes. A value of zero means that the node is the
root of the tree. The various steps of the algorithm in a generic
pseudocode language are shown in Fig. 15.

Boundary Smoothing Once segmentation is complete, to compute a smooth continuous

contour from the disjoint edges of the boundary we employ a
series of morphological operations, as follows: (a) if connected
component labeling is not already performed, we use dilation,
erosion, and bridging to fill in the holes within the lesion and
make it homogeneous; (b) alternatively, we employ parametric
curve modeling to obtain a smooth continuous contour from the
disjoint edges of the original boundary [84]. Specifically, for each
pixel in the image the sum of the square root of distances from
each boundary pixel is calculated, and this value is assigned to the
corresponding pixel in a new 2-D image, which defines a “basis
function.” Then, a new 2-D binary array representing the “minor
ridges” is formed from the basis function from the gradients of
the pixels. We then remove spurs from the boundary using mor-
phological operations and use skeletonization to get a boundary
that is only one pixel thick. Optionally, these points can be fitted
with a spline to get an even smoother boundary. Boundary
smoothing is resource intensive and not absolutely necessary, so it
is typically implemented only in the Matlab version of the soft-
ware, but not on the iPhone. As an example, Fig. 16 shows the
various steps needed to obtain a smooth lesion boundary: Fig. 16a
original boundary, Fig. 16b boundary from the basis functions,
Fig. 16c minor ridges, Fig. 16d smooth boundary, and Fig. 16e
smooth boundary superimposed on the original color image.
478 George Zouridakis et al.

Fig. 15 Various steps needed by the connected component algorithm to identify

a contiguous lesion area

3.3.4  Stage 4: Feature After segmentation, we extract a large number of low-level features
Extraction that relate to a lesion’s shape, texture, and color. Typically, this
requires image sampling as a first step. The sampling operator
Lesion Sampling Strategies
selects N pixels inside a lesion and then it centers a patch of p × p
pixels at each location. The number of patches sampled affects the
memory required and the speed of algorithm execution; therefore,
minimization of both parameters is of paramount importance in
handheld devices (see Note 2).
The full desktop system relies on a combination of heuristic
rules, autoregressive modeling, Laws masks, Gabor features, and
wavelet transforms (WTs) to define features that relate to the high-­
level dermoscopic criteria of Subheading 2 [51, 52, 70, 71].
However, through exhaustive parameter exploration, we found
[30, 85] that the handheld implementation can use only a subset
of features that constitute a reasonable trade-off between classifica-
tion accuracy and computation speed.
 Skin Lesion Screening using Smartphones 479

Fig. 16 Various steps to obtain a smooth lesion boundary after segmentation: (a) original boundary, (b) basis
function boundary, (c) minor ridges, (d) smooth boundary, and (e) smooth boundary superimposed on the origi-
nal image

Lesion Geometry Analysis Lesion shape, one of the most important for human perception
low-level lesion features, is highly irregular in several lesions.
Therefore, we extract three measures for estimating lesion irregu-
LesionArea
larity [67]: Total Irregularity, given by IRtot = 4π ;
LesionPerimeter 2
1 4
Sector Irregularity, given by IR sec = ∑ min area ( Si ) − area ( S j ) ,
4 i −1 j =1...4, j ≠ i
where Si, i  = 1…4, are four possibly overlapping sectors
defined by four points on the lesion boundary that intersect
with the minimum rectangle bounding the lesion; and
n −1
larity, given by
Contour Irregu­ IRcon = ∑ d j − d j −1 , where
j =0
y j − y j−w y j −1 − y j − w−1
d j = arctan − arctan represents the curvature
x j − x j−w x j −1 − x j − w−1
at a boundary point pj = (xj, yj), when the boundary is represented
by a series of points {p0, p1, …, pn − 1}, with p0 = pn.

Lesion Texture Analysis Texture analysis relies on wavelet [86] and local binary pattern
[87] analysis. Wavelets are used to hierarchically decompose a
given continuous-time signal into different frequency components.
In general, a continuous function can be represented as the sum of
480 George Zouridakis et al.

a scaling, φ, and wavelet, ψ, functions. The scaling function φ

provides an approximation to the original signal whereas the wave-
let function ψ is added to the approximation to provide detail. For
two-dimensional signals, the discrete wavelet transform (DWT)
scaling, φ, and wavelet, ψ, functions are defined by

φ j , m, n = 2 j / 2 φ ( 2 j x − m, 2 j y − n )

ψ ij ,m ,n = 2 j / 2ψ i ( 2 j x − m, 2 j y − n )

where the index i = {H,V,D} identifies horizontal (H), vertical (V),
and diagonal (D) directional wavelets. The DWT of image f(x,y) of
size M × N pixels is given by
M −1 N −1
1
Wφ ( j ,m,n ) =
MN
∑∑ f ( x,y )φ j 0, m , n ( x ,y )
x =0 y =0

1 M −1N −1
Wψi ( j ,m,n ) = ∑∑
MN x = 0 y = 0
f ( x,y )ψ ij 0,m ,n ( x,y )

Typically we set j0 = 0 and N = M = 2j, with j = 0, 1 … J − 1 and m,
n = 0, 1 … J − 1, so that the image is decomposed into sub-bands
that are dyadically subsampled. After one level of decomposition,
four sub-band images are obtained, but the approximation image
can be further decomposed using the same procedure. In general,
a DWT with d decomposition levels will yield a total of 3d + 1 sub-­
bands. DWT provides information on the various structural pat-
terns present in a skin lesion used by the dermoscopic criteria of
Subheading 2, such as pigmentation networks, vascular structures,
and dots and globules [88].

Haar Wavelet As wavelet function we use the Haar transform, which is given by
ψij(x) = ψ(2jx − i), where i  =  0, …, 2j − 1 and
 1 if 0 ≤ t < 1 / 2

ψ ( t ) = −1 if 1 / 2 ≤ t < 1
0 otherwise

Previous work [45] and our own studies [30] have demonstrated
the effectiveness of Haar wavelet coefficients and local binary pat-
terns [89] for melanoma detection. Images are decomposed into
three levels, each resulting in ten sub-bands images. The extracted
features used by the classifier are the mean and standard deviation
of each sub-band image. The Haar wavelet transform algorithm for
two-dimensional images is given in Fig. 17, and the feature extrac-
tion process is summarized in Fig. 18. For local feature extraction,
the image is divided into overlapping patches of K × K pixels cen-
tered on a square grid of size M pixels. When the algorithm runs
 Skin Lesion Screening using Smartphones 481

Fig. 17 Algorithm for the two-dimensional Haar wavelet transform of an image

optimized for the iPhone

on a handheld device we want to reduce computation time, so we

choose M = 10 pixels for the grid size, K = 24 pixels for the patch
size, and L = 200 as the number of clusters in the feature space.

Local Binary Pattern (LBP) LBP is a robust texture operator [89] defined on gray scale images.
It is invariant to both monotonic transformation of intensity and
to rotation. It can be derived using a symmetric neighborhood of
P members on a circle of radius R, where the parameter P repre-
sents the quantization of angular space in the circular neighbor-
hood, and R represents the spatial resolution. A limited number of
transitions or discontinuities (0/1 changes in LBP) are allowed to
reduce the noise and improve discrimination of features; however,
we restrict the maximum number of transitions to P, so that transi-
tions greater than P are considered equal. An occurrence histo-
gram, denoted by LBPPRriu, provides statistical and structural
information on the image and is computed following the steps
shown in Fig. 19.

Lesion Color Analysis Several dermoscopic criteria listed in Subheading 2 analyze color
intensity and patterns of color distribution to characterize a skin
Color Histograms
lesion. Examples include the presence of brown dots, blue-whitish
veil, and regression, which consist of mixtures of certain colors.
482 George Zouridakis et al.

Fig. 18 Feature extraction process based on Haar Wavelets optimized for the
iPhone

Fig. 19 Computation of the local binary pattern histogram on a gray-scale image

To extract the color information of a lesion, we compute several

color histograms from the intensity values of all foreground pixels
using the original image, which is almost always acquired in the
RGB color space, as well as two additional images obtained by
­transforming the RGB image into the HSV and LAB color spaces.2
The latter are invariant to illumination changes [90], and thus they
can reduce variability due to the lighting conditions under which
the original RGB dermoscopic images were taken. The various
steps for color histogram feature extraction are summarized in
Fig. 20. The intensity range in each channel, for each of the three-
color spaces, is divided into P fixed-length bins and a histogram is
build by counting the number of pixels belonging to each bin.
From the resulting nine histograms, we extract statistical features,
such as standard deviation and entropy, which are used for classifi-
cation. In particular, entropy is defined as entropy = ∑ i = 1Pf(histogr
am[i]), where histogram[i] is the normalized pixel count of i-th
n log 2 ( n ) , n > 0
bin and f ( n ) =  .
 0, n = 0

2
 The RGB color space uses separate channels for red, green, and blue color
intensity, HSV uses separate channels for hue, saturation, and brightness,
whereas LAB uses one channel for luminance L, and two color channels,
A and B, that correspond to perceived intensity of complementary colors.
 Skin Lesion Screening using Smartphones 483

Fig. 20 Feature extraction based on color histograms

Fig. 21 Various steps used during the training and testing phases of classifier
development

3.3.5  Stage 5: All the features extracted from a lesion, such as shape Si, color
Classification histogram Ci, Haar wavelet Hi, and local binary patterns Li are
combined to form a singe feature vector Fi =  {Si,  Ci, Hi, Li}.
Depending on what particular dermoscopic criteria we want to use
for lesion assessment, we filter the feature vector to select a subset
of Fi keeping only those features with the highest classification
accuracy for the particular dermoscopic rule. Correlation coeffi-
cient values between Fi and each criterion are used to rank the fil-
ters. Each criterion requires both training and testing before it is
deployed. To build a support vector machine (SVM) classifier, typi-
cally a dataset is divided into several folds, each containing both
malignant and benign lesions. The majority of the folds are used
for training of the classifier, whereas the remaining folds are used
for testing. The optimum number of features and parameters of
the linear SVM classifier are obtained during the training phase by
exhaustive search. The two phases of building a classifier are shown
in Fig. 21.
Even though the full version of the system employs several clas-
sification algorithms, the handheld device uses only the so-­called
“2C” formulation of the SVM algorithm [91] that can successfully
484 George Zouridakis et al.

control both the sensitivity and specificity of classification [52].

The classifier parameters are identified during the training phase
using a set of images of known labels. For a given set ­containing m
benign lesions (x1, y1), …, (xm, ym) and n malignant lesions (xm + 1, y
m + 1), …, (xm + n,ym + n) the algorithm minimizes the following objective
function,
m m+n
1 2
min w,b ,ξ w + C1 (C2 ∑ξi + (1 − C2 ) ∑ ξi )
2 i =1 i =1+ m
subject to the constraint that yi(wT ⋅ ϕ(xi) + b) ≥ 1 − ξi, i = 1, 2, …, m + 
n and ξi ≥ 0, i = 1, 2, …, m + n, where φ(xi) is the feature vector of the
i-th lesion xi and yi the corresponding label. C1 represents the trad-
eoff between the regularization term and the empirical cost esti-
mated from the training set, whereas C2 varies continuously in the
open interval (0, 1) and represents the weights of the two classes.

3.3.6  Problems Having imbalanced classes in a small sample size is a common

with Small Sample problem encountered in image classification, as the imbalance is
Imbalance Classification often an inherent part of the data. For instance, in the case of skin
cancer, at any time there are many more living people without
melanoma than with melanoma. Yet, typical machine learning and
data mining algorithms are designed based on the assumption that
the classes are balanced and that the unseen test data, on which the
classifier will make a prediction, are drawn from the same distribu-
tion as the training data [92]. If the test data samples are drawn
from a different distribution, the performance of the algorithm is
hindered, since a classifier cannot generalize the data characteristics
without a large training set, or it may overfit the training data and
be misled during testing due to the high dimensionality of feature
space. When the sample size is small, these problems are more
severe [93, 94], but special techniques can be employ to address
potential problems (see Note 3).

