COMPUTATIONAL

IMAGING AND ANALYTICS IN

BIOMEDICAL ENGINEERING
Algorithms and Applications
AAP Series on Digital Signal Processing, Computer Vision and Image Processing:
Advances and Applications

COMPUTATIONAL
IMAGING AND ANALYTICS IN
BIOMEDICAL ENGINEERING
Algorithms and Applications

Edited by
T. R. Ganesh Babu, PhD
U. Saravanakumar, PhD
Balachandra Pattanaik, PhD
First edition published 2024
Apple Academic Press Inc. CRC Press
1265 Goldenrod Circle, NE, 2385 NW Executive Center Drive,
Palm Bay, FL 32905 USA Suite 320, Boca Raton FL 33431
760 Laurentian Drive, Unit 19, 4 Park Square, Milton Park,
Burlington, ON L7N 0A4, CANADA Abingdon, Oxon, OX14 4RN UK

© 2024 by Apple Academic Press, Inc.

Apple Academic Press exclusively co-publishes with CRC Press, an imprint of Taylor & Francis Group, LLC
Reasonable efforts have been made to publish reliable data and information, but the authors, editors, and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors are solely responsible for all the chapter
content, figures, tables, data etc. provided by them. The authors, editors, and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has
not been obtained. If any copyright material has not been acknowledged, please write and let us know so we may rectify in any
future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any
form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and
recording, or in any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.com or contact the Copyright
Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that are not available on
CCC please contact [email protected]
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are used only for identification
and explanation without intent to infringe.

Library and Archives Canada Cataloguing in Publication

Title: Computational imaging and analytics in biomedical engineering : algorithms and applications / edited by T.R.
Ganesh Babu, PhD, U. Saravanakumar, PhD, Balachandra Pattanaik, PhD.
Names: Babu, T. R. Ganesh, editor. | Saravanakumar, U., 1984- editor. | Pattanaik, Balachandra, 1971- editor.
Description: First edition. | Series statement: AAP series on digital signal processing, computer vision and image
processing: advances and applications | Includes bibliographical references and index.
Identifiers: Canadiana (print) 20230581536 | Canadiana (ebook) 20230581587 | ISBN 9781774914717 (hardcover) |
ISBN 9781774914700 (softcover) | ISBN 9781032669687 (ebook)
Subjects: LCSH: Diagnostic imaging—Mathematical models. | LCSH: Medical informatics—Mathematical models.
Classification: LCC RC78.7.D53 C66 2024 | DDC 616.07/54—dc23
Library of Congress Cataloging-in-Publication Data
Names: Babu, T. R. Ganesh, editor. | Saravanakumar, U., 1984- editor. | Pattanaik, Balachandra, 1971- editor.
Title: Computational imaging and analytics in biomedical engineering : algorithms and applications / edited by T.R. Ganesh
Babu, U. Saravanakumar, Balachandra Pattanaik.
Other titles: AAP series on digital signal processing, computer vision and image processing.
Description: First edition. | Palm Bay, FL : Apple Academic Press Inc., 2024. | Series: AAP series on digital signal processing,
computer vision and image processing: advances and applications | Includes bibliographical references and index. |
Summary: "Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications focuses on
mathematical and numerical methods for medical images and data. The book presents the various mathematical modeling
techniques, numerical analysis, computing and computational techniques, and applications of machine learning for medical
images and medical informatics. It also focuses on programming concepts using MATLAB and Phython for medical image
and signal analytics. The volume demonstrates the use of various computational techniques and tools such as machine
learning, deep neural networks, artificial intelligence and human-computer interaction, fusion methods for CT and pet
images, etc. for diagnosis of brain disorders, cervical cancer, lung disease, melanoma, atrial fibrillation and other circulatory
issues, dental images, diabetes, and other medical issues. Key features: Addresses the various common challenges related
to biomedical image analysis Presents a variety of mathematical models for medical images Discusses applications of
algorithms on medical images for various medical issuses Describes the development of intelligent computing machines
such as embedded systems Explores the programming techniques using MATLAB and Phython for biomedical applications
This book presents a plethora of uses of algorithms and applications in computational imaging and analytics for the
medical/health field. It will serve as a resource on recent advances and trends in the field of computational imaging, where
computation is playing a dominant role in imaging systems"-- Provided by publisher.
Identifiers: LCCN 2023053511 (print) | LCCN 2023053512 (ebook) | ISBN 9781774914717 (hardcover) | ISBN 9781774914700
(paperback) | ISBN 9781032669687 (ebook)
Subjects: MESH: Diagnostic Imaging--methods | Medical Informatics Computing | Biomedical Engineering--methods
Classification: LCC RC78.7.D53 (print) | LCC RC78.7.D53 (ebook) | NLM WN 180 | DDC 616.07/54--dc23/eng/20240109
LC record available at https://lccn.loc.gov/2023053511
LC ebook record available at https://lccn.loc.gov/2023053512
ISBN: 978-1-77491-471-7 (hbk)
ISBN: 978-1-77491-470-0 (pbk)
ISBN: 978-1-03266-968-7 (ebk)
AAP SERIES ON DIGITAL SIGNAL
PROCESSING, COMPUTER VISION,
AND IMAGE PROCESSING:
ADVANCES AND APPLICATIONS

BOOK SERIES EDITORS:

Dr. Manoj Gupta, PhD
Associate Professor, Department of Electronics and Communication
Engineering, JECRC University, Jaipur (Rajasthan), India

Dr. Pradeep Kumar, PhD

Discipline of Electrical, Electronic and Computer Engineering,
Howard College Campus, University of KwaZulu-Natal,
Durban, South Africa

Brief description of the Book Series:

Digital signal processing, computer vision and image processing as a whole

is considered to be one of the most rapidly evolving areas of research and
technology today with growing applications in almost all disciplines of
engineering. Medical imaging, computer vision, healthcare, medical appli­
cations, remote sensing, agriculture, robotics, communication systems and
space exploration are some of the applications of digital signal processing,
computer vision and image processing, to name a few. The present day
curriculum covers many aspects of digital signal processing, computer
vision and image processing, addressing the theoretical aspects in particular.
This book series is intended to supplement the theoretical knowledge with
special emphasis on the practical side. The series content has been specifi­
cally chosen to give a thorough understanding of the fundamental aspects
to advanced applications of digital signal processing, computer vision and
image processing.
The series is open to monograph. handbooks, authored books, edited
volumes, and conference proceedings. This series aims to embrace all
vi AAP Series on Digital Signal Processing, Computer Vision, and Image Processing

aspects, sub-fields and new challenges in the followings research domains

(related topics) but are not limited to:
• Image and Video Processing: Image filtering, restoration and
enhancement, image segmentation, video segmentation and tracking,
morphological processing, feature extraction and analysis, interpola­
tion and super-resolution, motion detection and estimation, computer
vision, pattern recognition, content-based image retrieval, image/
signal computations and services, features and models for image/
signals, machine learning based image and signal processing, data
mining techniques, imaging algebra, mathematical morphology,
probabilistic, statistical, optimization, approximation theory, models
in imaging science, video signal processing, visualization, water­
marking, video surveillance, video compression and streaming, video
analysis and event recognition, biometrics , medical image analysis,
artificial intelligence and related areas.
• Signal Processing: Filters theory, spectral analysis, time-frequency
and time-scale representation, EEG/ECG signals, FIR/IIR and
adaptive filters, statistical signal processing, filtering, detection and
estimation, nonlinear signal processing, radar, antennas, telecommu­
nications systems, acoustics. Signal processing theory and methods,
high signal processing: integrating 5G and IoT with satellite networks,
hands-free speech communication and microphone arrays, wearable
sensor signal processing, architecture and frameworks, audio/speech
processing and coding, watermarking, data mining techniques, statis­
tical and optical signal processing, communication signal processing,
DSP system and embedded systems, multimedia processing, artificial
intelligence , IoT, cyber physical systems and related areas.
• Computer Vision: Algorithms, feature extraction and pattern
recognition, ML and deep learning in vision, CBIR, object and face
recognition, AR/VR, object detection and localization, 3D object
extraction, tracking and visual navigation, cognitive and biological
inspired vision, artificial intelligence, machine learning , ubiquitous
computing and related areas.
• Applications Areas: Biometric, bioinformatics and biomedical
imaging, medial images and applications, healthcare applications,
agriculture applications, augmented and mixed reality, mental health
and cognitive stimulations, security and video-surveillance, quality
control and inspection, archaeology, remote sensing, embedded
systems and applications, automated vehicles, speech and music
AAP Series on Digital Signal Processing, Computer Vision, and Image Processing vii

processing, robotics, rehabilitation, occupational therapy and tele­

medicine, artificial intelligence and machine learning/deep learning
based applications, cyber physical systems, Internet of Things (IoT),
Industry 4.0, Medicine 5.0 and other related applications.

BOOKS IN THE SERIES:

Artificial Intelligence and Machine Learning Techniques in Image
Processing and Computer Vision
Editors: Karm Veer Arya, PhD, Ciro Rodriguez Rodrigues, PhD,
Saurabh Singh, PhD, and Abhishek Singhal, PhD

Computational Imaging and Analytics in Biomedical Engineering:

Algorithms and Applications
Editors: T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD, and
Balachandra Pattanaik, PhD
ABOUT THE EDITORS

T. R. Ganesh Babu, PhD,

Professor, Department of Electronics and Communication Engineering at
Muthayammal Engineering College (Autonomous), Rasipuram,
Namakkal District, India.
T. R. Ganesh Babu, PhD, is currently working as a Professor in the Depart­
ment of Electronics and Communication Engineering at Muthayammal
Engineering College (Autonomous), Rasipuram, Namakkal District. He has
to his credit about 25 years of teaching experience in various engineering
colleges across India. He has over 118 research publications in international
and national journals, conferences, books, and book chapters. He has
authored six engineering books on topics in communication engineering,
linear integrated circuits, digital communication, digital image processing,
control systems, and FAQs in analog and digital signals. He has published
three edited books as a chief editor under Dipti Press and one book from
IGI Global. He has to his credit one patent granted and has filed nine more
patents and publishing. He has already guided 10 research scholars, and
seven more are pursuing their degrees at Anna University India. He is a
reviewer for Springer, Elsevier, and Inderscience Publisher. He received his
Distinguished Faculty Award from Venus International Foundation, Chennai.
He received a PhD from Anna University, Chennai, India, in 2014 in the field
of medical image processing.