3.4  Application This section illustrates skin lesion analysis on the Apple iPhone.
Examples For this particular example, we used images from the Interactive
CD of Dermoscopy,3 a large commercial library of skin cancer
3.4.1  Detection
images annotated by expert dermatologists. The library contains
of Skin Cancer
360 lesions, 270 benign and 90 melanomas; however, 13 images
that did not produce satisfactory segmentation results were
excluded from this analysis.
All analyses were done on an iPhone 4, using our menu-driven
application that implements all the algorithms of the automated
procedure for lesion classification described in the previous sections.
The user can select any dermoscopic image from the database or
take a new one using the DermLite attachment, analyze it using

3
 http://www.dermoscopy.org/atlas/cd-review.asp
 Skin Lesion Screening using Smartphones 485

Fig. 22 Example of automatic skin lesion analysis on the iPhone: (a) splash screen and (b) classification meth-
ods available for lesion analysis; (c) final lesion classification: the black line represents the border of the seg-
mented lesion; the score according to the 7-point checklist is shown at the bottom of the screen

the ABCD rule, Menzies’ method, or the 7-point checklist, and

see the results displayed on the screen in quasi real time.
Screen shots from the analysis of a malignant melanoma are
shown in Fig. 22. In the example, we demonstrate the use of the
7-point checklist. The application starts with the plash screen seen
in Fig. 22a. The user can perform lesion analysis using several der-
moscopic methods from the segmentation-method selection screen
shown in Fig. 22b. In the analysis results screen shown in Fig. 22c,
the black line represents the border of the segmented lesion,
whereas the final classification and the score for each criterion on
the 7-point checklist are shown at the bottom of the screen. Total
analysis takes only a few seconds.
To assess classification performance, we followed a tenfold
cross-validation procedure by dividing the 347 good images into
10 folds, 9 folds with 11 melanomas and 23 benign lesions, and 1
fold with 11 melanomas and 30 benign lesions. We performed ten
rounds of validation, each time choosing nine folds for training
and one for testing, to get a total of ten experimental results.
In Table 2 we present the confusion matrix computed as the
average of tenfold cross-validation across the ten test sets, along
with the corresponding overall sensitivity and specificity of the
automated analysis procedure. We compare the sensitivity and
specificity of our algorithms with the decision made by expert clini-
cians via dermoscopy.
486 George Zouridakis et al.

Table 2
Confusion matrix computed as the average of tenfold cross-validation
across ten test sets and the resulting average sensitivity and specificity
of the automated analysis procedure on the iPhone

iPhone

Expert Melanoma Benign

Melanoma 96  14
Benign 68 169
Sensitivity: 87.27 % Specificity: 71.31 %

Table 3
Classification results using individual and combined criteria of the 7-point
checklist on the iPhone

Sensitivity (%) Specificity (%)

Atypical Pigment Network 72.86 70.40
Blue-Whitish Veil 79.49 79.18
Atypical Vascular Pattern 75.00 69.66
Irregular Streaks 76.74 79.31
Irregular Pigmentation 69.47 74.21
Irregular Dots and Globules 74.05 74.54
Regression Structures 64.18 67.86
Combined features 87.27 71.31

In Table 3 we report the classification accuracy, in terms of

sensitivity and specificity, separately for each criterion of the 7-point
checklist and for all of the criteria combined (last row).

Execution Time We contrast the time needed for classification using the ISODATA
segmentation algorithm as implemented on the Apple iPhone 3G
against a typical desktop computer (2.26 GHz Intel Core 2 Duo
with 2GB RAM). The choice of the iPhone 3G would provide the
worst-case scenario for the execution time for our algorithm, since
all the following iPhone models are much faster.4 Classification
time includes time needed for feature extraction.
4
 The iPhone 3G features a single core 0.6 GHz ARM Cortex processor,
0.25 GB of RAM memory, and quad core 150 MHz PowerVR SGX535 GPU
with screen resolution of 164 ppi, whereas the latest model iPhone 5s features
a dual core 1.3 GHz ARM Cyclone (64 bit), 1 GB of RAM memory, and
quad core 300 MHz, PowerVR Series 6 Rogue G6430 GPU with screen reso-
lution of 325 ppi (source: http://goo.gl/88YB3U).
 Skin Lesion Screening using Smartphones 487

Table 4
Average computation time in seconds for the iPhone and desktop
platforms

Mean time (s) Apple iPhone 3G Desktop computer

Segmentation 2.3910 0.1028
Classification 7.4710 0.2415

Table  4 shows computation time in seconds for both the

iPhone and the desktop platforms. It can be seen that the whole
procedure takes under 10 s to complete. This proves that the appli-
cation is light enough to run on a smartphone with limited com-
putation power and memory compared to desktop computers.

3.4.2  Detection of Buruli The procedures presented in the previous sections can be adapted
Ulcer Disease and applied to other skin diseases that require early detection.
We have recently developed [59, 62] an application for automatic
detection of Buruli ulcer [63, 95, 96], a devastating flesh-eating
bacterial infection that, each year, affects thousands of people,
mostly children under the age of 15 years [97, 98], and results in
life debilitating amputations. However, if detected early, BU can
be easily treated with oral medication and cured [99].
BU image analysis starts with image acquisition, using either a
camera or the iPhone with a DermLite IIm attachment, which can
acquire multispectral images. Once an image is loaded, the system
can automatically perform lesion preprocessing, segmentation, fea-
ture extraction, feature selection, classification, and graphically
provide a probability of the lesion being Buruli, as shown in Fig. 23
(see Note 4).

3.5  New Framework The process of implementing new methods on mobile devices is
to Facilitate rather challenging, as a new algorithm typically undergoes a series
Development of steps before it can be implemented on the handheld device: it is
on Mobile Device first implemented in Matlab (for testing), then it is ported to the
C/C++ language and compiled for use by Matlab as a “mex”
library, and finally it is ported to the host language for the target
device. These steps also require modifying the C code to use mem-
ory management and threading functions from the operating system,
and writing or modifying a substantial amount of Objective-C
code to expose the new algorithm to the user interface and display
the results. If, as is often the case, the original algorithm is later
modified/improved using Matlab, some or all of the intermediate
steps must be repeated.
488 George Zouridakis et al.

Fig. 23 Automatic Buruli ulcer recognition system

Fig. 24 Overview of the new framework that optimizes code development and
maintenance for the Apple iOS 7.1 operating system and iPhone 5s architecture

To make porting of new software to a mobile device more effi-

cient, we are developing a software framework that abstracts away
the details of the underlying operating system [100]. An overview
of the new framework architecture is shown in Fig. 24. Briefly, the
new framework provides a complete abstraction of the operating
system and is capable of handling memory management and stor-
age; provides parallel process management without negatively
affecting overall performance; allows the development of
processing-­code libraries in C, C++, Objective-C, Ruby, or any
other language that can be compiled for the target device; allows
the interchange of processing components with minimal code
modifications; minimizes or eliminates the need to program using
 Skin Lesion Screening using Smartphones 489

the native programming language of the host device; and takes full
advantage of multiprocessor platforms, graphical processing units
(GPUs), and other recently developed hardware features.

4  Notes

1. Alternatively, the VS6502 is a VGA resolution Small Optical

Package (SmOP) sensor module that combines the image sen-
sor and fixed focus lens system in a single module. The sensor
outputs 10-bit raw digital image data that directly interface to
the microprocessor.
2. There are different sampling strategies that can be separated
into three groups: square grid sampling (selects patches on a
regular grid of fixed size within the lesion, but this method is
memory and computationally intensive) [101], random sam-
pling (selects patches randomly within the lesion, but it may
miss important lesion features and thus in general provides
decreased discriminative power) [102], and sparse sampling
based on key points (selects patches centered on special points,
such as blobs, corners, dots, and globular structures) [103],
or sparse sampling based on salient features (selects patches
based on feature co-occurrence and measures how well an
individual patch can predict the occurrence of other patches in
the same image) [104]. We have developed a new method for
nonuniform sampling that extracts more patches from dermo-
scopically interesting regions, which are defined by segmenta-
tion based on patch saliency [59]. As an example, various
strategies for sampling a malignant lesion are shown in
Fig. 25a–d, and include square grid based sampling, random
sampling, key-­point based sampling, and saliency based sam-
pling, respectively.
3. To address the problem of imbalanced classes in a small sample
size classification, several technique have been developed
based on resampling, namely, undersampling the majority
class [100, 105], oversampling the minority class [100–107],
or both [108]; one-class learning methods built to recognize
samples from a given class and reject samples from other classes
[109–111]; and feature selection, whereby algorithms use a
subset of features that provide optimal performance [112].
We have addressed this problem by developing a new resam-
pling method that increased the population of the minority
class by splitting lesions directly in the data space and treating
them as separate samples, rather than synthesizing new samples
in the feature space [59].
4. To make the application transferable to our mobile operating
system, we have built a library and translated all code from
490 George Zouridakis et al.

Fig. 25 Examples showing different lesion sampling strategies based on (a) dense square-grid, (b) random,
(c) key-point, and (d) saliency sampling. Blue circles correspond to the patch centers selected by various
sampling methods

Matlab to the C language, in discrete modules, which can be

easily incorporated into the new mobile computing framework
currently under development, as explained in Subheading 3.5.

5  Concluding Remarks

During the past decades, dermoscopy has resulted in a reduction

of unnecessary excisions [5] by providing physicians with a reli-
able tool to analyze skin lesions and help them decide if a mole is
potentially a melanoma that should be removed. Several studies
have shown that dermatologists that are users of dermoscopy in
general achieve higher rates of melanoma detection compared to
nonusers [3–5].
Over the years, digital dermoscopes that analyze lesions auto-
matically have undergone remarkable improvements in terms of
size, portability, ease of use, speed of analysis, and overall accu-
racy. Availability and ease of use are especially true for the latest
generation of dermoscopes that consist of a readily available
smartphone, like the iPhone, that runs a software app and an
external attachment that provides polarized illumination and
lesion magnification. The accuracy, however, of these systems is
still lacking. In a recent study assessing the diagnostic accuracy of four
 Skin Lesion Screening using Smartphones 491

smartphone apps for melanoma detection, three of these apps

incorrectly classified 30 % of melanomas as unconcerning [113].
Even though this study tested only four apps, its findings suggest
that further improvements are needed before smartphone apps
are can be deployed as a user friendly and clinically useful tool for
diagnosis of skin cancer. This is one of the reasons why we have
not released our application for general use yet.
A second reason has to do with the issues recently identified by
another study [114] as barriers for clinical adoption of current smart-
phone apps, namely, a variety of technological issues and the time-
consuming nature of data collection and patient documentation.
A third reason relates to our belief that smartphone apps and
other automated tools of this nature are clinically more useful
when used for screening rather than diagnostic purposes. That is,
instead of trying to maximize the sensitivity of a tool in detecting
correctly all true cancerous lesions, one should maximize its speci-
ficity and its ability to identify correctly the maximum number of
healthy subjects. In theory, a device can achieve 100 % sensitivity by
sending all lesions screened to biopsy, but the value of such a device
as a screening tool is low. In contrast, higher specificity on one
hand has the advantage of avoiding a huge number of unnecessary
biopsies, as most skin pigmentations are benign, and on the other,
makes the use of the device by a nonspecialist less dangerous.
Accepting a larger number of false positives (lower sensitivity) is
less risky than accepting a larger number of false negatives (lower
specificity). In general, smartphone-based dermoscopes can be
wonderful tools in the hands of experts; however, only a physician
should make a decision whether or not a suspicious lesion needs addi-
tional attention. Otherwise, in the hands on nonexperts, smart-
phone apps, especially when used as a replacement for medical
advice, may give a false sense of security, delay diagnosis of a malig-
nant lesion, and ultimately harm the patient [113].
In conclusion, smartphone-based systems can be used as
­assistive devices by primary care physicians during routine screen-
ing examinations, and may play a significant role by affording
health-­care providers a rapid and valid “second opinion” at the
point of care or by offering easy access to teleconsultation [115]
when health-care infrastructure is limited, as in the case of under-
served areas and in developing countries.

Acknowledgment 

This research has been supported in part by NSF grant 521527 and
NIH grant 5R21AR057921-02. This chapter was developed while
GZ was a Visiting Professor at the Basque Center on Cognition,
Brain, and Language, in Donostia-San Sebastián, Spain, and a
Fellow of the Ikerbasque Foundation, Basque Country, Spain.
492 George Zouridakis et al.

References
1. Argenziano G, Soyer HP, Chimenti S, 16. Pariser RJ, Pariser DM (1987) Primary care
Talamini R, Corona R, Sera F et al (2003) physicians errors in handling cutaneous disor-
Dermoscopy of pigmented skin lesions: ders. J Am Acad Dermatol 17:239–245
results of a consensus meeting via the Internet. 17. Cascinelli N, Ferrario M, Tonelli T, Leo E
J Am Acad Dermatol 48(5):679–693 (1987) A possible new tool for clinical diag-
2. Argenziano G, Soyer HP, De Giorgi V, nosis of melanoma: the computer. J Am Acad
Piccolo D, Carli P, Delfino M (2002) Dermatol 16(2):361–367
Dermoscopy: a tutorial. EDRA, Medical 18. Stolz W, Riemann A, Cognetta AB, Pillet L,
Publishing & New Media, Milan Abmayr W, Holzel D et al (1994) Abcd rule
3. Bafounta ML, Beauchet A, Aegerter P et al of dermatoscopy: a new practical method for
(2001) Is dermoscopy (epiluminescence early recognition of malignant-melanoma.
microscopy) useful for the diagnosis of mela- Eur J Dermatol 4(7):521–527
noma? Arch Dermatol 137:1343–1350 19. Nachbar F, Stolz W, Merkle T, Cognetta A,
4. Kittler H, Pehamberger H, Wolff K et al Vogt T, Landthaler M, Bilek P, Braun-Falco
(2002) Diagnostic accuracy of dermoscopy. O, Plewig G (1994) The ABCD rule of der-
Lancet Oncol 3:159–165 matoscopy: high prospective value in the
5. Menzies SW (2013) Evidence-based dermos- diagnosis of doubtful melanocytic skin lesions.
copy. Dermatol Clin 31(4):521–524 J Am Acad Dermatol 30(4):551–559
6. Menzies SW, Bischof L, Talbot H, Gutenev A 20. Menzies SW, Ingvar C, Crotty KA, McCarthy
et al (2005) The performance of SolarScan: WH (1996) Frequency and morphologic
an automated dermoscopy image analysis characteristics of invasive melanomas lacking
instrument for the diagnosis of primary mela- specific surface microscopic features. Arch
noma. Arch Dermatol 141(11):1388–1396 Dermatol 132(10):1178
7. ACS: American Cancer Society, Cancer Facts & 21. Argenziano G, Fabbrocini G, Carli P, De
Figures 2013. http://goo.gl/rgD56, Accessed Giorgi V, Sammarco E, Delfino M (1998)
31 Sep 2013 Epiluminescence microscopy for the diagno-
8. Rogers HW, Weinstock MA, Harris AR et al sis of doubtful melanocytic skin lesions: com-
(2010) Incidence estimate of nonmelanoma parison of the ABCD rule of dermatoscopy
skin cancer in the United States, 2006. Arch and a new 7-point checklist based on pattern
Dermatol 146(3):283–287 analysis. Arch Dermatol 134(12):1563
9. Parkin D, Bray F, Ferlay J, Pisani P (2005) 22. Johr RH (2002) Dermoscopy: alternative
Global cancer statistics, 2002. CA Cancer J melanocytic algorithms—the ABCD rule of
Clin 55(2):74–108 dermatoscopy, menzies scoring method, and
10. Thorn M, Penten F, Johansson B et al (1998) 7-point checklist. Clin Dermatol 20(3):
Rapid increase in diagnosis of cutaneous mel- 240–247
anoma in situ in Sweden, 1968–1992. Cancer 23. Silveira M, Nascimento JC, Marques JS,
Detect Prev 22:430 Marçal AR, Mendonça T, Yamauchi S et al
11. Pollitt R, Geller A, Brooks D, Johnson T, (2009) Comparison of segmentation meth-
Park E, Swetter S (2009) Efficacy of skin self-­ ods for melanoma diagnosis in dermoscopy
examination practices for early melanoma images. IEEE J Sel Top Signal Process 3(1):
detection. Cancer Epidemiol Biomarkers Prev 35–45
18:3018–3023 24. Wang H, Chen X, Moss RH, Stanley RJ,
12. Brady MS, Oliveria SA, Christos PJ et al Stoecker WV, Celebi ME et al (2010)
(2000) Patterns of detection in patients with Watershed segmentation of dermoscopy
cutaneous melanoma. Cancer 89(2):342–347 images using a watershed technique. Skin Res
13. Epstein DS, Lange JR, Gruber SB, Mofid M, Technol 16(3):378–384
Koch SE (1999) Is physician detection associ- 25. Celebi ME, Kingravi HA, Uddin B, Iyatomi
ated with thinner melanomas? JAMA H, Aslandogan YA, Stoecker WV, Moss RH
281(7):640–643 (2007) A methodological approach to the
14. McGuire ST, Secrest AM, Andrulonis R, classification of dermoscopy images. Comput
Ferris LK (2011) Surveillance of patients for Med Imaging Graph 31(6):362–373
early detection of melanoma: patterns in der- 26. Iyatomi H, Oka H, Celebi ME, Hashimoto
matologist vs patient discovery. Arch M, Hagiwara M, Tanaka M, Ogawa K (2008)
Dermatol 147(6):673–678 An improved internet-based melanoma
15. AAD: American Academy of Dermatology. screening system with dermatologist-like
How to perform a self-exam. http://goo. tumor area extraction algorithm. Comput
gl/6D9E1d. Accessed 27 Oct 2013 Med Imaging Graph 32(7):566–579
 Skin Lesion Screening using Smartphones 493

27. Nascimento JC, Marques JS (2005) Adaptive 39. Green A, Martin N, Pfitzner J, O’Rourke M,
snakes using the EM algorithm. IEEE Trans Knight N (1994) Computer image analysis in
Image Process 14(11):1678–1686 the diagnosis of melanoma. J Am Acad
28. Seidenari S, Pellacani G, Grana C (2005) Dermatol 31(6):958–964
Pigment distribution in melanocytic lesion 40. Ercal F, Chawla A, Stoecker WV, Lee HC,
images: a digital parameter to be employed Moss RH (1994) Neural network diagnosis
for computer‐aided diagnosis. Skin Res of malignant melanoma from color images.
Technol 11(4):236–241 IEEE Trans Biomed Eng 41(9):837–845
29. Situ N (2011) Melanoma recognition using 41. Rubegni P, Cevenini G, Burroni M, Perotti R,
mid-level features and high-level annotations. Dell'Eva G, Sbano P et al (2002) Automated
Doctoral dissertation, University of Houston diagnosis of pigmented skin lesions. Int J
30. Situ N, Yuan X, Chen J, Zouridakis G (2008) Cancer 101(6):576–580
Malignant melanoma detection by bag-of-­ 42. Blum A, Luedtke H, Ellwanger U, Schwabe R,
features classification. In 30th annual interna- Rassner G, Garbe C (2004) Digital image anal-
tional conference of the IEEE Engineering in ysis for diagnosis of cutaneous melanoma.
Medicine and Biology Society, 2008 (EMBS Development of a highly effective computer
2008). IEEE, pp 3110–3113 algorithm based on analysis of 837 melanocytic
31. Zhou H, Chen M, Rehg JM (2009) lesions. Br J Dermatol 151(5):1029–1038
Dermoscopic interest point detector and 43. Ganster H, Pinz P, Rohrer R, Wildling E,
descriptor. In IEEE international symposium Binder M, Kittler H (2001) Automated mela-
on biomedical imaging: from nano to macro, noma recognition. IEEE Trans Med Imaging
2009. ISBI'09. IEEE, pp 1318–1321 20(3):233–239
32. Celebi ME, Aslandogan YA, Bergstresser PR 44. Caetano Ruela AM (2012) What is the role of
(2005) Unsupervised border detection of color in dermoscopy analysis? Master of Science
skin lesion images. In International confer- Thesis, Biomedical Engineering, Instituto
ence on information technology: coding and Superior Técnico, Lisbon, Portugal, 2012
computing, 2005 (ITCC 2005), vol 2. IEEE, 45. Elbaum M, Kopf AW, Rabinovitz HS, Langley
pp 123–128 RG, Kamino H, Mihm MC Jr et al (2001)
33. Hoffmann K, Gambichler T, Rick A, Kreutz Automatic differentiation of melanoma from
M, Anschuetz M, Grünendick T et al (2003) melanocytic nevi with multispectral digital
Diagnostic and neural analysis of skin cancer dermoscopy: a feasibility study. J Am Acad
(DANAOS). A multicentre study for collec- Dermatol 44(2):207–218
tion and computer‐aided analysis of data from 46. Monheit G, Cognetta AB, Ferris L, Rabinovitz
pigmented skin lesions using digital dermos- H, Gross K, Martini M et al (2011) The per-
copy. Br J Dermatol 149(4):801–809 formance of MelaFind: a prospective multi-
34. Rubegni P, Burroni M, Cevenini G, Perotti R, center study. Arch Dermatol 147(2):188
Dell'Eva G, Barbini P et al (2002) Digital 47. Moncrieff M, Cotton S, Claridge E, Hall P
dermoscopy analysis and artificial neural net- (2002) Spectrophotometric intracutaneous
work for the differentiation of clinically atypi- analysis: a new technique for imaging pig-
cal pigmented skin lesions: a retrospective mented skin lesions. Br J Dermatol 146(3):
study. J Investig Dermatol 119(2):471–474 448–457
35. Serrano C, Acha B (2009) Pattern analysis of 48. Karargyris A, Karargyris O, Pantelopoulos A
dermoscopic images based on Markov random (2012) DERMA/care: an advanced image-­
fields. Pattern Recognit 42(6):1052–1057 processing mobile application for monitoring
36. Tanaka T, Torii S, Kabuta I, Shimizu K, skin cancer. In IEEE 24th international con-
Tanaka M (2008) Pattern classification of ference on tools with artificial intelligence
nevus with texture analysis. IEEE Trans Elect (ICTAI), vol 2, 2012. IEEE, pp 1–7
Electron Eng 3(1):143–150 49. Doukas C, Stagkopoulos P, Kiranoudis CT,
37. Andreassi L, Perotti R, Rubegni P, Burroni Maglogiannis I (2012) Automated skin lesion
M, Cevenini G, Biagioli M et al (1999) assessment using mobile technologies and
Digital dermoscopy analysis for the differen- cloud platforms. In annual international con-
tiation of atypical nevi and early melanoma: a ference of the IEEE Engineering in Medicine
new quantitative semiology. Arch Dermatol and Biology Society (EMBC), 2012. IEEE,
135(12):1459 pp 2444–2447
38. Green A, Martin N, McKenzie G, Pfitzner J, 50. Celi LA, Sarmenta L, Rotberg J, Marcelo A,
Quintarelli F, Thomas BW et al (1991) Clifford G (2009) Mobile care (moca) for
Computer image analysis of pigmented skin remote diagnosis and screening. J Health
lesions. Melanoma Res 1(4):231–236 Inform Dev Ctries 3(1):17
494 George Zouridakis et al.

51. Yuan X, Situ N, Zouridakis G (2008) 63. Johnson PD, Azuolas J, Lavender CJ, Wishart
Automatic segmentation of skin lesion images E, Stinear TP, Hayman JA et al (2007)
using evolution strategies. Biomed Signal Mycobacterium ulcerans in mosquitoes cap-
Process Control 3:220–228 tured during outbreak of Buruli ulcer, south-
52. Situ N, Wadhawan T, Hu R, Lancaster K, eastern Australia. Emerg Infect Dis 13(11):1653
Yuan X, Zouridakis G (2011). Evaluating 64. Gonzalez R, Woods R (2002) Digital image
sampling strategies of dermoscopic interest processing. Prentice Hall, Upper Saddle
points. In IEEE international symposium on River, NJ, pp 349–408
biomedical imaging: from nano to macro, 65. Tukey J (1977) Exploratory data analysis.
2011. IEEE, pp 109–112 Addison-Wesley, Menola Park, CA
53. Yuan X, Situ N, Zouridakis A (2009) A nar- 66. Huang T, Yang G, Tang G (1979) A fast two-­
row band graph partitioning method for skin dimensional median filtering algorithm. IEEE
lesion segmentation. Pattern Recognit Trans Acoust Speech Signal Process 27(1):
42(6):1017–1028 13–18
54. Hu R, Queen CM, Zouridakis G (2012) 67. Huang PW, Lai YH (2010) Effective segmen-
Lesion border detection in Buruli ulcer tation and classification for HCC biopsy
images. In annual international conference of images. Pattern Recognit 43(4):1550–1563
the IEEE Engineering in Medicine and 68. Acharya T, Ray AK (2005) Image processing:
Biology Society (EMBC), 2012. IEEE, principles and applications. Wiley-­
pp 5380–5383 Interscience, New York
55. Wadhawan T, Situ N, Rui H, Lancaster K, 69. Fries R, Modestino J (1979) Image enhance-
Yuan X, Zouridakis G (2011) Implementation ment by stochastic homomorphic filtering.
of the 7-point checklist for melanoma detec- IEEE Trans Acoust Speech Signal Process
tion on smart handheld devices. In annual 27(6):625–637
international conference of the IEEE 70. Zouridakis G, Doshi M, Mullani N (2004)
Engineering in Medicine and Biology Society, Early diagnosis of skin cancer based on seg-
EMBC, 2011. IEEE, pp 3180–3183 mentation and measurement of vasculariza-
56. Zouridakis G, Yuan X, Chen J, Stotzer E, Wu tion and pigmentation in nevoscope images.
Y (2012) Device and software for screening In 26th annual international conference of
the skin. University of Houston, July 2012. the IEEE Engineering in Medicine and
US 08213695 Biology Society, 2004. IEMBS’04, vol 1.
57. Zouridakis G, Yuan X, Chen J, Stotzer E, Wu Y IEEE, pp 1593–1596
(2008) Device and software for screening the 71. Zouridakis G, Doshi M, Duvic M, Mullani NA
skin, September 2008. US 20080226151-A1 (2005) Transillumination imaging for early
58. Wu Y, Chen J, Yuan X, Zouridakis G (2009) A skin cancer detection. Biomedical Imaging
programmable DSP development platform for Lab, Department of Computer Science,
automated detection of melanoma. Int J Intel University of Houston, Houston, TX. http://
Comput Med Sci Image Process 3(1):1–8 goo.gl/NPkLBp
59. Hu, R (2013) Automated recognition of 72. Song B, Chan T (2002) A fast algorithm for
Buruli ulcer images on smart handheld devices. level set based optimization. UCLA Cam
Doctoral dissertation, University of Houston Report, 02-68
60. Lancaster K, Zouridakis G (2014) A software 73. Sapiro G (2000) Segmenting skin lesions with
framework to facilitate development of image partial-differential-equations-based image
analysis applications on mobile and embed- processing algorithms. IEEE Trans Med
ded devices. PECCS 2014: International con- Imaging 19(7):763–767
ference on pervasive and embedded 74. Erkol B, Moss RH, Joe Stanley R, Stoecker
computing and communication systems, 7–9 WV, Hvatum E (2005) Automatic lesion
Jan 2014, Lisbon, Portugal boundary detection in dermoscopy images
61. Wadhawan T, Situ N, Lancaster K, Yuan X, using gradient vector flow snakes. Skin Res
Zouridakis G (2011) SkinScan©: a portable Technol 11(1):17–26
library for melanoma detection on handheld 75. Kass M, Witkin A, Terzopoulos D (1988)
devices. In IEEE International Symposium Snakes: active contour models. Int J Comput
on biomedical imaging: from nano to macro, Vision 1(4):321–331
2011. IEEE, pp 133–136 76. Ridler TW, Calvard S (1978) Picture thresh-
62. Hu R, Queen CM, Zouridakis G (2013) olding using an iterative selection method.
Detection of Buruli ulcer disease: preliminary IEEE Trans Syst Man Cybern 8(8):630–632
results with dermoscopic images on smart 77. Qureshi S (2005) Embedded image process-
handheld devices. In Point-of-Care Healthcare ing on the TMS320C6000TM DSP: exam-
Technologies (PHT), 2013 IEEE. IEEE, ples in code composer studioTM and
pp 168–171 MATLAB. Springer, New York
 Skin Lesion Screening using Smartphones 495