U. Saravanakumar, PhD
Professor and Head of Electronics and Communication Engineering,
Muthayammal Engineering College (Autonomous), Namakkal, India

U. Saravanakumar, PhD, is working as a Professor and Head of Electronics

and Communication Engineering in Muthayammal Engineering College
(Autonomous), Namakkal. His research interests are in the field of VLSI
design, silicon photonics, reconfigurable computing, system on chip, and
embedded systems. He worked in various institutions designated in various
x About the Editors

positions. He graduated in Electronics and Communication Engineering

from Anna University, Chennai, in 2006. In the year 2008, he received
his master’s degree in VLSI Design from Anna University, Chennai. He
received his doctoral degree in the faculty of Information and Communica­
tion Engineering, specializing in VLSI Design and on-chip communication
from Anna University in 2014.

Balachandra Pattanaik, PhD

Balachandra Pattanaik, PhD, is currently working as a full-time Professor
in the Department of Electrical and Computer Engineering, in the College
of Engineering and Technology, Wollega University, Ethiopia, Africa, Dr.
Pattanaik holds PhDs in Electrical Engineering (Embedded Systems) and in
Mathematics, and has an ongoing postdoctoral fellowship on electric vehi­
cles. He holds 20 innovation patents issued by the Indian Government. He
has the following: innovation patent grants: two, international patent grants
in Australia and Germany: four. He has published 13 engineering books and
six book chapters as well as many national and international publications.
He is an advisor to the IEEE, Power and Energy Student Chapter at Bule
Hora University in Ethiopia, a senior member of IEEE, and a recipient of
the Chartered Engineer Award from the Institution of Engineers of India.
He holds many administrative positions, such as Head of Department of
Electrical Engineering, Principal, and Anna University Chief Superinten­
dent for university examinations, NSS coordinator, research coordinator,
accreditation management representative, editorial board member, and chair/
vice chair for conference sessions. He has guided many undergraduate and
postgraduate research projects and has been a foreign PhD thesis evaluator
for many universities.
CONTENTS

Contributors...........................................................................................................xiii
Abbreviations ........................................................................................................ xvii
Preface ................................................................................................................... xxi

1. Statistical Analysis of Seizure Data to Support Clinical Proceedings ........1

Vajiram Jayanthi, Sivakumar Shanmugasundaram, and Syed Khasim

2. Spatial Preprocessing in Segmentation of Brain MRI

Using T1 and T2 Images...............................................................................21
S. Jacily Jemila and A. Brintha Therese

3. Comparative Volume Analysis of Pediatric Brain with

Adult Brain Using T1 MRI Images .............................................................41
S. Jacily Jemila and A. Brintha Therese

4. Comparison of Region of Interest and Cortical Area Thickness of

Seizure and Hemosiderin-Affected Brain Images......................................63
Vajiram Jayanthi, Sivakumar Shanmugasundaram, and C. Vijayakumaran

5. Design and Analysis of Classifier for Atrial Fibrillation and

Deep Neural Networks with ECG ...............................................................77
M. Sivakumar, S. Omkumar, B. Arputhamary, and Yohannes Bekuma Bakare

6. Design and Analysis of Efficient Short Time Fourier

Transform Based Feature Extraction for Removing
EOG Artifacts Using Deep Learning Regression .......................................87
M. Sivakumar, K. Umapathy, P. Srinivasan, and S. Punitha

7. Machine Learning for Medical Images.......................................................95

N. Gopinath, M. Suresh Anand, and M.V. Ishwarya

8. Innovations in Artificial Intelligence and Human Computer

Interaction in the Digital Era.....................................................................105
M. V. Ishwarya, M. Suresh Anand, A. Kumaresan, and N. Gopinath

9. Computer-Aided Automatic Detection and Diagnosis of

Cervical Cancer by Using Feature Markers.............................................147
P. Sukumar, R. Murugasami, A. Rajan, and S. Sharmila
xii Contents

10. A Study on Sentiment Analysis ..................................................................163

M. Moorthy, Yohannes Bekuma Bakare, and Balachandra Pattanaik

11. Applications of Magnetic Resonance Imaging Techniques and

its Advancements.........................................................................................175
V. Ramesh Babu, S. Mary Cynthia, K. Savima, and G. Lakshmi vara Prasad

12. A Hybrid Clustering Approach for Medical Image Segmentation.........187

M. Malathi, K. Sekar, Mahendrakan K, and P. Sinthia

13. Approaches for Analyzing Dental Images with Medical

Image Processing with Its Statistics ..........................................................201
G. Balanagireddy, P. Subramanian, J. Martin Sahayaraj, and S. Muthukumar

14. An Investigation on Diabetes Using Multilayer Perceptron ...................215

J. Surendharan, A. Karunamurthy, R. Praveena, and K. Shebagadevi

15. Dermoscopic Implementation and Classification on Melanoma

Disease Using Gradient Boost Classifier ...................................................229
B. Balakumar, K. Sakthi Murugan, N. Sureshkumar, A. Purushothaman

16. Image Processing and Deep Learning Techniques for

Lung Disease Segmentation Using KNN Classifier..................................235
G. Soniya Priyatharsini, N. Naveenkumar, Balachandra Pattanaik, and M. Kumarasamy

17. Design Detecting and Classifying Melanoma Skin Cancer

Using CNN with K Means Clustering .......................................................243
R. Sumathi and Sridhar P. Arjunan

18. Detection of Lung Cancer Using Fusion Methods for CT and

PET Images..................................................................................................255
V. Ramesh Babu, G. Gunasekaran, and Sreenithi R.

19. A Framework Promoting Position Trust Evaluation System in

Cloud Environment ....................................................................................275
S. R. Sridhar, S. Pragadeeswaran, and M. Ganthimathi

20. Efficient Machine Learning Techniques for Medical Images .................299

Gururama Senthilvel P., K. Sampath Kumar, and T. Ganesh Kumar

Index .....................................................................................................................321
CONTRIBUTORS

M. Suresh Anand
Department of Computing Technologies, School of Computing, SRM Institute of Science &
Technology, Kattankulathur, India
Sridhar P. Arjunan
Department of Electronics and Instrumentation Engineering, SRM Institute of Science and Technology,
Kattankulathur, India
B. Arputhamary
Department of Computer Applications, Bishop Heber College, Tiruchirappalli, India

V. Ramesh Babu
Department of Computer Science and Engineering, Sri Venkateswara College of Engineering,
Sriperumbudur, India
Yohannes Bekuma Bakare
Department of Electrical and Computer Engineering, College of Engineering and Technology,
Wollega University, Ethiopia, Africa
G. Balanagireddy
Department of Electronics and Communication Engineering, Rajiv Gandhi University of Knowledge
Technologies-Ongole Campus, Ongole, India

B. Balakumar
Centre for Information Technology and Engineering, Manonmaniam Sundaranar University,
Tirunelveli, India

S. Mary Cynthia
Department of ECE, Jeppiaar Institute of Technology, Chennai, India

M. Ganthimathi
Department of CSE, Muthayammal Engineering College, Namakkal, Tamil Nadu, India

N. Gopinath
Department of Computer Science and Engineering, Sri Sairam Engineering College, Chennai, India

G. Gunasekaran
Department of Computer Science and Engineering, Dr. M. G. R. Educational and Research Institute,
Maduravoyal, Chennai, India

M. V. Ishwarya
Department of Artificial Intelligence and Data Science, Agni College of Technology, Chennai, India
Vajiram Jayanthi
SENSE, Vellore Institute of Technology, Chennai, India

S. Jacily Jemila
Vellore Institute of Technology, Chennai, India
xiv Contributors

Mahendrakan K.
Department of Electronics and communication Engineering, Hindusthan Institute of Technology,
Coimbatore, Tamil Nadu, India

A. Karunamurthy
BWDA Arts and Science College, Vilupuram, India

Syed Khasim
Department of CSE, Dr. Samuel George Institute of Engineering &Technology, Andhra Pradesh, India

T. Ganesh Kumar
Department of Computing Science and Engineering, Galgotias University, Greater Noida,
Uttar Pradesh, India

K. Sampath Kumar
Department of Computing Science and Engineering, Galgotias University, Greater Noida,
Uttar Pradesh, India

M. Kumarasamy
Department of Computer Science, College of Engineering and Technology, Wollege University,
Ethiopia, Africa

A. Kumaresan
School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, India

M. Malathi
Department of Electronics and communication Engineering, Rajalakshmi Institute of Technology,
Chennai, India

M. Moorthy
Muthayammal Engineering College, Rasipuram, India

K. Sakthi Murugan
Department of ECE, PSN College of Engineering and Technology, Tirunelveli, India

R. Murugasami
Department of Electronics and Communication Engineering, Nandha Engineering College
(Autonomous), Erode, India

S. Muthukumar
Department of CSE, B.S. Abdur Rahman Crescent Institute of Science and Technology, Chennai, India

N. Naveenkumar
Department of CSE, Muthayammal Engineering College, Rasipuram, India

S. Omkumar
Department of ECE, SCSVMV (Deemed University), Kanchipuram, India

Gururama Senthilvel P.
Department of Computing Science and Engineering, Galgotias University, Greater Noida,
Uttar Pradesh, India

Balachandra Pattanaik
Department of Electrical and Computer Engineering, College of Engineering and Technology,
Wollege University, Ethiopia, Africa

S. Pragadeeswaran
Department of CSE, Muthayammal Engineering College, Namakkal, Tamil Nadu, India
Contributors xv

G. Lakshmi Vara Prasad

Department of IT, QIS College of Engineering and Technology, Andhra Pradesh, India

R. Praveena
Department of ECE, Muthayammal Engineering College, Namakkal, India

G. Soniya Priyatharsini
Department of ECE, DR. M. G. R. Educational and Research Institute, Maduravoyal, Chennai, India

S. Punitha
Department of ECE, Muthayammal Engineering College, Rasipuram, India

A. Purushothaman
Department of ECE, Hindhusthan Institute of Technology, Coimbatore, India

Sreenithi R.
Department of Computer Technology, Madras Institute of Technology, Chennai, India

A. Rajan
Department of ECE, Sreerama Engineering College, Thirupathi, India

J. Martin Sahayaraj
Department of Electronics and Communication Engineering, Sri Indu College of Engineering and
Technology, Hyderabad, Telangana, India
K. Savima
Department of Computer Science, S.T.E.T. Women’s College, Mannargudi, India