78. Bezdek JC (1981) Pattern recognition with international joint conference on artificial
fuzzy objective function algorithms. Kluwer intelligence, vol 1999. pp 55–60
Academic Publishers, Norwell 92. Wasikowski M, Chen XW (2010) Combating
79. Dunn JC (1973) A fuzzy relative of the the small sample class imbalance problem
ISODATA process and its use in detecting using feature selection. IEEE Trans Knowl
compact well-separated clusters. J Cybern Data Eng 22(10):1388–1400
3(3):32–57 93. Forman G, Cohen I (2004) Learning from
80. Chan TF, Vese LA (2001) Active contours little: comparison of classifiers given little train-
without edges. IEEE Trans Image Process ing. In: Knowledge discovery in databases:
10(2):266–277 PKDD 2004. Springer, Berlin, pp 161–172
81. Chan TF, Sandberg BY, Vese LA (2000) 94. Cui YJ, Davis S, Cheng CK, Bai X (2004) A
Active contours without edges for vector-­ study of sample size with neural network. In
valued images. J Vis Commun Image Proceedings of 2004 international conference
Represent 11(2):130–141 on machine learning and cybernetics, 2004,
82. Ramlakhan K, Shang Y (2011) A mobile vol 6. IEEE, pp 3444–3448
automated skin lesion classification system. In 95. Rezze GG, Sá BCSD, Neves RI (2006)
23rd IEEE international conference on tools Dermoscopy: the pattern analysis. An Bras
with artificial intelligence (ICTAI), 2011. Dermatol 81(3):261–268
IEEE, pp 138–141 96. WHO, World Health Organization: Buruli ulcer
83. Shapiro LG, Stockman GC (2001) Computer disease, Fact sheet N 199, Revised Mar 2007
vision. Prentice Hall, Upper Saddle River, NJ 97. Boleira M, Lupi O, Lehman L, Asiedu KB,
84. Pehamberger H, Steiner A, Wolff K (1987) In Kiszewski AE (2010) Buruli ulcer. An Bras
vivo epiluminescence microscopy of pig- Dermatol 85(3):281–301
mented skin lesions. I. Pattern analysis of pig- 98. Debacker M, Aguiar J, Steunou C, Zinsou C,
mented skin lesions. J Am Acad Dermatol Meyers WM, Guédénon A et al (2004)
17(4):571–583 Mycobacterium ulcerans disease (Buruli
85. Situ N, Yuan X, Zouridakis G, Mullani N ulcer) in rural hospital, Southern Benin,
(2007) Automatic segmentation of skin 1997–2001. Emerg Infect Dis 10(8):1391
lesion images using evolutionary strategy. In 99. Drummond C, Butler JR (2004) Myco­
IEEE international conference on image bacterium ulcerans treatment costs, Australia.
processing, 2007 (ICIP 2007), vol 6. IEEE, Emerg Infect Dis 10(6):1038
pp VI–277 100. Kubat M, Holte R, Matwin S (1997)

86. Stollnitz EJ, DeRose AD, Salesin DH (1995) Learning when negative examples abound.
Wavelets for computer graphics: a primer. 1. In: Machine learning: ECML-97. Springer,
IEEE Comput Graph Appl 15(3):76–84 Berlin, pp 146–153
87. Marghoob AA, Swindle LD, Moricz CZ, 101. Leung T, Malik J (2001) Representing and
Sanchez Negron FA, Slue B, Halpern AC, recognizing the visual appearance of materials
Kopf AW (2003) Instruments and new using three-dimensional textons. Int J
technologies for the in vivo diagnosis of Comput Vision 43(1):29–44
melanoma. J Am Acad Dermatol 49(5): 102. Nowak E, Jurie F, Triggs B (2006) Sampling
777–797 strategies for bag-of-features image classifica-
88. Binder M, Schwarz M, Winkler A, Steiner A, tion. In: Computer vision–ECCV 2006.
Kaider A, Wolff K, Pehamberger H (1995) Springer, Berlin, pp 490–503
Epiluminescence microscopy: a useful tool for 103. Lowe DG (2004) Distinctive image features
the diagnosis of pigmented skin lesions for from scale-invariant keypoints. Int J Comput
formally trained dermatologists. Arch Vision 60(2):91–110
Dermatol 131(3):286 104. Parikh D, Zitnick CL, Chen T (2008)

89. Ojala T, Pietikainen M, Maenpaa T (2002) Determining patch saliency using low-level
Multiresolution gray-scale and rotation context. In: Computer vision–ECCV 2008.
invariant texture classification with local Springer, Berlin, pp 446–459
binary patterns. IEEE Trans Pattern Anal 105. Chen XW, Gerlach B, Casasent D (2005)
Mach Intell 24(7):971–987 Pruning support vectors for imbalanced
90. Van De Sande KE, Gevers T, Snoek CG data classification. In Proceedings. 2005
(2010) Evaluating color descriptors for object IEEE international joint conference on neural
and scene recognition. IEEE Trans Pattern networks, 2005 (IJCNN'05), vol 3. IEEE,
Anal Mach Intell 32(9):1582–1596 pp 1883–1888
91. Veropoulos K, Campbell C, Cristianini N 106. Kubat M, Matwin S (1997) Addressing the
(1999) Controlling the sensitivity of sup- curse of imbalanced training sets: one-sided
port vector machines. In Proceedings of the selection. In: Proceedings of the fourteenth
496 George Zouridakis et al.

international conference on machine learn- 112. Kira K, Rendell LA (1992) The feature selec-
ing, vol 97. Morgan Kaufmann, San Mateo, tion problem: traditional methods and a new
CA, pp 179–186 algorithm. In: Proceedings of the tenth
107.
Chawla NV, Bowyer KW, Hall LO, national conference on artificial intelligence.
Kegelmeyer WP (2011) SMOTE: synthetic AAAI Press, Menlo Park, CA, pp 129–134
minority over-sampling technique. arXiv pre-
113. Wolf JA, Moreau JF, Akilov O, Patton T,
print arXiv:1106.1813 English JC, Ho J, Ferris LK (2013) Diagnostic
108. Estabrooks A, Jo T, Japkowicz N (2004) A inaccuracy of Smartphone applications for mel-
multiple resampling method for learning anoma detection. JAMA Dermatol 149(4):
from imbalanced data sets. Comput Intell 422–426
20(1):18–36 114. Chao JT, Loescher LJ, Soyer HP, Curiel-­

109. Schölkopf B, Platt JC, Shawe-Taylor J, Smola Lewandrowski C (2013) Barriers to mobile
AJ, Williamson RC (2001) Estimating the teledermoscopy in primary care. J Am Acad
support of a high-dimensional distribution. Dermatol 69(5):821–824
Neural Comput 13(7):1443–1471 115.
Kroemer S, Frühauf J, Campbell TM,
110. Raskutti B, Kowalczyk A (2004) Extreme re-­ Massone C, Schwantzer G, Soyer HP,
balancing for SVMs: a case study. ACM Hofmann‐Wellenhof R (2011) Mobile tele-
SIGKDD Explor Newslett 6(1):60–69 dermatology for skin tumour screening: diag-
111. Manevitz LM, Yousef M (2002) One-class
nostic accuracy of clinical and dermoscopic
SVMs for document classification. J Mach image tele-evaluation using cellular phones.
Learn Res 2:139–154 Br J Dermatol 164(5):973–979
 INDEX

A Thomale Guide App ..................................................411

“Voice Recorder”......................................... 329, 331, 333
Accelerometer (ACT) data ...................... 299, 392, 394–396, Aptamer ........................................................... 100, 101, 106
400, 403 Array detector ...................................................................261
ACT data. See Accelerometer (ACT) data Assay plate fabrication ......................................................234
Active Contours ...............................................................473 Assessment, tremor...........................................................364
Amplification Attention
air bubbles.....................................................................34 EEG bands .................................................................378
detect fluorescent signals ........................................15–16 eSense™ measures ......................................................382
enzymatic......................................................................21 eSense meter data .......................................................384
fluorescence detection ...................................................26 frontal lobe .................................................................377
gel electrophoresis.........................................................33 intensity ......................................................................383
isothermal .....................................................................18 performance ................................................................377
membrane-bound nucleic acid......................................30 REM/non-REM sleep ...............................................384
microfluidic chip ...........................................................17 Audacity software .....................................................300, 301
“multifunctional”.....................................................18, 19 Audio processing
NATs ............................................................................16 accelerometers.............................................................299
negative samples/controls. ............................................36 amplification ...............................................................298
Android Audacity .............................................................300, 301
Architect .....................................................................449 AVC ...........................................................................298
and iOS devices ..................................................442, 449 clear sound ..................................................................299
mobile application ...................................... 452, 453, 465 conversation ................................................................299
Anti-BCG crowd sourcing ............................................ 294, 300–301
biotin ............................................................................64 heartbeat filtered.........................................................298
fluorescein.....................................................................64 heartbeat pure .............................................................298
Antibiotic resistance, M. tuberculosis ................................... 42 high pass filter ............................................................298
Apple App Store.......................................................408, 409 low pass filter ......................................................296, 298
Application, handheld devices spectrogram ................................................................300
Buruli ulcer diseas .......................................................487 Audio recording
skin cancer (see Skin cancer) iPhone ........................................................................328
Apps MP3 ...........................................................................331
API libraries ...............................................................294 quality .........................................................................334
Apple App Store.........................................................409 STL ............................................................................330
auscultation .........................................................296–300 Voice Recorder............................................ 329, 331, 333
better-quality audio recording ....................................334 Auto volume control (AVC)
Dropbox and cloud services ........................................334 clear sound ..................................................................299
iPhysioMeter ..............................................................309 description ..................................................................298
iStethoscope Pro .........................................................295 filtering and amplification ..........................................300
iTunes App Store by Apple ................................307, 309
launching ............................................................295–296 B
Mel-App.....................................................................444 Beer–Lambert law ....................................... 48, 54, 157–158,
modern cell phones .......................................................96 163, 180
OnlineDermClinic App .............................................444 Benzene, toluene, ethylbenzene and xylenes (BTEX)
smartphone .................................................................491 and MIP .....................................................................205
Store ................................................................... 294, 295 separation and detection, air samples ..................208, 209

Avraham Rasooly and Keith E. Herold (eds.), Mobile Health Technologies: Methods and Protocols, Methods in Molecular Biology,
vol. 1256, DOI 10.1007/978-1-4939-2172-0, © Springer Science+Business Media New York 2015

497
 MOBILE HEALTH TECHNOLOGIES
498 Index

Benzene, toluene, ethylbenzene and xylenes (BTEX) (cont.) Charge-coupled device (CCD)
and VOCs...................................................................203 detector ............................................... 248, 250, 255, 258
wearable detection system...........................................208 device .................................................................. 140, 144
Biotinylated anti-PfHRP2 IgG synthesis...........................78 Chemical sensors ..............................................................202
Blood analysis Circularity index (CIRC) .................................................439
cell-phone based imaging ...................................187–188 Circulating tumor cells (CTCs)
cost-effective translation .............................................172 auxiliary electronics.....................................................131
smart application ................................................181–182 auxiliary fluidics ..........................................................131
static microscopic imaging cancer..........................................................................124
hemoglobin concentration ............................186–188 challenges ...................................................................123
imaging process methods ..............................182–184 clinical applications.....................................................151
optical design ................................................177–180 detection .............................................................133–134
red blood cell density ....................................185–186 drug response ......................................................134–136
smart application ..........................................181–182 electronics ...........................................................128–129
white blood cell density ................................184–185 engineers .....................................................................123
Botulinum neurotoxin A (BoNT-A) ................................252 finite element simulations-fluidic .......................132–133
Buruli ulcer finite element simulations-magnetic...................131–132
automatic detection ............................................487, 488 giant magnetoresistance ..............................................126
skin cancer ..................................................................468 immunomagnetic detection ........................................124
magnetic labeling ........................................................131
C
magnetic nanomaterials ......................................129–130
Camera. See Mobile camera magnetic nanomaterials and micro-Hall detectors ...........126
Cancer detection....................................... 436, 465, 467, 468 μHall chip (see μHall chip)
Capillary μHall sensors .......................125, 126, 128, 129, 132, 133
3D detection systems ..................................................232 microfluidics ...............................................................125
fluorophore .................................................................237 mobile mobile flow cytometer ....................................140
imaging detector .........................................................237 superparamagnetic MNPs ..........................................126
LED ...........................................................................237 Clinical assays, artificial sera samples ...........................96–97
optical amplification ...................................................232 Cloud infrastructures ........................................ 436, 451, 457
Capillary-driven microfluidics ............................................86 CMOS camera. See Complementary metal oxide
Cardiovascular diseases (CVDs) .......................................293 semiconductor (CMOS) camera
CCD. See Charge-coupled device (CCD) Colorimetric method
Cell-phone based diagnostics paper-based mobile platform ..................................42–43
application ..................................................................174 tuberculosis diagnostics.................................................42
blood analysis (see Blood analysis) Colorimetric non-cross-linking assay
chemical reagents........................................................174 gold nanoprobe characterization .............................49–50
cost-effective translation .............................................172 MTBC DNA detection ....................................46, 50–52
flow-cytometry (see Cell-phone based Commercial off-the-shelf instrumentation.
flow-cytometry) See Spectrophotometry, CMOS cameras
imaging cytometer ......................................................189 Complementary metal oxide semiconductor (CMOS)
manual counting .........................................................172 camera.....................................................140, 144
micro-fluidic chip .......................................................172 “iOS Default” .............................................................310
optical components .....................................................174 iPhysioMeter ..............................................................316
opto-fluidic design ......................................................188 LED flash ...................................................................320
opto-fluidic pumping/excitation scheme ....................173 lnNPV values ..............................................................313
PDMS based micro-fluidic chip .................................188 PPG measurement..............................................306, 309
ratio of f/f 2.................................................................188 Computational image enhancement .........................239, 240
telemedicine devices ...................................................172 Computer aided drawing (CAD) .....................................250
Cell-phone based flow-cytometry Computer control .............................................................233
nHD mode .................................................................175 Connected component algorithm .....................................478
optical design ..............................................................175 CTCs. See Circulating tumor cells (CTCs)
and optical microscopy ...............................................172 Cyan magenta yellow key (CMYK) .................................261
SYTO16 solution .......................................................174 Cytometry. See also Mobile flow cytometer
WBC density ......................................................176–177 imaging flow ...............................................................145
Cell phone camera ..............................................................94 POC clinical flow .......................................................151
 MOBILE HEALTH TECHNOLOGIES
Index
499

D EEG recording
eSenses, attention and meditation ......................383–384
Deoxyribonucleic acid (DNA) modes .................................................................382–383
amplification components.......................................45, 49 noise frequencies .........................................................378
and control samples ................................................45, 46 physiological and behavioral .......................................381
fluorescent dyes.............................................................35 poor signal ..................................................................382
isolation Electric field
components.......................................................44–45 and flow circulation ......................................................58
extraction with QIAamp DNA mini kit ...........48–49 with and without vibration ...........................................66
sample decontamination .........................................48 Electrocardiograph (ECG)
LAMP ..........................................................................18 ACT ................................................................... 396, 400
Loopamp™ DNA amplification kit .............................21 HRV/PRV ..................................................................322
on-chip isolation ...........................................................25 and iPhysioMeter .......................................................323
Depolymerization .............................................................215 laboratory NIR PPG ..................................................322
Developing world and PPG instruments .................................................321
HRME system ...........................................................422 and QRS complex detection .......................................394
oral and cervical cancer (see Oral and cervical cancer) Electrochemiluminescence (ECL). See Paper microfluidics
Diagnostics. See also Improve lateral-flow immunoassay Electroencephalogram (EEG)
(LFIA) diagnostics ADHD ...............................................................378, 387
health care services .........................................................3 adverse symptoms .......................................................378
plastic microfluidic cassette ............................................6 Ag/AgCl electrodes ....................................................377
point-of-care........................................... 71, 72, 119, 121 brain electrical activity ................................................375
polycarbonate (PC) tubing .............................................6 computerized test .......................................................387
Diaphragm-based Fabry–Perot interferometric disruption, visual processing .......................................378
(DFPI) ....................................................160–162 frontal bands and accommodative lag .........................387
Diblock copolymer synthesis ..............................................76 frontal lobe .................................................................377
Dielectrophoresis (DEP) ..............................................58, 59 Mindset/ThinkGear (see MindSet)
Diffuse reflectance spectroscopy (DRS) recording.............................................................382–384
calibration process ......................................................158 SPSS syntax (see SPSS syntax)
self-calibration channels .....................................160, 163 types of waves .....................................................375–376
unreliable and defective ..............................................168 Electrohydrodynamic (EHD) flow ...............................59, 60
visible wavelength range (VIS-DRS) .................156, 158 Environmental monitoring .......................................201, 207
Digital auscultation Enzymatic
amplification ...............................................................298 amplification .................................................................17
audio playback ....................................................300, 301 cellulose-based FTA membrane ...................................30
AVC ...........................................................................298 detect fluorescent signals ........................................15–16
high pass filter ............................................................298 molecular assays ............................................................16
internal processing ..............................................296, 297 Enzyme-linked immunosorbent assay (ELISA).................58
listening modes ...................................................298–299 capture ..........................................................................93
low pass filter ......................................................296, 298 indirect..........................................................................93
optimal positions, body ...............................................299 μTADs ....................................................................86–88
DMP-pNIPAAm synthesis..........................................77–78 Essential tremor................................................ 359, 362, 371
Dorsolateral prefrontal cortex (DLPFC) ..................377, 378
3D printing technologies, stethoscope F
audio processing and analysis..............................331–332
Fabrication
iPhone (see iPhone)
μTADs ....................................................................89–90
shapeways ...................................................................330
printed microplates .................................................88–89
test and record ............................................................331
Fabrication methods .........................................................234
Dream diary ..................................................... 392, 400–402
FCM. See Fuzzy C Means (FCM)
DRS. See Diffuse reflectance spectroscopy (DRS)
FDM. See Fused deposition modeling (FDM)
Dysplastic nevi .................................................................437
Fiber-optic sensor
E mobile .................................................................160–162
pressure test ................................................................164
ECG. See Electrocardiograph (ECG) smartphone based spectrometers ................................168
EEG. See Electroencephalogram (EEG) Field of view (FOV) .........................................................146
 MOBILE HEALTH TECHNOLOGIES
500 Index

Flow-cytometry colorimetric non-cross-linking assay

opto-fluidic fluorescent...............................................175 (see Colorimetric non-cross-linking assay)
Fluidics system data analysis ............................................................52–53
flow cell ......................................................................141 Lambert–Beer law ........................................................54
flow cytometers...........................................................139 LSPR ............................................................................42
optofluidic fluorescent imaging cytometry .................140 MgCl2 concentrations...................................................54
Fluorescein morphological characterization.....................................54
CCD detector .....................................................255–256 mutation analysis ..........................................................55
imageJ .........................................................................255 nanodiagnostics ......................................................41–42
LED illuminator .........................................................255 nanoprobe functionalization .........................................54
metal mold master/injection molding .........................256 non-cross-linking assay components ............................45
mixing and detection ..................................................255 paper platform (see Paper platform)
syringe ........................................................................255 PCR product quantification .........................................54
three-layer design, ADU.............................................256 printing technology ......................................................55
Fluorescein isothiocyanate (FITC)...............................62, 64 RGB analysis ................................................................55
Fluorescence detection ........... 139, 144. See also Smartphone synthesis ...........................................................44, 47–48
Fluorescence detector .......................................................255 thiol-modified ssDNA .................................................53
Fluorescent microscopy tuberculosis .....................................................................2
opto-fluidic pumping/excitation scheme ....................173 Gold-streptavidin conjugates
real-time .....................................................................172 diblock copolymer synthesis .........................................77
Fused deposition modeling (FDM) ..................................333 mCTA synthesis ...........................................................76
Fuzzy C Means (FCM) .................................... 473, 475, 476 temperature-responsive gold colloids ......................76–77
Gray Level Co-occurrence Matrix (GLCM) ...................441
G GSM/satellite communication .......................................4, 11
Gait analysis
H
advanced algorithms ...................................................339
anatomical mounting positions ...................................340 Haar wavelet .............................................................480–481
biomedical applications ......................................336–337 Handheld devices
characteristics .............................................................344 application (see Application, handheld devices)
experimental protocol .........................................344–345 cart and battery-operated ...........................................460
GaitShoe ....................................................................338 DermLite II multispectral device ...............................460
healthy and hemiplegic ...............................................338 dermoscopy .........................................................460, 490
Internet .......................................................................336 diagnostic criteria ...............................................461–462
monitoring activity .....................................................339 digital communications...............................................460
parameters ..................................................................345 digital dermoscopy systems......................... 459–460, 464
Parkinson’s disease (see Parkinson’s disease) facilitate development on mobile device .............487–489
patient care .................................................................337 general practitioner .....................................................460
post-processing ...........................................................340 hardware implementation ...................................465–467
progressive evolution ..................................................345 image analysis .....................................................462–463
real-time biofeedback .................................................338 image-based screening methods .................................461
remote distance ...........................................................346 imbalanced classification ............................................489
rheumatoid arthritis ....................................................340 lesion magnification....................................................459
sensors, microelectronics and telecommunications .....337 lesion sampling strategies ...................................489, 490
tendon reflex response (see Patellar tendon reflex) MelaFind ....................................................................460
time-averaged acceleration ......................... 338, 346–347 melanoma detection....................................................460
triaxial accelerometer ..................................................338 methods
virtual proprioception .........................................339, 346 boundary smoothing .....................................477–478
WBAN .......................................................................337 classification..................................................483–484
GLCM. See Gray Level Co-occurrence Matrix (GLCM) connected component labeling .....................476–477
Global health ...................................................... 72, 151, 248 Haar wavelet .................................................480–481
Gold (Au) nanoparticles image segmentation ......................................472–475
biomolecular moieties, bio-recognition.........................42 imbalance classification.........................................484
centralized off-site laboratory .................................43–44 ‘jpeg’ format image........................................470–471
colorimetric method .....................................................42 lesion color analysis.......................................481–483
 MOBILE HEALTH TECHNOLOGIES
Index
501
lesion geometry analysis........................................479 HIV-1 and HIV-2 ..........................................................6
lesion sampling strategies..............................478–479 LAMP testing ..............................................................26
lesion texture analysis....................................479–480 in saliva ...................................................................19–20
local binary pattern ...............................................481 tuberculosis ...................................................................57
postprocessing.......................................................476 viral RNA targets..........................................................25
preprocessing ................................................471–472 Hybrid microfluidic ..........................................................125
procedure ......................................................468–470
system ...................................................................468 I
mobile operating system .....................................489–490 Image analysis
noninvasive diagnostic procedures ..............................459 approaches ..................................................................463
nonuniform sampling .................................................489 Buruli ulcer disease .....................................................487
random sampling ........................................................489 feature extraction ........................................................463
saliency sampling ........................................................489 lesion classification .....................................................463
skin cancer risk ...........................................................461 melanoma detection systems.......................................462
smartphone apps .................................................490, 491 segmentation ......................................................462–463
smartphone-based systems ......................... 464–465, 491 skin cancer detection system .......................................468
software implementation ....................................467–468 supervised classification ..............................................463
special light sources ....................................................459 Image capturing
square grid sampling ...................................................489 data stream .................................................................239
Heart rate (HR) JPEG ..........................................................................238
“Graph” windows ........................................................319 luminance ...........................................................238, 239
HRV/PRV ..................................................................322 mobile phone ..............................................................238
iPhysioMeter ......................................................320, 322 video processing amplifier...........................................238
and LF/HF ratio ........................................................398 Webcam ..............................................................238–239
and NPV............................................. 306–307, 312, 313 ImageJ analysis
PPG measurement......................................................317 color separation...........................................................147
pulse volume ...............................................................305 data processing............................................................263
Scatter and Bland–Altman plots.................................321 3D image ....................................................................240
variability ............................................................322, 400 image visualization .....................................................257
HE4 biomarker MATLAB® ...................................................................28
micro-plate ELISA.....................................................116 orthographic projection optics ....................................236
unknown concentrations .....................................117–118 RGB components .......................................................243
in urine ...............................................................112, 119 video files ....................................................................141
High resolution imaging...................................................423 Imager qualification
High-resolution microendoscope (HRME) camera gain setting .....................................................242
cervical cancer .............................................................421 capture application..............................................242–243
cervical imaging procedure .................................429–430 infrared .......................................................................242
contrast agents ............................................................434 pixels ...........................................................................242
debris on the fiber ...............................................433–434 sensitive fluorescent measurements.............................243
dichroic mirror alignment ...........................................433 Imaging. See also Blood analysis
epithelial cells .............................................................422 cell-phone based .........................................................177
illumination ................................................................424 dark-field background ................................................172
imaging components...................................................425 imageJ analysis (see ImageJ analysis)
imaging system ...........................................................423 Immunoassay
LEEP..........................................................................422 BGM-based................................................ 101, 102, 106
in mobile health ..................................................430–433 lateral flow (LF) strip ...................................................16
optical filters .......................................................423–424 LFIA (see Improve lateral-flow immunoassay
optical lenses and mirrors ...........................................423 (LFIA) diagnostics)
optics assembly ...................................................425–429 microfluidic
optomechanics ............................................................424 assay operation ..........................................................8
Papanicolaou...............................................................421 COC cassette............................................................7
proflavine solution ..............................................422, 425 materials ...................................................................6
HR. See Heart rate (HR) reagent loading .....................................................7–8
Human immunodeficiency virus (HIV) molecular assays ............................................................16
blood plasma samples ...................................................19 in vitro bioanalytical methods.......................................16
 MOBILE HEALTH TECHNOLOGIES
502 Index