K. Sekar
Department of Electrical and Electronics Engineering, Hindusthan College of Engineering and
Technology, Coimbatore, Tamil Nadu
Sivakumar Shanmugasundaram
SENSE, Vellore Institute of Technology, Chennai, India

S. Sharmila
Department of Civil Engineering, Nandha Engineering College, Erode, India

K. Shebagadevi
Department of ECE, Muthayammal Engineering College, Namakkal, India

P. Sinthia
Department of Biomedical Engineering, Saveetha Engineering College, Chennai, India

M. Sivakumar
Department of ECE, Mohamed Sathak A. J. College of Engineering, Chennai, India

S. R. Sridhar
Department of CSE, Muthayammal Engineering College, Namakkal, Tamil Nadu, India

P. Srinivasan
Department of CSE, Muthayammal Engineering College, Rasipuram, India
P. Subramanian
Department of CSE, Mohamed Sathak A. J. College of Engineering, Chennai, India

P. Sukumar
Department of Computer Science and Engineering, Nandha Engineering College (Autonomous),
Erode, India
xvi Contributors

R. Sumathi
Department of Computer Science and Engineering, Kalasalingam Academy of Research and Education,
Krishnankoil, India

J. Surendharan
HKBK College of Engineering, Bangalore, Karnataka, India

N. Sureshkumar
Department of ECE, Muthayammal College of Engineering, Rasipuram, India

A. Brintha Therese
Vellore Institute of Technology, Chennai, India

K. Umapathy
Department of ECE, SCSVMV (Deemed University), Kanchipuram, India

C. Vijayakumaran
Department of CSE, SRM Institute of Science and Technology, Kattankulathur, Chennai, India
ABBREVIATIONS

AD axial diffusivity
AGI artificial general intelligence
AI artificial intelligence
AML acute myeloid leukemia
ANN artificial neural network
ARR arrhythmia
AS attacker scale
ASD autism spectrum disorder
ATS attack target scale
BBB blood–brain barrier
BC binary classifier
BFC bias field corrector
BN Bayesian network
BOLD blood-oxygen-level-dependent
BSE brain surface extractor
BSL British sign language
CAD computer-aided diagnostic
CAF cooperation attack frequency
CAPEX capital expenditure
CHO channeled hotelling observer
CNN convolution neural network
CNNPL convolutional neural network with a prototype learning
CNS central nervous system
CRM customer relationship management
CS collusion set
CSF cerebrospinal fluid
CSL Chinese sign language
CT computer tomography
DCNN deep convolutionary neural network
DL deep learning
DT decision tree
DTI diffusion tensor imaging
ECG electrocardiography
EEG electroencephalography
xviii Abbreviations

EOG electro-oculogram
ERP enterprise resource planning
EV exploration views
FA fractional anisotropy
FCM fuzzy C-means
FL feedback limit
FLAIR fluid attenuated inversion recovery
fMRI functional magnetic resonance imaging
FP false positive
FS feedback set
FWHM full width half maximum
GLM general linear model
GOFAI good old fashioned artificial intelligence
GSL Greek sign language
HCI human computer interaction
HE histogram equalization
HPV human papilloma virus
IaaS infrastructure services
IdM identity management service
IQA image quality assessment
ISTFT inverse signals using short time Fourier transform
KNN K-nearest neighbor
LASSO least absolute shrinkage and selection operator
LR logistic regression
LSF French sign language
LSTM long short-term memory
MCT manual contour tracing
MD mean diffusivity
ME maximum entropy
MEG magnetoencephalogram
ML machine learning
MRI magnetic resonance imaging
NB Naïve Bayes
NMR nuclear appealing resonation
NN neural network
NSCT non-subsample contourlet transform
OFR out-of-field recurrence
OLAP online analytical processing
OPEX operational expenditure
Abbreviations xix

PaaS platform as a service

PCA principle component analysis
PET positron emission tomography
PReDicT predicting response to depression treatment
PSNR peak signal noise ratio
PTSD posttraumatic stress disorder
PVC partial volume classifier
RBFN radial basis function network
RF radio frequency
RF random forest
RMSE root mean square error
RNN recurrent neural network
ROI region of interest
SaaS software as a service
SD standard deviation
SIM similarity index measure
SLAs service-level agreements
SMEQ subjective mental effort question
sMRI structural MRI
SNR signal-to-noise ratio
SPECT single photon emission computed tomography
SPM statistical parametric mapping
STC slice timing correction
STFT signals using short time Fourier transform
STIs sexually transmitted infections
SVM support vector machine
SvReg surface and volume registration
TAI traumatic axonal injury
TCGA tissue from the Cancer Genome Atlas
TD typically developed
TDO total dermoscopic value
TE time to echo
TES trust management system
TF time-frequency
TR repetition time
PREFACE

Medical images are useful in analyzing the internal physical anatomy of the
human body and therefore, they are treated as a core component in medical
diagnosis and treatment. These medical images contain information about the
human anatomy such as cell and tissue arrangements, any growth in human
cells, and so on. These information cannot be interpreted easily by human
eyes and need some medical devices, tools, and software to analyze in-depth.
On the other side, medical images are stored in databases, forming big-data
for population groups, and thus, handling of medical data (or images) is also
a crucial task.
To simplify these issues, various image-processing techniques are applied
on medical images and datasets with suitable algorithms to automatically
or semiautomatically extract useful information about the patients. For
further simplification and optimization, emerging techniques such as neural
networks, machine learning, deep learning (DL), and artificial intelligence
are adapted for feature extraction of medical images, medical image segmen­
tation, and image-based bio-models.
These advancements enable appropriate use of medical images in the
field of healthcare including: (1) medical knowledge of biological systems
through detailed analysis of structures and functions of particular biological
part or organs; (2) computer-assisted frameworks for medical treatments;
and (3) development of image datasets of human body (bodies) for diagnosis,
guiding doctors, and research activities.
This book comprises 20 chapters, and they are organized in the following
manner.
Chapter 1 discusses statistical analysis of seizure data to support clinical
proceeding. The radiographic imaging is a powerful and clinically important
tool within oncology. Artificial intelligence is used for the quantification of
radiographic characteristics of images, using predefined algorithms. The
strategies of clinical imaging data include normalization, robust models,
and statistical analyses. They also help in improving quality and enhancing
perfection of medical image analysis and supporting clinicians diagnose,
treat, plan, and cover changes, and execute procedures more safely and
effectively. Statistical models give a surgery-specific environment to the
xxii Preface

problem from a training set of cases conforming to the problem along with
the result.
Chapter 2 presents a spatial preprocessing in segmentation of brain MRI
using t1 and t2 images. Preprocessing is a very important process before
segmentation in medical image processing. In this chapter, T1 and T2 MRI
images of brain are segmented before and after preprocessing using SPM-12
neuroimaging software, and the importance of preprocessing is discussed
by using five important quality metrics MSE, PSNR, SNR, EPI, and SSIM
(Structural Similarity Index Metric).
Chapter 3 discusses a comparative volume analysis of pediatric brain
with adult brain using t1 MRI images segmentation of pediatric brain is very
useful in medical image analysis to supervise the brain growth and develop­
ment in infants. Atlas-based methods are used for automatic segmentation
of MRI images of brain. In this chapter, automatic segmentation of four
pediatric and three adult brains are done using Brainsuite19a brain imaging
software, and the volumes of different brain parts are computed.
Chapter 4 presents a comparison of region of interest and cortical area
thickness of seizure and hemosiderin-affected brain images. Brain disorder
and tumors are caused by the severe neurological abnormality in their func­
tions. Hemosiderin is induced by changes in the characteristic functional
magnetic field and can be detected using susceptibility weighted T1 images.
Clinical syndromes and its relevance are unclear. Hemosiderin and epilepsy
brain images are preprocessed with available software tool. Based on the
region of interest, mean thickness area and cortical thickness area of the
different parts of the brain are determined.
Chapter 5 presents a design and analysis of classifier for atrial fibrilla­
tion detection and ECG classification based on deep neural networks (DNN)
atrial fibrillation (AF), which is the most common cardiac arrhythmia as
well as a significant risk factor in heart failure and coronary artery disease.
Deep learning is the current interest of different healthcare applications
that includes the heartbeat classification based on ECG signals. AF can be
detected by using a short ECG recording. This chapter describes constructing
a classifier for detection of atrial fibrillation in signals of ECG using deep
neural networks.
Chapter 6 discusses design and analysis of efficient short time Fourier
transform-based feature extraction for removing EOG artifacts using
deep learning regression. It also discusses how to eliminate electro­
oculogram (EOG) noise from the signals of electroencephalogram (EEG)
by employing the benchmark dataset of EEGdenoiseNet. This work involves
Preface xxiii

EOG-contaminated signals of EEG signals for training a regression model

based on deep learning in order to eliminate the artifacts of EOG. Initially,
original input signals are used to train the model of regression and then trans­
formed signals using short-time Fourier transform (STFT) are employed.
Chapter 7 discusses that research in medical imaging is increasingly
turning to machine learning approaches. To recognize patterns in medical
images, this method relies on pattern recognition. In order to enhance medical
imaging applications, these algorithms are now being used to detect faults in
the sickness diagnosing system, which may lead to significantly ambiguous
medical therapy.
Chapter 8 discusses about the innovations in artificial intelligence (AI)
and human computer interaction (HCI) in the digital era. AI and HCI have
frequently been described as having contradicting perspectives on how
people and PCs ought to interface. As the two advance, there is a profound
differentiation that cuts across these networks, in how scientists imagine the
connection between information and plan. Evolution in AI has prompted
improvement in strategies and apparatuses that have influenced regions
outside AI's core functions.
Chapter 9 presents a computer-aided automatic detection and diagnosis
of cervical cancer by using feature markers. The proposed system constitutes
reprocessing, feature succession, and nuclei partition. Internal structure
processes are used to partition cell center area. The gray equal, wavelet, and
GLCM structures are mined from standard plus diseased cell center. The
mined structures are proficient and categorized using ANFIS categorization.
Chapter 10 presents a detailed study of sentiment analysis. It explains the
basics of sentiment analysis, its types, and different approaches of sentiment
analysis and more. The open-source tools and datasets are also discussed.
Magnetic resonance imaging (MRI) techniques and its advancements in
biomedical research and clinical applications are presented in Chapter 11.
The emergence of new techniques increases the applications of MR imaging
remarkably. MRI is performed with the use of magnetization properties.
To capture MRI, a strong magnetic field is applied to randomly oriented
protons of human body available in water molecules to make changes in
the alignment. Then this magnetization is disordered by applying external
radio frequency wave. This absorbed RF energy is then emitted with the help
of several relaxation processes and the protons realigned. Subsequently, the
emitted signals are determined.
Chapter 12 presents a hybrid clustering approach for medical image
segmentation. The predominant algorithms used in hybrid clustering
xxiv Preface

techniques are K-means, adaptive K-means, spatial fuzzy c-means algorithm.

By combining the K-means algorithm along with a spatial fuzzy c-means
algorithm, the hybrid clustering segmentation yields higher accuracy level
for tumor detection and also minimizes the processing time required for the
segmentation process.
Chapter 13 presents an analysis of dental image with medical image
processing. This chapter deals with such aspects considering radio graphical
images. Measuring the distance between mandible inferior border and the
superior border of “alveolar” for image is done manually. Thresholding via
various methods such as “Haung,” “Otsu” and their corresponding values are
provided for analysis.
Chapter 14 discusses an investigation on diabetes using multilayer
perceptron. The goal of this work is to identify, detect, and forecast the
emergence of diabetes in its earliest stages by employing machine learning
techniques and algorithms. When it comes to diabetes classification, an MLP
is used.
Chapter 15 elaborates a dermoscopic implementation and classification
on melanoma disease using gradient boost classifier. In this study, the authors
proposed deep learning techniques for classification of deadly diseases.
Dermoscopic images are used in this study as a dataset directory. The authors
of this chapter used the classifier of gradient boost and feature extraction by
NSCT to predict an accurate and specific value of melanoma disease.
Chapter 16 presents a segmented lung disease prediction using deep
learning techniques such as convolutional neural network (CNN) and KNN
classifier.
Chapter 17 describes a design detecting and classifying melanoma skin
using CNN with K-means clustering. An automated approach with cuckoo
search optimization and K-means clustering for detecting the skin cancer
and for classifying the cancer as normal and abnormal by support vector
machine have been proposed. For preprocessing, median filters are used to
reduce noise inference in the input image. For validating the accuracy, the
authors utilize the IISC-DSI dataset.
Chapter 18 describes detection of lung cancer using fusion methods for
CT and PET images. In the first part of the work, segmentation of lungs,
a tumor in CT image is used to spatially weigh fuzzy c-means clustering
(SWFCM) techniques. In the second part of the work, segmentation of lungs,
a tumor in PET image is used to spatially weigh fuzzy c-means clustering
(SWFCM) techniques. In the third part, the diagnosis is strengthened for
mass screening; the CT images and the PET images are fused effectively.
Preface xxv

The four fusion methods namely wavelet transform, curvelet transform, non­
subsample contourlet transform (NSCT), and multimodal image fusion are
applied. The performance analysis entropy, peak signal noise ratio (PSNR),
standard deviation (SD), structural similarity index measure (SIM), and root
mean square error (RMSE) are computed.
Chapter 19 presents a framework promoting position trust evaluation
system in the cloud environment. The design and development of Cloud
Armor, a legacy trust evaluation framework that offers a set of functions
to provide trust as a service, is discussed in this chapter (TaaS). It also
discusses the problems in calculating trust based on input from cloud users.
This technology effectively protects cloud services by detecting hostile and
inappropriate behavior by the use of trust algorithms which may recognize
on/off assaults and colluding attacks by using different security criteria.
Chapter 20 presents machine learning for medical images. This chapter
comprises key phases of digital image processing applications of medical
images, usages, and difficulties of artificial intelligence in healthcare
applications.
Our deep gratitude to all the contributors for providing quality content
to the readers of the book. We extend our sincere thanks to all the reviewers
for their suggestions which helped us to pack this book with its merits. From
the bottom of our heart, we thank our family members, employers, research
scholars, and colleagues for their unrestricted support. We wish to thank the
tireless team of Apple Academic Press for their wonderful support and guid­
ance throughout the process of making this book.
— T. R. Ganesh Babu
U. Saravanakumar
Balachandra Pattanaik
CHAPTER 1

STATISTICAL ANALYSIS OF SEIZURE

DATA TO SUPPORT CLINICAL
PROCEEDINGS
VAJIRAM JAYANTHI1, SIVAKUMAR SHANMUGASUNDARAM2, and
SYED KHASIM3
1
SENSE, Vellore Institute of Technology, Chennai, India
2
SENSE, Vellore Institute of Technology, Chennai, India
Department of CSE, Dr. Samuel George Institute of Engineering
3

&Technology, Andhra Pradesh, India

ABSTRACT

Radiographic imaging is highly valued and widely utilized in the field of

oncology. It plays a crucial role in assisting clinicians with diagnostic, treat­
ment planning, and procedural decision-making tasks. Artificial intelligence
(AI) technology is employed to analyze radiographic images by quantifying
various characteristics through predetermined algorithms. Strategies for
managing clinical imaging data involve normalization techniques, robust
modelling approaches, and statistical analyses. These methods aim to improve
the quality and accuracy of medical image analysis, thereby enhancing the
precision of clinical outcomes. Statistical models provide a customized
framework for addressing specific surgical concerns by training on a dataset
comprising cases with corresponding outcomes. Such models are particularly
useful for image recovery tasks, assessing intensity, graph curves, and global
shapes to guide clinical recommendations toward the most likely outcome.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
2 Computational Imaging and Analytics in Biomedical Engineering

Visual interpretation by the medical team is currently the prevailing method

for analyzing clinical images, heavily reliant on subjective interpretations.
Therefore, automated analysis of these images using computer algorithms
offers a more reliable option to ensure a highly precise analysis. This chapter
focuses on the statistical analysis of seizure data to support clinical decision-
making processes.

1.1 INTRODUCTION

Visual interpretation is not always the right choice in the medical process,
so automated image analysis using computers with a suitable algorithm
is only the perfect choice.1 The radiographic analysis of disease images
needs more attention. Because data are collected during routine clinical
practice, undergone, pre- and post-processing work. The artificial intelli­
gence (AI), statistical methodology, of image processing capabilities, with
big data analysis of datasets, is growing exponentially.2,3 The automated
quantification of the radiographic process is more useful in the discovery,
characterization, and monitoring of conditions are called radiomics,4 it
uses the predefined features of the images5 based on the aspects of texture
or features, shape, and intensity used for the image illustration.6 The early
success of radiomics giving clinical opinions of different brain-related
diseases7 has increased the rapid-fire expansion in this medical field.8
The analysis of radiologic multitudinous challenges has been addressed
in quantitative fields of biostatistics.9 Microarray data normalization,
transformation,10 batch effects in high-throughput data,11 Bayes methods
used for the adjusting batch effects in microarray expression,12 a statis­
tical method in DNA hybridization,13 the study of cancer subtypes.14
Researchers in biostatistics have also contributed to the development of
statistical methods for radiologic analysis, addressing challenges such as
normalizing and transforming microarray data, mitigating batch effects
in high-throughput data, and studying cancer subtypes.15 15 In an earlier
state, data analysis in radiology faces a lot of challenges and avoidable
data analysis pitfalls.16 The more than 150 biomedical image analyses are
discussed up to 2016. To quantify the robustness of a ranking, Kendall’s
tau of statistical methods is used such as statistical analysis with radiomic
quantification, biomarker identification, and validation.17 The different
ways used in AI are especially useful when it comes to the field of “big
data,” which describes large volumes of complex and variable data.
Statistical Analysis of Seizure Data to Support Clinical Proceedings 3

1.2 MATERIALS AND METHODS

1.2.1 MEDCALC

MedCalc is a useful clinical calculator used for evidence-based medicine.

This tool provides hundreds of clinical decisions, which include risk scores,
algorithms, dosing calculators, diagnostic criteria, formulas, and classifica­
tions are all framed in the easy handling way. MedCalc serves for clinical
decision-making. It is designed to do statistical analysis and offer the relevant
clinical decision-making purpose. The use of MedCalc is recommended for
a fast and accurate method for analyzing patient data. It is very useful for
clinical decision-making. The tool was designed based on the calculating
confidence intervals of regression,18 correlation and residuals,19 regression,
and analysis of variance.20 Funnel plots are used for detecting bias in meta­
analysis. The tool used to find the statistical analysis such as meta-analysis,
clustering, and regression.21 The receiver-operating characteristic (ROC)
method is used for the evaluation purpose.22

1.2.2 CLUSTER ANALYSIS

Clustering is a method of grouping the set of objects in the same cluster

(group) that are similar to each other clusters. The clustering algorithms
operate in noisy, deficient, and high dimensional data, their performance
based on the datasets. The literature shows the two cluster algorithms, cluster
ensembles method, covariance matrix, comparison of supervised classifiers,
clustering data mining, simulated annealing, genetic algorithms, the problem
of overfitting, data clustering, statistical computing, financial risk analysis,
classification purpose, SOM neural network with fuzzy means, k means,
and traditional clustering algorithms.23 Researchers in biostatistics have also
contributed to the development of statistical methods for radiologic analysis,
addressing challenges such as normalizing and transforming microarray
data, mitigating batch effects in high-throughput data, and studying cancer
subtypes.15 A modified interpretation29 was used to produce 400 distinct
datasets, used to quantify the clustering algorithms. The data generation
approaches are also used.30 The clustering algorithm relies on performance
and parameter values,31 and overfitting datasets.32 The new dataset gives
good to bad performance sometimes. Clustering algorithms are used in
several R programming languages.33 The algorithms were estimated based
on the number of features, classes, objects for each class, and the average
4 Computational Imaging and Analytics in Biomedical Engineering

distance between classes. The data generation and performance measures are
used to compare the algorithms are shown in this paper.