Immunosensor microtip ...............................................61, 67 sampling rate ......................................................320–321

Improve lateral-flow immunoassay (LFIA) diagnostics use.......................................................................313–316
antibody molar concentration .......................................82 iPod
binary reagent system ...................................................72 Internet connectivity...........................................336, 343
biomarker enrichment process ......................................75 and iPads ....................................................................295
biotinylated anti-PfHRP2 IgG.....................................78 iPhone ................................................ 316, 317, 324, 343
carboxylates...................................................................82 patellar tendon reflex ..................................................341
clinical laboratory assays ...............................................71 smartphones and portable media devices ....................339
extinction coefficient ....................................................82 wireless accelerometer reflex quantification ........347–349
gold-antibody conjugates ..............................................73 ISODATA algorithm ...............................................473, 474
gold-streptavidin conjugates (see Gold-streptavidin Isothermal
conjugates) enzymatic amplification................................................27
1
H NMR results ...........................................................81 on-site applications .......................................................18
iron(0) pentacarbonyl ...................................................82 proteinase K and lysing agents......................................35
lateral flow device modification ..............................78–79 iStethoscope Pro
magnetic affinity reagents .............................................72 Apple iOS devices ..............................................294, 295
materials .................................................................74–75 App Store ...................................................................295
mobile devices...............................................................72 auscultation (see Digital auscultation)
NIPAAm recrystallization ............................................81 awareness ....................................................................302
PfHRP2 enrichment ..............................................79–81 crowd sourcing data ............................................300–301
pNIPAAm magnetic nanoparticles ..............................72 CVDs .........................................................................293
point-of-care diagnostic test .........................................72 description ..................................................................293
temperature-responsive magnetic nanoparticles development ...............................................................294
(see Magnetic nanoparticles) internal microphone, iPhones .....................................295
Infectious diseases ..............................................................30 iPhone app..........................................................302–303
iOS devices ................................436, 442, 461, 465, 468, 488 launching ............................................................295–296
iPhone limitations ...................................................................302
3D printing (see 3D printing technologies, stethoscope) “Medical Apps”...........................................................294
gait acceleration waveform..................................344, 345 software samples .........................................................294
gait parameters ...................................................345, 346 iTunes App Store by Apple ......................................307, 309
Internet connectivity...................................................336
Parkinson’s disease tremor ..................................350–352 L
test and evaluation ......................................................341 Lab-on-a-chip (LOC)
iPhone app ...............................................................302–303 alignment ....................................................................257
iPhysioMeter bonding.......................................................................257
after measurement ..............................................318–320 capillary force valves ...................................................248
calculation, HR and NPV...................................310–313 controlling noise .........................................................258
CMOS camera ...................................................309–310 data analysis ................................................................249
description ..........................................................306–307 design and fabrication
and electrocardiograph .......................................321, 323 BoNT-A ...............................................................252
features ...............................................................307–308 chamber ................................................................252
finger ..........................................................................310 PMMA .................................................................253
heart rate (see Heart rate (HR)) V-junction design .................................................252
HRV/PRV ..................................................................322 engraving ....................................................................257
indices iPR and iCPV.................................................322 equipment ...................................................................249
iOS 7 and iPhone 5s ...................................................324 fabrication approach ...................................................248
materials .....................................................................309 image visualization .....................................................257
measurement.......................................................316–317 imaging conditions .....................................................257
during measurement ...........................................317–318 introduction ........................................................247–248
oxygen saturation ........................................................322 LOM approach (see Laminated object
photoplethysmography ...............................................307 manufacturing (LOM) approach)
physiological variables.................................................305 mHealth technology ...................................................247
PPG measurements ....................................................309 microfluidics designs...................................................248
PPG use......................................................................321 micromachining approaches .......................................248
 MOBILE HEALTH TECHNOLOGIES
Index
503
optical filtering ...........................................................257 screening systems ........................................................464
optical signal ...............................................................258 spectrophotometric intracutaneous analysis ................464
reagents.......................................................................248 symptoms....................................................................442
sample size ..........................................................257–258 TMS320DM642-based automated ............................466
thermoplastic polycarbonate film................................249 Visual Risk Calculator ................................................465
valves (see Passive capillary valves) Melanoma screening.........................................................464
Labsel-free detection ........................................................192 Mel-App ..........................................................................444
Laminated object manufacturing (LOM) approach Menzies method .......................................................440, 462
assembly......................................................................252 μHall chip
bonding.......................................................................251 clinical use ..................................................................133
CO2 laser ...................................................................250 CTC counts ................................................................134
3D device....................................................................250 fine needle aspirates ....................................................136
laser cutting ................................................................250 μΗD sensor.................................................................130
LAMP. See Loop-mediated amplification (LAMP) tumor detection ..........................................................135
Laser μHall sensors .............................125, 126, 128, 129, 132, 133
engraving ....................................................................250 mHealth
excitation ....................................................................143 applications ...........................................................72, 142
Lesion color analysis.................................................481–483 CMOS/CCD cameras ...............................................140
Lesion geometry analysis ..................................................479 flow cytometry ............................................................139
Lesion texture analysis..............................................479–480 smartphone (see Smartphone)
Light-emitting diode (LED) ...............61, 234, 240, 249, 255 Microchip ELISA
Localized surface plasmon resonance (LSPR) ....................42 anti-HE4-rabbit primary antibody .............................115
Loop-mediated amplification (LAMP) anti-rabbit-HRP.........................................................115
isothermal amplification ...............................................18 cell phone imaging, black box .....................................114
real-time fluorescence emission intensity......................33 challenges ...................................................................112
thermocouple junction ..................................................23 design and fabrication...........................................113–14
virus concentration .......................................................26 Eppendorf tube...........................................................120
Low-cost sensing. See Paper microfluidics homogeneous color distribution .................................120
Low-resource settings.......................................................422 materials .....................................................................113
Luminescence spectrometry. See Spectrophotometry, microchannels .............................................................115
CMOS cameras micro-plate ELISA.....................................................115
ovarian cancer patients................................................111
M POC testing ...............................................................112
Magnetic nanomaterials ................................... 126, 129–130 pure HE4 peptide antigen ..........................................114
Magnetic nanoparticles quantitative image processing by Matlab ............116–117
DMP-pNIPAAm .........................................................77 quantitative image processing by mobile
synthesis .......................................................................78 application ..............................................118–119
mCTA synthesis .................................................................76 serum CA125 and HE4 .............................................111
Median power frequency (MPF) ......................................365 spectrophotometer ......................................................112
Meditation eSense ....................................................383, 384 unknown concentrations, HE4 ...........................117–118
Melafind ...........................................................................464 urine sample ...............................................................120
Melanoma Microelectronic chips ...............................................125, 337
benign nevus ...............................................................451 Microfluidic devices
cutaneous ....................................................................436 Adober Acrobatr .........................................................219
details..........................................................................448 biocompatibility ..........................................................125
detections schemes......................................................462 capillary micropipettes ................................................217
detect signs .................................................................446 CCD camera ..............................................................196
vs. dysplastic and non-dysplastic ................................454 CleWin Layout Editor and Adober Illustratorr .........217
early diagnosis ............................................................436 cofactor and immobilized enzymes.............................218
epidermis ....................................................................437 cross section ................................................................132
handheld devices (see Handheld devices) fabricating three-dimensional paper-based .................218
iABCD rule ................................................................465 fabrication...........................................................197–198
melanocytes on skin ....................................................438 food coloring, patterned paper ............................219, 220
non-melanoma skin cancers........................................435 hydrophobic oligomer solutions..........................217, 221
 MOBILE HEALTH TECHNOLOGIES
504 Index

Microfluidic devices (cont.) Microtip

hydrophobic regions ...........................................218–219 concentrator assembly...................................................63
immobilized enzymes .........................................215, 220 EHD flow.....................................................................60
label-free optical detection method ............................192 fabrication...............................................................62, 63
LOC platform ............................................................256 functionalization .....................................................64–65
mobile device (see Mobile device) immunofluorescence sensor ....................................58, 59
NATs (see Nucleic acid-based tests (NATs)) immunosensor system ...................................................61
optical-coding system .................................................196 mHealth system ............................................................61
PDMS based ..............................................................128 microfabrication process ...............................................62
point-of-care applications ...........................................192 MindSet
polycarbonate layer .....................................................251 analogue data ..............................................................378
polymers and adhesives...............................................248 description ..........................................................378, 379
printed paper ..............................................................219 EEG signals................................................................380
reagents and function..................................................220 electrical potential.......................................................378
scattering signal waveforms ........................................194 electrode locations ..............................................378, 379
silicon PIN photodetector ..................................196, 197 good quality signal ......................................................388
spatial mask pattern ............................................196–197 interference, signal ......................................................382
synthetic food dyes .............................................217–218 NeuroSkyLab MATLAB module ..............................381
three-dimensional devices NeuroView software ...................................................381
Adober Illustratorr ........................................222, 224 packets ........................................................................380
application, spray adhesive ............................221, 223 poor signal quality ..............................................381–382
assembled and laminated devices ..................223, 224 power density distribution ..........................................380
CleWin Layout Editor .........................................222 severity........................................................................382
Epilog Engraver laser cutter .........................222, 223 MindSet Research Tools (MRT). See MindSet
glass sheet, tweezers ......................................222, 223 Mobile application
plastic laminate .............................................223, 224 AES symmetrical encryption ..............................450–451
scissors ..........................................................223, 224 Android ......................................................................453
wax printer and black wax ..........................................217 contextual information................................................450
Whatman chromatography papaer .............................217 feature extraction and classification ....................455, 456
Microfluidic toner-based analytical devices (μTADs) image processing and data report................................118
advantages.....................................................................85 MATLAB ..................................................................119
capture ELISA .............................................................93 Mean Opinion Score ..................................................456
clinical applications.......................................................87 Mel-App.....................................................................444
clinical diagnostics ........................................................88 quantitative image processing .............................118–119
colorimetric detection ...................................................87 skin lesion region ........................................................450
delimit microzones .......................................................86 tremor assessment .......................................................361
detection system ...........................................................94 yield R pixel values .....................................................118
direct-printing technology ............................................86 Mobile camera
electrospray ...................................................................86 CMOS .......................................................................232
ELISA ....................................................................86–88 factors .........................................................................239
fabrication (see Fabrication) image analysis .....................................................243–244
flow rate ........................................................................10 ImageJ ........................................................................243
indirect ELISA .............................................................93 photomultiplier-based detector...................................239
iron and silicon oxide ....................................................86 sensor ..........................................................................242
miniaturized devices .....................................................86 Mobile computing ....................................................461, 490
point-of-care diagnostics (see Point-of-care Mobile dermoscopy .......................................... 464–465, 491
(POC) diagnosis) Mobile device
simultaneous colorimetric tests ...............................93–94 assay operation ................................................................8
Micro-Hall detector (μHD) 8-bit microcontroller ......................................................7
cancer patients ............................................................134 COC cassette..................................................................7
fabrication method .....................................................127 compact satellite modem ................................................7
magnetic moment .......................................................129 data acquisition and analysis .......................................8–9
MNPs .........................................................................125 data confidentiality, transmission ............................11–12
mobile diagnostics in mind .................................128–129 dilution .........................................................................12
Micromachining ...............................................................234 GSM/GPRS module ......................................................7
 MOBILE HEALTH TECHNOLOGIES
Index
505
handheld device ..............................................................4 aptamers .....................................................................100
hardware and software set-up ...................................9–10 biomarkers ..................................................................100
health care services .........................................................3 body fluids ..................................................................100
immunoassays ...............................................................12 cartridge......................................................................102
mChip device..............................................................4–5 iBG STAR® and Telcare® blood glucose
mChip platform............................................................12 monitoring systems .........................................102
messaging protocol .................................................10–11 IFN-γ .......................................................................... 100
microcontroller .............................................................12 immunoassay ..............................................................101
microfluidic cassettes ..................................................5–6 non-glucose biomarkers ..............................................100
microfluidic immunoassay ..........................................6–7 person’s physiological state............................................99
mobile microfluidic chip .................................................4 Mobile phone-based sensing. See Paper microfluidics
photodiodes ....................................................................7 Molecular diagnostics ....................................... 18–19, 33, 34
public health and patient care .........................................3 Molecularly imprinted polymer (MIP) ............ 202, 205, 210
reagent loading ...........................................................7–8 Monitoring, tremors
real-time observation and measurement .......................13 health, individuals .......................................................360
resource-limited settings .................................................3 patient populations .....................................................359
satellite-based communication........................................4 MPF. See Median power frequency (MPF)
satellite modem and GSM/GPRS module .....................6 Multiple sclerosis ..............................................................359
short-burst data (SBD) service .......................................4 Mycobacterium tuberculosis (MTB)
signal detection module ..............................................5–6 BCG suspension .....................................................63–64
super-bright red light-emitting diodes ...........................7 detection .......................................................................59
Mobile diagnostics. See Mobile device and non-MTBC Mycobacteria cultures .......................49
Mobile flow cytometer pulmonary specimens ...................................................58
adhesive layer ..............................................................152 single nucleotide sequence ............................................42
application ..................................................................139 sputum samples ......................................................66, 67
autofluorescent emission rate ......................................152 thiol-modified ssDNAs ................................................45
background reduction .........................................147–148 Mycobacterium tuberculosis complex (MTBC)
cell culture DNA detection .................................................46, 50–52
and fluorescence staining ......................................141 non-MTBC Mycobacteria cultures ..............................49
staining and dilution .....................................144–145 rpoB gene .......................................................... 42, 45, 46
cell fabrication ............................................................143
cell streaking .......................................................148–149 N
CMOS and CCD.......................................................140 Nanoparticles
color separation...........................................................147 DMP-pNIPAAm .........................................................78
computer control and data analysis .............................141 magnetic microbeads ....................................................72
CTC detection ...................................................140–141 magnetic nanoparticles .................................................78
current flow cytometers ..............................................140 pNIPAAm ..............................................................72, 73
device dimensions and imaging Nanoprobes, Au
conditions ...............................................145–146 characterization ......................................................49–50
fast focal ratio .............................................................151 colorimetric changes .....................................................42
flow cell and fluidic system .........................................141 functionalization ...........................................................54
fluidic system ..............................................................139 MTBC .............................................................46, 50–52
high pressure ports......................................................152 synthesis components .......................................44, 47–48
imager qualification ....................................................152 Near High Definition (nHD) mode.................................175
imaging flow cytometry ..............................................145 Neighboring gray-level dependence
mHealth applications .................................................142 matrix (NGLDM) ..........................................441
narrow-field detection ................................................139 Neuronavigation .......................................................406, 407
optical components .....................................................141 NeuroSky. See MindSet
POC analytical tools...................................................140 Neurosurgery. See Ventricular catheter, neurosurgery
rare cell counting ................................................150–151 NeuroView
rare cell detection ........................................................147 attention eSense meter ...............................................383
THP-1 human monocytes..................................140–141 Bluetooth connection .................................................381
webcam-based cytometer platform (see Webcam) data types ....................................................................384
Mobile health (mHealth) technologies live recording window .........................................382, 383
applications .................................................................100 MRT ..........................................................................381
 MOBILE HEALTH TECHNOLOGIES
506 Index