1.2.3 META-ANALYSIS

Meta-analysis solves the genetic studies issues by identifying inter-study

heterogeneity and two-stage process. The first measure of treatment effect
is 95% confidence intervals (CI) used to measure odds ratios (OR), relative
risks (RR), and risk differences. Meta-analysis can be qualitative or quanti­
tative (meta-analysis). A meta-analysis uses statistical methods to perform:
data acquisition, problem formulation, quality, statistical analysis of quanti­
tative data, and clinical interpretation. Meta-analyses of the diagnostic test
measure the accuracy, specificity, and sensitivity of the uncertainty.34 These
measures are used in radiology and biomedical studies. These studies provide
synthesized, quantitative information on certain topics to get meta-analyses
of diagnostic accuracy.35 To evaluate the quality, statistical tools are used36
to access the impact of evidence37 and the statistical results.38 The study of
100 top-cited meta-analyses of diagnostic accuracy, published in radiology
journals.39 The meta-analysis literature is based on PubMed.40 The bivariate
analysis of sensitivity and specificity was published in 2005 also used for
this chapter analysis.41 The meta-analyses to perform a quantitative imaging
modality of X-ray, ultrasound computed tomography, magnetic resonance
imaging, used for detection of a specific condition or disease, concerned
with diagnostic accuracy measures.

1.2.4 GENETIC META ANALYSIS

Meta-analyses of genotyping studies provide advanced techniques of

genotyping at untyped HapMap loci, enabling the combination of genotype
data from datasets with millions of variants. HapMap loci refer to specific
locations on the human genome that were examined and genotyped as part of
the HapMap project. The loci (plural of locus) in the HapMap were selected
based on their representation of common genetic variations across different
human populations. These loci were genotyped using various genetic tech­
niques to determine the specific genetic variation present at each location.
The data from HapMap loci has been widely used by researchers to study
the genetic basis of diseases, drug responses, and other complex traits. These
analyses can be carried out in a sequential, cumulative manner to enhance
Statistical Analysis of Seizure Data to Support Clinical Proceedings 5

statistical power. The establishment of consortia has greatly facilitated

advancements in various fields. Numerous non-overlapping consortia exist
in certain genetic diseases, each contributing to the epidemiological studies
of specific aspects of their designs, data, and sample collections42. The geno­
type statistics can currently be combined throughout datasets with millions
of variants, can be conducted in a sequential and cumulative manner. The
development of consortia43 was advanced in many fields.44 In some genetic
sicknesses, numerous non-overlapping consortia exist, GWA datasets used
for epidemiological studies of vital elements of their layout, statistics, and
sample series cannot be altered. E-book bias, shows the traditional meta­
analysis literature,45 the advantage of GWA research in genome studies.46
Public repositories have GWA datasets, especially those funded by the
public.47 But public statistics availability poses numerous challenges. The
“genome-wide statistical significance” used for phenotype analyses.48 A
couple of analytical alternatives for the identical records are extraordinary
genetic fashions, used for various adjustments. Population stratification was
used for the structured mapping.49 The unavoidable overlap and covariance
accounted in the meta-analysis. The dataset variance increases the correla­
tion which is protected within the calculations.
The restrained overlap between specific genotyping platforms and evalu­
ation of imputed versions is done in all experiments separately and managed
for population stratification of genotype assignments. The heterogeneity is
Cochran’s Q, which tries to answer statistically, the threshold of heteroge­
neity based on Q is p < 0. The statistic is used to quantify the percentage
of variation across studies, indicating inconsistency. It is calculated as (Q-
degree of freedom)/Q ratio.51 The variance, τ2, heterogeneity, is one kind of
meta-analyses, besides heterogeneity biases in a different way affecting the
outcomes of various datasets, genetic results, and distinct biological placing
of virtually informative heterogeneity. The meta-analysis aims to determine
the association at a genome-wide level. The information may be odds ratios,
standardized impact sizes, or other metrics together with variance or 95%
self-belief C program language period of p-values for different datasets.52

1.2.5 META ANALYSIS RISK AND DIFFERENCE IN IMAGE

PROCESSING

The risk difference between two proportions is a measure of effect size in

studies where the variable is a dichotomous response. In the meta-evaluation
of binomial proportions, the impact is the odds ratio (OR), usually analyzed
6 Computational Imaging and Analytics in Biomedical Engineering

as log (OR), risk ratio (RR), and its logarithm interpretation. The log (OR)
and log (RR) are each unbounded, observe-stage occasion in the C program­
ming language (0, 1). The binomial generalized linear combined fashions
(GLMMs) and beta-binomial (BB) distribution, with the logit hyperlink
feature, was used for the meta-analysis. The Cochrane meta-analyses of RR
bias effects from the conventional inverse-variance–weighted technique. The
epidemiologic research is executed with meta-regression. The linear rela­
tion is between publicity and the algorithm, a nonlinear relation among an
exposure and relative chance and statistical solution also effortlessly carried
out.53 The relative dangers are equal to referent class and the meta-regression
contributions of meta-evaluation. Greenland and Longenecker54 developed a
technique of dose–response data with meta-analysis.

1.2.6 META ANALYSIS OF RISK RATIO DIFFERENCE

The risk ratio (RR, or relative risk) of two event groups, whereas the odds
ratio (OR) of the odds of an event, measures a value of 1 that estimates
effects for both interventions. The measures regarding patient characteristics,
comorbidities, symptoms crucial signs, and symptoms have been extracted
from a meta-analysis. The risk metrics of meta-analysis became carried out
to assess the two consequences: excessive and mortality. A meta-analysis
was conducted using a random-effects model to pool the regression data
on abnormal ratios (ORs). The ORs for mortality studies were calculated
for any severe illness (patient case history, ICU-Incentive care unit admis­
sion) and the same scientific variable estimation, whether multivariate or
univariate. The data were analyzed using R statistical software along with
the meta-analysis and plots generated using the R package meta values.55,56

1.2.7 META ANALYSIS OF ROC CURVE

The ROC curve is used for the estimation of the individual curve. The param­
eters are then pooled with bivariate random effects. Receiver-running feature
(ROC) is used for evaluating the overall diagnostic performance tests and
for comparing the accuracy of statistical version of logistic regression, linear
discriminant analysis that classifies subjects into 1 of 2 categories, diseased
or not diseased. Predictive modeling to estimate anticipated outcomes
consisting of mortality on patient risk traits in research with ROC analysis.
The measures of accuracy, sensitivity, specificity, and location underneath
Statistical Analysis of Seizure Data to Support Clinical Proceedings 7

the curve (AUC) that use the ROC curve. To estimate accuracy with ROC
strategies, the ailment reputation of the affected person is measured without
blunders is called the gold trendy. The diagnostic accuracies are specificity
(i.e., true bad charge) and sensitivity (i.e., true fine charge). The final results
of a threshold of the diagnostic test used to classify subjects. The ROC
curve is discussed in the coronary restenosis and peak oxygen consumption
analysis articles (57–58).

1.2.8 NON-LINEAR REGRESSION

Nonlinear regression is a mathematical modeling technique that relates two

variables, X and Y, with a nonlinear curved relationship rather than a straight
line (y = mx + b). This method is commonly used in image processing,
particularly for analyzing images that contain a combination of shading
and noise. When applied to images, nonlinear regression can be used to
measure both qualitative and quantitative aspects, with a particular focus
on addressing shading problems. In this context, an image can be seen as
a combination of a noise image, which represents the characteristics of the
pixels in the scene, and a shadowing image, which describes the illumination
and shape of the objects in the scene. To model a specific scene image within
a certain range, nonlinear regression is applied to decompose the image into
its natural factors, allowing for the calculation of the inherent characteristics
of the scene59–64. This process includes the utilization of a weighting func­
tion to adjust the original estimate, leading to a more precise estimate that
reduces errors. Nonlinear regression can be applied to image patches without
encountering any issues with dimensionality. In such cases, the estimation of
the shadowing image involves modifying the image constraints in the most
effective manner. Moreover, the Retinex algorithm, which employs Poisson's
equation and secondary estimates, is commonly used to estimate the image
derivatives and enhance the accuracy of the original estimates.

1.3 RESULTS AND DISCUSSION

MedCalc is a fast and reliable statistical software that includes all features
and with more than 200 statistical tests, procedures, and graphs. Medical
calculators help physicians’ memory work and calculation skills to the
clinical test.
8 Computational Imaging and Analytics in Biomedical Engineering

Method 1: Seizure patient treatment based on cluster analysis

The cluster analysis is based on the treatment of seizure-affected persons and
their clinical proceedings are measured as shown in Table 1.1 and Figures
1.1 and 1.2 are given below

TABLE 1.1 Input Data of Seizure Patient Treatment Based on Cluster Analysis.
Gender Treatment Measurment 1 Measurement 2
Female A 21 25
Male B 22 26
To be continued

FIGURE 1.1 Cluster comparison graph.

FIGURE 2.2 Cluster multiple variable graph.

Statistical Analysis of Seizure Data to Support Clinical Proceedings 9

Method 2: Meta-analysis risk ratio and difference

Average assumption of seizure-treated positive and the control positive
patient’s details are discussed in Table 1.2 and Figure 1.3.

TABLE 1.2 Meta-Analysis Risk Ratio and Difference.

Seizure study Treated_positive Treated_Total Controls_positive Controls_total

Study 1 95 103 47 104

Study 2 119 127 34 129

Study 3 51 223 12 76

Study 4 122 139 61 142

Study 5 47 53 10 51

Study 6 121 135 29 68

Study 7 337 378 170 376

FIGURE 1.3 Meta-analysis risk ratio and difference.

10 Computational Imaging and Analytics in Biomedical Engineering

Method 3: Generic meta analysis

The variable for estimate and variable for standard error based on heteroge­
neity calculation, and the graph was shown in Figure 1.3.

TABLE 1.3 Generic Meta-Analysis.

Seizure Hazard_Ratio__Log_ SE_of_LOG_HR (Generic meta-analysis)

Study A 1 − 0.077 0.212

Study B 2 0.012 0.221

Study C 3 0.323 0.426

Study D 4 0.154 0.23

Study E 5 0.051 0.348

Study F 6 − 0.661 0.232

Study G 7 − 0.199 0.337

FIGURE 1.4 Meta-analysis generic invariance

Statistical Analysis of Seizure Data to Support Clinical Proceedings 11

FIGURE 1.5 Meta-analysis generic invariance.

Method 4: Meta-analysis continuous measure

The result of continuous measure of the meta-analysis is shown in Table 1.4
and the output in Table 1.5 and Figure 1.6.

TABLE 1.4 Input Data to Meta-Analysis Continuous Measure.

Seizure type_% FLE PLE OLE MTLE ME TOTAL

Subclinical 18 3 2 85 20.12 135

SPS 109 41 10 60 23.5 258

CPS 35 22 21 213 19.7 541

SGS 65 61 5 62 23.21 317

TOTAL 227 127 28 420 24.9 1251

FLE, frontal lobe epilepsy; PLE, peritial lobe epilepsy; OLE, occipital lobe epilepsy; MTLE,
mesial temporal lobe epilepsy; ME, mesial epilepsy.
12 Computational Imaging and Analytics in Biomedical Engineering

TABLE 1.5 Meta-Analysis: Continuous Measure-OUTPUT.

Variable for studies SeizureType_%
1. Intervention groups
Variable for number of cases FLE
Variable for mean PLE
Variable for SD OLE
2. Control groups
Variable for number of cases MTLE
Variable for mean ME
Variable for SD TOTAL
Test for heterogeneity
Q 1.2429
DF 4
Significance level P = 0.8710
I2 (inconsistency) 0.00%
95% CI for I2 0.00 to 37.00
Publication bias
Egger’s test
Intercept − 0.663
95% CI − 2.6639 to 1.3379
Significance level P = 0.3691
Begg’s test
Kendall’s Tau − 0.6
Significance level P = 0.1416

FIGURE 1.6 Meta-analysis continuous measure.

Notes: SPS, simple partial seizure; CPS, complex partial seizure; SGS, secondarily general­
ized seizure.
Statistical Analysis of Seizure Data to Support Clinical Proceedings 13

Method 5: Meta-analysis based on the correlation

The result of meta-analysis based on the correlation is shown in Table 1.7 as
referred from the literature survey and Figure 1.7.

TABLE 1.7 Meta-Analysis Based on the Correlation.

Seizure_study Correlation_coefficient Number of samples
Alickovic et al. 0.56 133
Fasil and Rajesh 0.43 149
Manzouri et al. 0.53 131
Parvez and Paul 0.51 120
Siddiqui et al. 0.66 111
Hussein et al. 0.46 152
Mursalin et al. 0.33 60

FIGURE 1.7 Meta-analysis based on correlation co-efficient.

Method 6: Meta-analysis risk ratio and difference

The result of meta-analysis based on risk ratio and difference are shown in
Table 1.8 and Figure 1.8.

TABLE 1.8 Meta-Analysis Based on Risk Ratio and Difference.

Study Treated_positive Treated_Total Controls_positive Controls_total
Study 1 95 103 47 104
Study 2 119 127 34 129
Study 3 51 223 12 76
14 Computational Imaging and Analytics in Biomedical Engineering

TABLE 1.8 (Continued)

Study Treated_positive Treated_Total Controls_positive Controls_total
Study 4 122 139 61 142
Study 5 47 53 10 51
Study 6 121 135 29 68
Study 7 337 378 170 376

FIGURE 1.8 Meta-analysis relative risk.

FIGURE 1.9 Meta-analysis relative difference.

Statistical Analysis of Seizure Data to Support Clinical Proceedings 15

Method 7: Non-linear regression method

The non-linear regression method is captured based on the dose given to the
patients and their responses.

TABLE 1.9 Meta-Analysis Based on Non-linear Regression Method.

Seizure dose Response

0 0.1

1.3 0.5

2.8 0.9

5 2.6

10.2 7.1

16.5 12.3

21.3 15.3

31.8 20.4

52.2 24.2

FIGURE 1.10 Scatter diagram of dose–response.

16 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 1.11 Non-linear regression dose–response.

1.4 CONCLUSION

Statistical analysis is very essential in the field of healthcare applications

core areas. However, meaningful and accurate data analyses are chal­
lenging and with errors and incorrect conclusions. This article presents the
radiographic medical image statistical analysis. The completely error-free
unbiased experimentation and handling of research-compromising pitfalls.
The statistical analysis will enhance the quality of clinical data for better
integration of imaging data with patient-specific.

KEYWORDS

• statistical analysis
• meta-analysis
• cluster analysis
• radiology

REFERENCES

1. Renukalatha, S.; Suresh, K. V.; A Review on Biomedical Image Analysis. Biomed. Eng.
App. Basis Commun. 2018.
2. Wang, W.; Krishnan, E. A Review on the State of the Science. Big Data Clin. JMIR Med.
Inform. 2014, 2, e1.
Statistical Analysis of Seizure Data to Support Clinical Proceedings 17

3. Luo, J.; Wu, M.; Gopukumar, D.; Zhao, Y. A Literature Review: Big Data Application in
Biomedical Research and Health Care. Biomed. Inform. Insights 2016, 8, 1–10.
4. Lambin, P.; Rios-Velazquez, E.; Leijenaar, R.; Carvalho, S.; van Stiphout, RGPM.;
Granton, P. Extracting More Information from Medical Images Using Advanced Feature
Analysis Radiomics. Eur. J. Cancer 2012, 48, 441–446.
5. van Griethuysen, J. J. M.; Fedorov, A.; Parmar, C.; Hosny, A.; Aucoin, N.; Narayan, V.
Radiomics System to Decode the Radiographic Phenotype Computational. Cancer
Res. 2017, 77, e104–e107.
6. Rusk, N. Deep Learning. Nat Methods 2015, 13, 35.
7. Huynh, E.; Coroller, T. P.; Narayan, V.; Agrawal, V.; Romano, J.; Franco, I. et al.
Associations of Radiomic Data Extracted from Static and Respiratory-Gated CT Scans
with Disease Recurrence in Lung Cancer Patients Treated with SBRT. PLoS One 2017,
12, e169–e172.
8. Kolossváry, M.; Kellermayer, M.; Merkely, B.; Maurovich-Horvat, P. A Comprehensive
Review on Radiomic Techniques: Cardiac Computed Tomography Radiomics. J.
Thorac. Imaging 2018, 33, 26–34.
9. O’Connor, J. P. B.; Aboagye, E. O.; Adams, J. E.; Aerts, H. J. W. L.; Barrington, S.
F.; Beer, A. et al. Imaging Biomarker Roadmap for Cancer Studies. Nat. Rev. Clin.
Oncol. 2017, 14, 169–186.
10. Quackenbush, J. Microarray Data Normalization and Transformation. Nat. Genet. 2002,
32, 496–501.
11. Leek, J. T.; Scharpf, R. B.; Bravo, H. C.; Simcha, D.; Langmead, B.; Johnson, W. E. et
al Tackling the Widespread and Critical Impact of Batch Effects in High-Throughput
Data. Nat. Rev. Genet. 2010, 11, 733–739.
12. Johnson, W. E.; Li, C.; Rabinovic, A. Adjusting Batch Effects in Microarray Expression
Data Using Empirical Bayes Methods. Biostatistics 2007, 8, 118–127.
13. Lee, M. L.; Kuo, F. C.; Whitmore, G. A.; Sklar, J. Importance of Replication in
Microarray Gene Expression Studies, Statistical Methods and Evidence from Repetitive
cDNA Hybridizations. Proc. Natl. Acad. Sci. USA 2000, 97, 9834–9839.
14. Neve, R. M.; Chin, K.; Fridlyand, J.; Yeh, J.; Baehner, F. L.; Fevr, T. et al. A Collection
of Breast Cancer Cell Lines for the Study of Functionally Distinct Cancer Subtypes.
Cancer Cell 2006, 10, 515–527.
15. Allison, D. B.; Cui, X.; Page, G. P.; Sabripour, M. From Disarray to Consolidation and
Consensus: Microarray Data Analysis. Nat. Rev. Genet. 2006, 7, 55–65.
16. Aerts, H. J. Data Science in Radiology—A Path Forward. Clin. Cancer Res. 2018, 24,
532–534.
17. Parmar, C.; Barry, J.; Hosny, D.; Ahmed, Q.; John, A.; Hugo, J. W. L. Data Analysis
Strategies in Medical Imaging. Clin. Cancer Res. 2018. clincanres.0385.2018.
18. Altman, D. G.; Gardner, M. J. Calculating Confidence Intervals for Regression and
Correlation. Br. Med. J. 1988, 296, 1238–1242.
19. Altman, D. G. Construction of Age-Related Reference Centiles Using Absolute
Residuals. Statistics Med. 1983, 12, 917–924.
20. Glantz, S. A.; Slinker, B. K.; Neilands, T. B. Primer of Applied Regression & Analysis
of Variance, 2nd ed.; McGraw- Hill.
21. Walter, S. D. The Partial Area Under the Summary ROC Curve. Stat. Med. 24,
2025–2040.
18 Computational Imaging and Analytics in Biomedical Engineering