Neutrophils on-site detection ...........................................................15

commercial flow cytometer .................................199, 200 operation.................................................................30, 31
digital signal processing ..............................................199 palm-sized cassette mates .............................................17
human whole blood ............................................197, 198 PCR technology ...........................................................18
label-free optical-coding technique ............................195 plasma-derived lysate....................................................20
lymphocytes and monocytes .......................................193 polymer chip materials stock ........................................20
optofluidic device vs.commercial flow cytometer........200 and protocol ............................................................19–20
velocity, cell .................................................................199 reagents
white blood cells .........................................................199 isothermal amplification ...................................25, 26
NGLDM. See Neighboring gray-level dependence on-chip isolation, DNA ..........................................25
matrix (NGLDM) on-chip isolation, RNA ..........................................25
Normalized PV (NPV) real-time detection
CMOS camera ...........................................................313 amplicons ..........................................................27–28
“Graph” windows ........................................................319 fluorescence signal ..................................................28
HR (see Heart rate (HR)) isothermal amplification .........................................29
iPhone ........................................................................316 LEDs ......................................................................28
iPhysioMeter ..............................................................308 portable fluorescence microscope ............................28
lnNPV ................................................................ 313, 315 Qiagen ESE reader .................................................28
peak and foot points, calculation ................................312 smart cell phone-based detection............................28
Nucleic acid-based tests (NATs) RNase inhibitor ............................................................21
air bubbles.....................................................................34 sealing tape ...................................................................20
analysis..........................................................................33 simple filtration separation device ..........................33–34
binding phases ........................................................20, 23 supporting instrumentation ..........................................17
blood sample molecular analysis .............................31–33 temperature...................................................................35
cell phone camera ...................................................15–16 template ........................................................................34
DC power supply..........................................................20 in vitro bioanalytical methods.......................................16
delamination/leaks ........................................................34 wax encapsulation .........................................................19
design and fabrication
autofluorescence......................................................23 O
chip fabrication and bonding methods ...................21 Object-space telecentric lens ............................................237
and end-point fluorescent detection .................23, 24 OnlineDermClinic ...................................................444–446
layers .......................................................................23 On-site ........................................................15, 18, 29, 33, 36
microcluidic ......................................................21, 22 Optical-coding technique .........................................192, 195
multifunctional amplification reactors ..............19, 21 Optical system
on-chip LAMP amplification...........................23, 24 excitation and emission filters .....................................237
temperatures ...........................................................23 light distribution .........................................................237
thermocouple junction ............................................23 orthographic projection ......................................236, 237
thermoplastic polymers .....................................21, 22 Optofluidic device
enzymatic amplification kits ...................................16, 21 equilibrium position............................................193–194
filtration module .....................................................29–30 human blood cells .......................................................191
fluorescence signals .......................................................16 hydrodynamic inertial effect .......................................192
fluorescent dyes.............................................................21 label-free cell detection methods ................................192
fluorescent reporters ...............................................26–27 label-free optical-coding technique ............................195
Harris Unicore™ ..........................................................20 materials
inhibitors ......................................................................35 polydimethylsiloxane kit .......................................195
isolation, nucleic acids ............................................20–21 sample solution .....................................................195
isothermal amplification ...............................................18 space-time coded optical scattering signal ............195
“lab-on-a-chip” technology ..........................................16 methods
limit of detection ....................................................35–36 fabrication, microfluidic device .....................197–198
lyophilized poly-A RNA ..............................................34 neutrophils ....................................................198–200
molecular analysis .........................................................17 setup, microfluidic device ..............................196–197
molecular assays ............................................................16 microfluidic devices ....................................................192
“multifunctional” amplification reactors..................18, 19 neutrophils ..................................................................193
negative samples/controls, amplification.......................36 optical coding method ................................................192
 MOBILE HEALTH TECHNOLOGIES
Index
507
parabolic velocity profile .............................................192 Samsung Galaxy S (i9000) .........................................280
photomultiplier tubes (PMTs) ....................................195 Samsung I8910 HD ...................................................280
point-of-care blood analysis .......................................192 sensor lamination ........................................................280
scattering signal waveforms ........................................194 Paper platform
softer cells ...........................................................193–194 components ..................................................................46
spatial modulation technique ......................................195 detection .......................................................................52
Oral and cervical cancer preparation....................................................................52
absorption and scattering coefficients .........................156 Parkinson’s disease
cases in poor countries ................................................155 acceleration waveform ................................................351
change with probe pressure.........................................158 adverse side effects ......................................................342
computer and software ...............................................159 cardinal sign................................................................359
cost of wireless technology..........................................168 description ..................................................................336
data analysis ........................................................162–163 drug therapy ...............................................................342
in developing countries ...............................................158 Essential tremor..........................................................371
DRS calibration process .............................................158 evaluation, tremor .......................................................362
emerging technologies ................................................168 evolution .............................................................342–343
fiber-optic attachment ................................................168 experimental protocol .................................................351
gag reflex in patient ....................................................156 inferential statistical analysis .......................................353
global mortality...................................................155–156 medical resources ................................................353–354
human hemoglobin powder ........................................167 post-processing ...........................................................350
light scattering ............................................................156 power distribution ......................................................365
light source .................................................................157 software application ....................................................343
mobile fiber-optic sensor system ........................160–162 software capabilities ....................................................354
optical systems and components .................................159 spectral analysis ..................................................351, 353
oral mucosa .................................................................165 surgical procedures......................................................342
photons .......................................................................157 test and evaluation ......................................................354
pressure sensor test .............................................163–164 time averaged acceleration ..................................352–353
probe pressure .............................................................167 Particle image velocimetry (PIV) .......................................60
smartphone .................................................................168 Passive capillary valves
software ......................................................................162 engraving ....................................................................254
spectrometers ..............................................................159 Hagen-Poiseuille’s equation........................................254
temperature fluctuations .............................................167 PMMA layer ..............................................................254
tissue morphological information .......................157–158 reagent ........................................................................254
tissue phantom test .............................................164–165 Patellar tendon reflex
VIS-DRS............................................................156, 158 anatomical mounting strategy.....................................347
working materials and solutions .................................160 automation software ...................................................349
capabilities ..................................................................349
P description ..................................................................336
Paper microfluidics. See also Quantitative POC assays evolutions....................................................................350
advantages...................................................................278 experimental protocol .................................................351
AKD-heptane hydrophobization................................280 iPod and iPhone .........................................................341
CAD drawing .....................................................280, 281 machine learning algorithms ......................................350
calibration curve .................................................285, 286 Matlab software program ...................................341, 349
chemicals and materials ......................................278–279 post-processing ...........................................................348
cut-and-seal, fabrication .....................................282–283 potential energy impact pendulum .....................340–341
cyclic voltammograms and emission profiles ......284, 285 wireless accelerometer nodes.......................................340
description ..................................................................277 Pathogen
detection strategies .....................................................278 drug-susceptible and drug-resistant ..............................17
inkjet printer ............................................... 278, 281–282 microfluidic cassette......................................................16
Java based android application ....................................280 on-site detection ...........................................................15
MEP instruments .......................................................279 point-of-care...............................................................188
optimisation experiments....................................279–280 viral/bacterial ................................................................20
perspex clamp .....................................................283–284 Personal exposure assessment ...........................................209
potentiostatic control and detection ...................286–288 PfHRP2 enrichment ....................................................79–81
 MOBILE HEALTH TECHNOLOGIES
508 Index

Photo ionization detection (PID).....................................202 glucose meter ..............................................................102

Photoplethysmography (PPG). See iPhysioMeter grade VII invertase .....................................................102
Physiological indices.........................................................305 grade X invertase ........................................................107
Point-of-care (POC) ........................................................140 and human serum .......................................................103
label-free cell detection methods ................................192 IFN-γ .......................................................................... 106
“lab-on-a-chip” technology ..........................................16 immobilization of invertase on magnetic
microfluidic devices ....................................................192 beads .......................................................104–105
optical-coding technique ............................................195 in industrial applications.............................................107
Point-of-care (POC) diagnosis.................................100, 213 oligonucleotides ..........................................................103
anti-BCG (see Anti-BCG) principles ....................................................................108
BCG and MTB (H37Ra).......................................63–64 streptavidin-coated magnetic beads ............................102
concentration and detection....................................65–67 sulfo-SMCC...............................................................107
diagnostic methods .......................................................58 thiol modified DNA to invertase ........................103–104
dielectrophoretic force ..................................................60 Pulse volume (PV)
EHD ......................................................................59, 60 cutaneous ....................................................................322
electrokinetic methods ..................................................58 heart rate (see Heart rate (HR))
electrophoresis ..............................................................58 logarithmically transformed normalized .............315, 321
ELISA ..........................................................................58 normalized .................................................. 308, 319, 322
LED .............................................................................61 PV. See Pulse volume (PV)
low-cost microtip sensor ...............................................61
Q
microtip ...........................................61 (see also Microtip)
μTADs Quantification
artificial sera samples ........................................96–97 gait (see Gait analysis)
dengue ..............................................................95–96 Parkinson’s disease .............................. 341–343, 350–354
MTB ............................................................................58 tendon reflex ....................................... 340–341, 347–350
particle image velocimetry ............................................60 Quantitative POC assays
PCR..............................................................................58 alkaline phosphatase, calibration curve .......................216
public healthcare sectors ...............................................57 aluminum foil .............................................................227
reagents...................................................................61–62 bead-bound glucose oxidase................................214, 215
smear microscopy....................................................57–58 biotin functionalized glucose oxidase..........................226
sputum samples ................................................60–61, 65 calibration curve .........................................................228
Point-of-need detection................................................42–43 CleWin Layout Editor ...............................................225
Poly(N-isopropylacrylamide) (pNIPAAm)...................72, 75 Drytacr JetMounterT JM26 laminator .......................228
Polymer ethyl actetate ...............................................................225
molecular weight ..........................................................76 glass capillary tubes.....................................................226
pNIPAAm ....................................................................72 glucose 6-phosphate ...................................................226
rubber septum ...............................................................77 hydrogen peroxide ..............................................214, 216
Polymerase chain reaction (PCR) .................................58, 60 hydrophilic channels ...........................................214, 215
Poly-methylmethacrylate (PMMA) ................. 250–251, 257 hydrophilic products ...................................................214
Portable instruments.........................................................260 hydrophobic oligomers .......................................226, 227
Portable media device hydrophobic regions ...................................................226
acceleration waveform ................................................336 left-hand channel................................................214, 215
description ..................................................................354 magnesium chloride....................................................226
materials .....................................................................343 mHealth .....................................................................217
Parkinson’s disease .............................. 341–343, 350–354 microfluidics (see Microfluidic devices)
software ......................................................................335 optimum concentrations .............................................225
tendon reflex ....................................... 340–341, 347–350 paper-based microfluidic devices ....................... 214, 215,
wireless accelerometer (see Gait analysis) 217, 225
Protein detection, BGM printer ................................................................. 227, 228
amicon-10K and 100K ...............................................102 properties, laser cutter .................................................228
aptamer/invertase coated magnetic beads ...................107 resource-limited environments ...........................213–214
buffers and solutions ...................................................103 right-hand channel .............................................214–215
DNA-modified invertase with magnetic beads ..........107 spray adhesive .............................................................228
glucose levels...............................................................107 Whatman Chromatography Paper No. 1 ...................225
 MOBILE HEALTH TECHNOLOGIES
Index
509