22. Zweig, M. H.; Campbell, G. Receiver-Operating Characteristic (ROC) Plots: A

Fundamental Evaluation Tool in Clinical Medicine. Clin. Chem. 39:561–577.
23. Fraley, C.; Raftery, A. E. How Many Clusters? Which Clustering Method? Answers via
Model-Based Cluster Analysis. Comput. J. 1998,41 (8), 578–588.
24. Halkidi, M.; Batistakis, Y.; Vazirgiannis, M. On Clustering Validation Techniques. J.
Intell. Inf. Syst. 2001,17 (2–3), 107–145.
25. Strehl, A.; Ghosh, J.; Cardie, C.; Cluster Ensembles—A Knowledge Reuse Framework
for Combining Multiple Partitions. J. Mach. Learn. Res. 2002, 3, 583–617.
26. Fowlkes, E. B.; Mallows, C. L. A Method for Comparing Two Hierarchical Clusterings.
J. Am. Stat. Assoc. 1983, 78 (383), 553–569.
27. Lawrence, H.; Arabie, P. Comparing Partitions. J. Classification 1985, 2 (1), 193–218.
28. Jaccard, P. Nouvelles recherches sur la distribution florale. Bulletin de la Sociète
Vaudense des Sciences Naturelles. 1908, 44, 223–270.
29. Hirschberger, M.; Qi, Y.; Steuer, R. E. Randomly Generating Portfolio-Selection
Covariance Matrices with Specified Distributional Characteristics. Eur. J. Operat. Res.
2007, 177 (3), 1610–1625.
30. Amancio, D. R.; Comin, C. H.; Casanova, D.; Travieso, G.; Bruno, O. M.; Rodrigues,
F. A.; et al. A Systematic Comparison of Supervised Classifiers. Plos One 2014, 9 (4),
e94137. pmid:24763312.
31. Berkhin, P.; Kogan, J.; Nicholas, C.; Teboulle, M. A Survey of Clustering Data Mining
Techniques; Springer: Berlin Heidelberg, 2006; pp 25–71.
32. Hawkins, D. M. The Problem of Overfitting. J. Chem. Inf. Comput. Sci. 2004, 44 (1),
1–12. pmid:14741005.
33. R Development Core Team. R: A Language and Environment for Statistical Computing;
2006.
34. Mallett, S.; Halligan, S.; Thompson, M.; Collins, G. S.; Altman, D. G. Interpreting
diagnostic Accuracy Studies for Patient Care. BMJ 2012.
35. McGrath, T. A.; McInnes, M. D.; Korevaar, D. A.; Bossuyt, P. M. Meta-Analyses of
Diagnostic Accuracy in Imaging Journals: Analysis of Pooling Techniques and Their
Effect on Summary Estimates of Diagnostic Accuracy. Radiology 2016, 281 (1), 78–85.
36. Choudhri, A. F.; Siddiqui, A.; Khan, N. R.; Cohen, H. L. Understanding Bibliometric
Parameters and Analysis. Radiographics 2015, 35 (3), 736–746.
37. Frank, R. A; Sharifabadi, A. D.; Salameh, J. P.; McGrath, T. A.; Kraaijpoel, N.; Dang,
W.; et al. Citation Bias in Imaging Research: Are Studies with Higher Diagnostic
Accuracy Estimates Cited More Often? Eur. Radiol. 2019, 29 (4), 1657–1664.
38. Letchford, A.; Moat, H. S.; Preis, T. The Advantage of Short Paper Titles. R. Soc. Open
Sci. 2019,2 (8), 150266.
39. Analytics C. Journal Impact Factor 2018, Journal Citation Reports 2019.
40. Reitsma, J. B.; Glas, A. S.; Rutjes, A. W.; Scholten, R. J.; Bossuyt, P. M.; Zwinderman,
A. H. Bivariate Analysis of Sensitivity and Specificity Produces Informative Summary
Measures in Diagnostic Reviews. J. Clin. Epidemiol. 2015, 58 (10), 982–990.
41. Hillner, B. E.; Tosteson, A. N.; Song, Y.;Tosteson, T. D.; Onega, T.; Goodman, D. C.
et al Growth in the Use of PET for Six Cancer Types After Coverage by Medicare:
Additive or Replacement? J. Am. Coll. Radiol. 2012, 9 (1), 33–41.
42. Anderson, C. A.; Pettersson, F. H.; Barrett, J. C. et al. Evaluating the Effects of Imputation
on the Power, Coverage, and Cost Efficiency of Genome-Wide SNP Platforms. Am. J.
Hum. Genet. 2008, 83 (1), 112–119.
Statistical Analysis of Seizure Data to Support Clinical Proceedings 19

43. Seminara, D.; Khoury, M. J.; O’Brien, T. R. et al. The Emergence of Networks in Human
Genome Epidemiology: Challenges and Opportunities. Epidemiology 2007, 18, 1–8.
44. Ioannidis, J. P.; Bernstein, J.; Boffetta, P. et al. A Network of Investigator Networks in
Human Genome Epidemiology. Am. J. Epidemiol. 2005, 162, 302–304.
45. Rothstein, H. R.; Sutton, A. J.; Borestein, M., Eds. Publication Bias in Meta-Analysis–
Prevention, Assessment and Adjustments; Wiley: Chichester, 2005.
46. Barrett, J. C.; Cardon, L. R. Evaluating Coverage of Genome-Wide Association Studies.
Nat. Genet. 2006, 38 (6), 659–662.
47. Manolio, T. A.; Rodriguez, L. L.; Brooks, L. et al. GAIN Collaborative Research Group.
New Models of Collaboration in Genome-Wide Association Studies: The Genetic
Association Information Network. Nat. Genet. 2007, 39 (9), 1045–1051.
48. Hoggart, C. J.; Clark, T. G.; De Iorio, M.; Whittaker, J. C.; Balding, D. J. Genome-Wide
Significance for Dense SNP and Resequencing Data. Genet. Epidemiol. 2008, 32 (2),
179–185.
49. Ioannidis, J. P. Why Most Discovered True Associations Are Nflated. Epidemiology
2008, 19 (5), 640–648.
50. Pritchard, J. K.; Stephens, M.; Rosenberg, N. A.; Donnelly, P. Association Mapping in
Structured Populations. Am. J. Hum. Genet. 2000, 67, 170–181.
51. Ioannidis, J. P.; Patsopoulos, N. A.; Evangelou, E. Uncertainty in Heterogeneity
Estimates in Meta-Analyses. BMJ. 2007, 335, 914–916.
52. Moonesinghe, R.; Khoury, M. J.; Liu, T.; Ioannidis, J. P. Required Sample Size and
Non-Replicability Thresholds for Heterogeneous Genetic Associations. Proc. Natl.
Acad. Sci. U.SA 2008, 105 (2), 617–622.
53. Il’yasova, D.; Hertz-Picciotto, I.; Peters, U. et al. Choice of Exposure Scores for
Categorical Regression in Meta-Analysis: A Case Study of a Common Problem. Cancer
Causes Control 2005, 16, 383–388.
54. Greenland, S.; Longnecker, M. P. Methods for Trend Estimation from Summarized
Dose-Response Data, with Applications to Meta-Analysis. Am. J. Epidemiol. 1992, 135,
1301–1309.
55. Metlay, J. P.; Waterer, G. W.; Long, A. C.; Anzueto, A.; Brozek, J.; Crothers, K. et al.
Diagnosis and Treatment of Adults with Community-acquired Pneumonia. An Official
Clinical Practice Guideline of the American Thoracic Society and Infectious Diseases
Society of America. Am. J. Respir. Crit. Care Med. 2019, 200, e45–e67. PMID:31573350.
56. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation
for Statistical Computing: Vienna, Austria, 2019. http://www.R-project.org/
57. Mauri, L.; Orav, J.; O’Malley, A. J.; Moses, J. W.; Leon, M. Z. B.; Holmes, D. R.;
Teirstein, P. S.; Schofer, J.; Breithardt, G.; Cutlip, D. E.; Kereiakes, D. J.; Shi, C.; Firth,
B. G.; Donohoe, D. J.; Kuntz, R. Relationship of Late Loss in Lumen Diameter to
Coronary Restenosis in Sirolimus-Eluting Stents. Circulation. 2005, 111, 321–327.
58. O’Neill, J.; Young, J. B.; Pothier, C. E.; Lauer, M. S. Peak Oxygen Consumption as
a Predictor of Death in Patient with Heart Failure Receiving β-Blockers. Circulation.
2005, 111, 2313–2318.
59. Tappen, M. F.; Freeman, W. T.; Adelson, E. H. Recovering Intrinsic Images from a
Single Image. IEEE Trans. Pattern Analy. Mach. Intell. 2005.
60. Kimmel, R.; Elad, M.; Shaked, D.; Keshet, R.; Sobel, I. A Variational Framework for
Retinex. Int. J. Comput. Vision 2003, 52 (1), 7–23.
20 Computational Imaging and Analytics in Biomedical Engineering

61. Bell, W.; Freeman, W. T. Learning Local Evidence for Shading and Reflectance. In
Proc. Int. Conf. Comput. Vision 2001.
62. Freeman, E. T.; Pasztor, E. C.; Carmichael, O. T.; Learning Low-Level Vision. Int. J.
Comput. Vision 2000, 40 (1), 25–47.
63. Land, E. H.; McCann, J. J. Lightness and Retinex Theory. J. Optic. Soc. Am. 1971, 61,
1–11.
64. Barrow, H. G.; Tenenbaum, H. T. Recovering Intrinsic Scene Characteristics from
Images. In Computer Vision Systems; Hanson, A., Riseman, E., Eds.; Academic Press,
1978; pp 3–26.
65. https://www.researchgate.net/figure/Seizure-outcomes-by-epilepsy-etiology_
tbl2_239066212[DATA]
CHAPTER 2

SPATIAL PREPROCESSING IN
SEGMENTATION OF BRAIN MRI
USING T1 AND T2 IMAGES
S. JACILY JEMILA and A. BRINTHA THERESE
Vellore Institute of Technology, Chennai, India

ABSTRACT

Preprocessing is a very important process before segmentation in medical

image processing. To upgrade the characteristics of the image by diminishing
the noise, preprocessing is employed. In this chapter, T1 and T2 MRI images
of the brain are segmented before and after preprocessing using SPM-12
neuroimaging software, and the importance of preprocessing is discussed
using five important quality metrics: peak signal to noise ratio (PSNR),
mean squared error (MSE), edge preservation index (EPI), signal to noise
ratio (SNR), and structural similarity index metric (SSIM). We can conclude
that if we preprocess the MRI brain T1 and T2 images before segmentation,
we can reduce the MSE and increase the PSNR, SNR, and SSIM, which
are very important in the evaluation of medical images. EPI value is also
retained.

2.1 INTRODUCTION

Preprocessing in medical images is of considerable importance for the

following motives:

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
22 Computational Imaging and Analytics in Biomedical Engineering

a. The image feature ought to be strengthened.

b. Noise in the image does require elimination.
c. Images are to be higher fit for further computation.
d. Marks or labels existing in the images can intrude in the post­
processing of these images.
Nowadays, medical imaging plays a very important role in human lives.
In medical imaging, MRI is the most preferable technique for diagnosing
most medical issues because it has the following advantages:
1. MRI functions in the RF range.
2. No ionizing radiation present.
3. Can produce wonderful soft tissue contrast.
4. Can generate 3-D images.
5. Capability of producing images at any orientation (axial, sagittal,
and coronal).
MRI is a risk-free diagnostic and uncomplicated procedure to scan the
internal organs and structure of the human body through strong magnets and
radiowaves. An MRI scan is often used to examine the brain, spinal cord,
bones, joints, breast, heart, and blood vessels. For the past 40 decades, it has
been used as a significant tool in scientific research and medical diagnosis. A
commercial MRI imaging machine is displayed in Figure 2.1.

FIGURE 2.1 MRI imaging machine.