R melanoma (see Melanoma)

MelApp ......................................................................444
Recording, tremor. See Tremor measurement, smartphone mobile application ..............................................450–451
Reflex response. See Patellar tendon reflex mole detective .....................................................442–444
Ribonucleic acid (RNA) NGLDM....................................................................441
amplification reactions ..................................................21 normal skin lesions and components ..........................437
DNA and RiboGreen™ ...............................................35 OnlineDermClinic .............................................444–446
NA isolation .................................................................34 processing method ......................................................449
on-chip isolation ...........................................................25 severity........................................................................456
skin cancer (see Skin cancer)
S
SkinLesion detector............................................446–449
Selective laser sintering (SLS) ..........................................328 SpotMole ....................................................................446
Sensors system implementation .......................................452–454
array microfabricated Hall ..........................................125 system’s architecture....................................................450
CTCs..........................................................................125 texture analysis............................................................440
giant magnetoresistance ..............................................126 Sleep diary ........................................................ 392, 400–402
μHall chip...........................................................128, 132 Sleep parameters
pseudomorphic ...........................................................126 computation................................................................392
Shapeways software and algorithm
3D printing service .....................................................330 accurate computation ............................................398
materials .....................................................................333 ACT features ................................................395–396
SIAscope V.......................................................................464 after preprocessing ................................................395
Signal-to-noise ratio (SNR) .................................... 145, 146, binary classifier C .................................................398
238, 240 biomedical datasets ...............................................397
Single-electrode wireless EEG. See Electroencephalogram confusion matrix ...................................................399
(EEG) ECG and Accelerometer data...............................400
Skin cancer ECG and QRS complex .......................................394
assess classification .....................................................485 Hypnogram ..........................................................400
classification ...............................................................486 multimodal dataset ...............................................394
diagnosis .....................................................................491 Non-REM ....................................................397, 398
execution time ....................................................486–487 positive and negative correction factors ................399
handheld detector ...............................................465–467 PSG data ..............................................................400
iPhone ................................................................ 484, 485 rejected samples ....................................................399
malignant melanoma ..................................................485 REM.............................................................397, 398
non-melanoma ...........................................................435 RR and ACT data ........................................394–395
risk ..............................................................................461 RR features ...........................................................395
screening and early detection ..............................436, 465 supervised learning ...............................................397
skin lesion analysis ......................................................484 Support Vector Classifier ..............................397–398
tenfold cross-validation across ............................485, 486 Wakefulness ..................................................397, 398
Skin lesions image analysis SLS. See Selective laser sintering (SLS)
advantages...................................................................457 Smartphone application
Android and iOS devices ............................................449 angulation measurements ...........................................409
artificial illumination and lenses system, Apple App Store.................................................408, 409
smartphones....................................................457 DICOM standard ......................................................409
ASM ...........................................................................441 iphone/ipad.........................................................406, 411
asymmetry ..................................................................438 Thomale Guide App ..........................................411–412
border .................................................................438–439 Smartphones
Cloud part ..........................................................451–452 air bubbles...................................................................241
colors .................................................................. 439, 440 artificial illumination and lenses system .....................457
differential structures ..........................................440, 442 based dermoscopy systems ..................................464–465
doctor mole application ..............................................442 components ................................................................234
GLCM Mean .............................................................441 configuration ..............................................................234
GLCM Standard Deviation .......................................441 dermoscopes ...............................................................491
initial evaluation .................................................454–456 description .................................................. 231–232, 460
melanocytes ................................................................438 device accessories ........................................................168
 MOBILE HEALTH TECHNOLOGIES
510 Index

Smartphones (cont.) stray light ....................................................................273

fabrication...................................................................241 teaching, students ...............................................266–267
image stacking ............................................................232 toy instrument ............................................................267
iPhysioMeter (see iPhysioMeter) wavelength calibration ........................................269, 272
laser machining ...........................................................234 webcam and cellcam detection ...................................273
LOD...........................................................................232 SpotMole..........................................................................446
materials .............................................................233–234 SPSS syntax
optical noise ................................................................240 beta wave ....................................................................386
photoplethysmography ...............................................307 computerized test .......................................................387
polycarbonate bottom .................................................234 data management................................................380–381
and portable media device (see Portable media device) frontal EEG bands and accommodative lag ...............387
RGB color channels....................................................243 MindView software ....................................................381
ROI management tool........................................243–244 power spectrum file ............................ 384–386, 388–389
skin lesions image analysis (see Skin lesions image transformation, power distribution .............................386
analysis) Sputum
sleep monitoring system .............................................392 MTB from pulmonary specimens.................................58
spectrometers ..............................................................168 sample preparation........................................................65
stethoscope recording (see Stethoscope) Standalone ........................................................ 360, 367, 369
user interface (see User interface) Stereolithography (STL) ..................................................330
Smartphones in health care ..............................................491 Stethoscope. See also iStethoscope Pro
SolarScan ..........................................................................464 adult chest sounds.......................................................328
Sound assemble, attachment ..........................................329, 331
adult chest...................................................................328 audio post-production software ..................................329
heartbeats ...........................................................329, 332 cardiovascular/respiratory issues .........................327–328
isolation ......................................................................331 components ................................................................327
voice recorder ..............................................................329 3D printing (see 3D printing technologies, stethoscope)
Spectrophotometry, CMOS cameras laser-sintered nylon.............................................329, 330
assembled instrument .........................................263, 264 standard stethoscope tubing ...............................329, 331
barrel distortion ..........................................................269 voice recorder .............................................. 329, 331, 333
biological measurements .....................................263–264 Stimuli-responsive ..............................................................72
capillary electrophoresis ..............................................270 STL. See Stereolithography (STL)
CCD, wavelength .......................................................269 Support vector machine (SVM) ...............................483–484
computational capabilities ..........................................271 Surface chemistry .........................................................64, 65
data processing............................................................263
description ..........................................................259–260 T
detectors .............................................................261–263 Tachogram................................................................394, 396
digital data ..........................................................271–272 Teledermatology ...............................................................436
focusing optics ....................................................260–261 Telemedicine ..............................................................96, 172
gratings, prisms and filters ..........................................261 Telephone ................................................. 259, 263, 365–367
high resolution ............................................................265 Tendon reflex response. See Patellar tendon reflex
impingent light ...........................................................269 TFs. See Tuning forks (TFs)
infrared filter.......................................................267, 268 Thomale Guide App ........................................................411
layout, fiber optics and webcam detector ............269, 270 Time series analysis, tremor
light sources ................................................................260 amplitude ....................................................................365
noise ...........................................................................271 Blackberry® Storm™ 9530, vibration ..........................364
pathologies..................................................................266 characteristics .............................................................373
PhotoShop® ........................................................ 265, 267 harmonic index ...........................................................366
Public Lab community ...............................................266 MPF ...........................................................................365
RGB data ...........................................................267–268 power dispersion .........................................................366
scientific enterprise .....................................................267 power distribution ......................................................365
signal-to-noise ratio....................................................270 power spectrum ..........................................................365
silicon photodetectors .................................................259 regularity.....................................................................365
spectral resolution .......................................................267 triaxial accelerometer ..................................................365
 MOBILE HEALTH TECHNOLOGIES
Index
511
TMS320DM642-based automated online analyses ............................................................366
melanoma detection system ............................466 statistical approach .....................................................366
Tremor. See Parkinson’s disease Ventricular catheter, neurosurgery
Tremor measurement, smartphone application software ............................................411–412
accelerometer data ......................................................361 brain tissue ..................................................................416
alert and vibration .......................................................362 computer technology ..................................................406
amplitudes ..........................................................371, 373 CT/MRI images.................................................410, 415
assessment ..................................................................364 description ..................................................................416
Blackberry® Java SDK .................................................360 guiding instrument .............................................411, 415
Blackberry® Storm™ 9530 .......................... 360, 370, 372 image acquisition ........................................................410
computer and data acquisition card ....................370–371 image data transfer .............................................411, 412
data collection .............................................................360 information resources .........................................405–406
Eclipse ........................................................................370 iphone/ipad application ..............................................406
evaluation devices .......................................................360 measurements, iPhone app .................................412, 413
features ...............................................................372–373 mobile device ..............................................................411
hardware and software components ....................360–361 neuronavigation ..................................................406, 407
Java programming language........................................371 patient data .................................................................415
NetBeans IDE ............................................................370 surgical procedure ...............................................406–407
neurological disorders .................................................359 surgical steps .......................................................413–415
protocols .............................................................371–372 technical advances.......................................................416
reminder/alarm ...........................................................372 trajectory’s angulation, coronal plane ..........................407
sampling frequency .....................................................371 Volatile organic compounds (VOCs)
start/repeat/stop buttons .............................................372 alkyl, aromatic and chlorinated
tasks ............................................................................362 hydrocarbons...................................................201
time data .....................................................................361 and BTEX ..................................................................203
time series analysis ..............................................364–366 and MIP .....................................................................202
user interface (see User interface) temperature.................................................................208
validation (see Validation process, tremor) and TFs ......................................................................202
Tropical diseases .................................................................87
Tuberculosis (TB) W
diagnosis (see Point-of-care (POC) diagnosis) Waveguide capillary array .................................................237
gold nanoparticles (see Gold (Au) nanoparticles) WBC density
MTB ............................................................................58 cell-phone based flow-cytometry ........................176–177
multidrug-resistant .......................................................42 static microscopic imaging based
Tuning forks (TFs) blood analysis ..........................................184–185
chemical sensors and wearable sensing Wearable detection system
systems ............................................................202 BTEX .........................................................................203
and MIP .............................................................202, 205 colorimetric tubes .......................................................202
quartz crystal resonator ...............................................202 complex environmental samples
wireless communication ..............................................206 and preconcentration module .........................202
device operation
U
data collection and analysis ...................................208
User interface desorption .............................................................208
data analysis ................................................................363 preconcentration ...........................................207–208
design, procedure ................................................361–362 sample injection ....................................................208
flowchart .............................................................362, 363 system cleaning .....................................................208
length of recording .....................................................363 environmental monitoring ..........................................201
participant identification ....................................362–363 industry setting ...................................................208, 209
integrated chemical detection system .........................203
V mapping BTEX exposure ...................................208–210
Validation process, tremor materials
amplitude ............................................................366–368 electronics module ................................................204
capacity ...............................................................366–368 fluidic and thermal control module.......................204
clinical status, patients ........................................368–369 sample detection module ......................................204
 MOBILE HEALTH TECHNOLOGIES
512 Index

Wearable detection system (cont.) sensors and electronic circuitry ...................................337

sample preconcentration module ..........................204 wireless accelerometer nodes.......................................337
sample separation module .....................................204 Webcam
user interface.........................................................204 detector and emission filter.................................143–144
monitoring system setup drivers and software ....................................................141
detection module ..................................................205 optical system
electronics module ........................................206, 207 detector and emission filter ...........................143–144
fluidic and thermal control module...............205–206 laser excitation ......................................................143
preconcentration module ......................................205 White blood cells (WBC)
separation module .................................................205 lymphocytes and monocytes .......................................193
user interface.........................................................207 microflluidic device .....................................................199
near-ubiquitous smartphone .......................................203 neutrophils ..................................................................193
networking capability .................................................203 Wide-field flow cell
PID.............................................................................202 fabrication...................................................................143
TFs .............................................................................202 3M 9770 double-sided adhesive transfer tape ............143
VOCs .........................................................................201 THP-1 stained with SYTO-9 dye .............................145
Wearable device Wireless accelerometer. See Gait analysis
gait quantification ...............................................344, 347 Wireless sensing. See Tuning forks (TFs)