Spatial Preprocessing in Segmentation of Brain MRI 23

Segmentation of MRI is a significant step in medical image analysis

using MRI. It impacts the conclusion of the overall analysis. Manual
segmentation is a time-consuming method, and there is also an option for
error. Preprocessing is a very important step before segmentation. If we do
preprocessing,1,2 then we could improve the quality of the segmented parts.
To make the images more suitable for further analysis and to upgrade the
characteristics of the image by reducing the noise, preprocessing is very
helpful.

2.2 NOISE IN MAGNETIC RESONANCE IMAGES

Many image attainment processes (MRI, PET, SPECT, etc.) sustain image
degradation by noise. For MRI, the foremost seedbed of random noise is
thermal noise, which frames a statistically-independent random source
piercing the MR data in the time domain. Thermal noise is white and can
be depicted by a Gaussian random field with a zero mean and constant vari­
ance. Hence, the noise is not associated with the signal or with itself. Aside
from thermal noise, structured noise ordinarily demotes the image standard
as well as MR system attributes, physiological pulses, and object move­
ment. The features of noise depend on their source. Noise, inhomogeneous
pixel intensity allocation, and abrupt boundaries in the medical MR images
invoked by the MR data acquisition procedure are the capital problems that
influence the grade of MRI segmentation. One leading origin of noise is
the ambient electromagnetic field groomed up by the radiofrequency (RF)
detectors acquiring the MR signal, and the other is the object or body being
imaged.
In MRIs, raw data are constitutionally complex, valued, and spoiled with
zero-mean Gaussian-distributed noise with equal variance. Next to the inverse
Fourier transformation, the real and imaginary images are quite Gaussian-
distributed, habituated to the orthogonality and linearity of the Fourier
transform. MR magnitude images are framed by exclusively fetching the
square root of the sum of the square of the two self-reliant Gaussian random
variables (real and imaginary images) pixel by pixel. After this nonlinear
transformation, MR magnitude data can be proven to be Rician-distributed.
There is a correlation between SNR, acquisition time, and spatial resolu­
tion in MR images. The SNR is fairly dominant in huge MRI usages, and this
is cultivated implicitly and clearly by averaging. The MRI data obtainment
procedure can be performed by two averaging methods: (1) Spatial volume
24 Computational Imaging and Analytics in Biomedical Engineering

averaging is mandatory because of the discrete manner of the obtainment

procedure, and (2) In the case of few usages, the signal for the same k-space
location is obtained several times and averaged in order to minimize noise.
The two averaging methods are related. When a larger sampling rate of
the frequency domain is applied, improved resolution images are acquired.
However, in order to get the desired SNR at dominant spatial resolution, a
lengthy acquisition time is needed, as is the extra time required for aver­
aging. Contrarily, the acquisition time, with the subsequent SNR and the
imaging resolution, is practically bounded by patient comfort and system
throughput. Consequently, high SNR MRI images can be obtained at the
cost of constrained temporal for spatial resolution. Also, improved resolution
MRI imaging is obtainable at a cost of lower SNR for long acquisition times.
A further considerable cause of noise in MRI imaging is thermal noise in
the human body. Common MRI imaging concerns sampling in the frequency
domain (also called “k-space”) and employing an inverse discrete Fourier
transform. Signal measures have features in both real and imaginary chan­
nels, and each channel is strained by additive white Gaussian noise. Thus,
the complex reconstructed signal contains complex white additive Gaussian
noise. Because of phase errors, the magnitude of the MRI signal is ordinarily
preferred for the MRI image reconstruction. The magnitude of MRI signal is
real-valued and is applied for the image processing tasks as well as for visual
inspection.
The method by which the magnitude MRI image is rebuilt results in a
Rician distribution of noise. Since the Rician noise is signal-based, split­
ting the signal away from the noise is the most challenging process. In
large-intensity areas of the magnitude image, the Rician distribution can
be approached as a Gaussian distribution, and in low-intensity regions, it
can be represented as a Rayleigh distribution. A practical effect is a lowered
contrast of the MRI image; as the noise increases, the mean intensity values
of the pixels in low-intensity regions also increase. As established, it is a
reality that Rician noise demotes the MRI images in both qualitative and
quantitative senses, making image processing, interpretation, and segmenta­
tion more complex.

2.3 MAJOR CHALLENGES IN BIOMEDICAL IMAGE PROCESSING

It is challenging to integrate a medical image straightaway into the automatic

computation of image processing. This is named the semantic gap, which
Spatial Preprocessing in Segmentation of Brain MRI 25

implies the disparity between the cognitive interpretation of a diagnostic

image by the physician and the uncomplicated structure of discrete pixels,
which is exploited in computer programs to characterize an image. In the
medical domain, there are three main facets that hinder bridging this gap.

2.3.1 HETEROGENEITY OF IMAGES

Medical images demonstrate organs, living tissues, or body parts. Although

acquired with the same modality and pursuing a standardized shape, acquisi­
tion protocol, size, and internal structures, these objects might differ notably
not simply from patient to patient but also among various views of the same
patient and similar views of the same patient at different times. Thus, the
universal formulation of prior knowledge is impossible.

2.3.2 UNKNOWN DELINEATION OF OBJECTS

Biological structures cannot be extracted from the background because

the diagnostically related object is pictured in the overall image. Though
determinable objects are noticed in biomedical images, their segmentation is
difficult. Hence, they are segmented at the texture level.

2.3.3 ROBUSTNESS OF ALGORITHMS

Automatic analysis of images in medicine should be accurate. This is

followed as a rule in medical image processing. The images that cannot be
treated properly must be inevitably discarded and withdrawn from further
processing.

2.4 CONCEPT OF MRI

Our body comprises beyond 70% of water. The water molecule is in the form
of H2O, where H is the proton. This proton is an important component in
MRI. Based on the resonance property of the proton, only MRI images will
be made. All protons in our body normally rotate in random directions with
a Larmor frequency of 63.855 MHZ, as shown in Figure 2.2, and they are in
an out-of-phase condition.
26 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.2 Proton spin at Larmor frequency.

Source: Reprinted from Radiology Expert. https://www.radiology.expert/en/modules/mri­
technique-introduction/mri-technique/

When the patient is positioned in a magnetic field, that is, in an MRI

scanner, then all the protons will align in the z direction as in Figure 2.3,
which is at the side of the magnetic field. Even though all the protons are in
the z direction, they are in an out-of-phase condition.

FIGURE 2.3 Position of protons when the patient is positioned in an MRI scanner.

Now, if we apply some RF pulse with a frequency equal to Larmor’s

frequency, then excitation occurs, and all the protons are aligned in an
in-phase condition. The protons will change their position from the z plane
Spatial Preprocessing in Segmentation of Brain MRI 27

(longitudinal plane) to the xy plane (transversal plane). Now the longitudinal

magnetization will reduce and the trasversal magnetization will increase.
If we stop the RF excitation, then relaxation will take place. In relaxation,
the protons revert back to the longitudinal plane from the transversal plane,
which is given in Figure 2.4.

FIGURE 2.4 Alignment of protons in in-phase and out-phase conditions.

Two independent processes take place during relaxation.

1. Longitudinal relaxation (T1 relaxation)
2. Transversal relaxation (T2 relaxation)

2.4.1 T1 RELAXATION
When the RF pulse is removed, protons try to come to the longitudinal plane
from the transversal plane by giving some energy to the surroundings, so
now longitudinal magnetization increases. T1 is the time required to achieve
63% of the original longitudinal magnetization. This is the time required for
T1 imaging.

2.4.2 T2 RELAXATION
When we stop the RF pulse, not only do the protons try to come to the longi­
tudinal plane, but they also change to an out-of-phase condition. T2 is the
time required to dephase up to 37% of the original value. This is the time
required for T2 imaging.
For different tissues, there are different T1 and T2 timing and relaxation
curves.
28 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.5 T1 relaxation curve.

FIGURE 2.6 T2 relaxation curve.

Figures 2.5 and 2.6 show T1 and T2 relaxation curves for fat and water,
respectively. From the figures, we can observe that fat and water contain
individual relaxation curves. The T1 and T2 relaxation times of water are
high when compared to those of fat.
Spatial Preprocessing in Segmentation of Brain MRI 29

2.5 T1-WEIGHTED VERSUS T2-WEIGHTED IMAGES

FIGURE 2.7 T1 versus T2 MRI.

Even though many kinds of MRIs are available, T1-weighted and

T2-weighted are the two fundamental types of MRI images, often called T1
and T2 images, which are given in Figure 2.7. In T1 images, fat is white, and
in T2 images, fat and water are white. In T1-weighted images, cerebrospinal
fluid (CSF) is black, gray matter is gray, white matter is white, and bones
are black. But in the case of T2-weighted images, CSF is white, gray matter
is gray, white matter is dark gray, and bones are black. Table 2.1 shows the
different tissue colors in T1-weighted and T2-weighted images.

TABLE 2.1 Different Tissue Colors in T1-Weighted and T2-Weighted Images.

Tissue T1-weighted T2-weighted
CSF Dark Bright
White matter Light Dark gray
Cortex Gray Light gray
Fat Bright Light
Inflammation Dark Bright

2.6 SPATIAL PREPROCESSING

SPM-12 follows spatial preprocessing. The main goals of spatial prepro­

cessing are:
1. To match all scans of a particular subject.
2. To match scans of all subjects into standard space.
30 Computational Imaging and Analytics in Biomedical Engineering

The most significant steps are:

1. Realign
2. Coregister
3. Normalization
4. Smoothing.

2.6.1 REALIGN

Realignment is most significant in preprocessing steps. Variations due to

movement and shape differences between series of scans are rectified in this
step. In the realignment step, it will compare our input with the mean image
or first slice of the input. To correct these errors, rigid-body transformation8
is used. It will allow movement and rotation only in the x, y, and z directions.
It minimizes the sum of squared differences by trial and error method. It is
used only within modalities.

2.6.2 COREGISTER

Coregister is used to match scans of different modalities, that is, T1 and T2

images can be matched by using this step. In this step, T1 and T2 images are
matched using mutual information.6 If we try to match the sum of squared
differences, it will not give an output because the variation is high between
T1 and T2 images. The matching is obtained by comparing the sharpness of
the lines in the 2-D joint histogram. The joint histogram is formed by placing
the gray level of the first scan in the x-axis and the gray level of the second
scan in the y-axis, so two similar images would thus result in a diagonal line
from bottom left to upper right. Similar to realigning, this step also allows
only rigid-body transformation.

2.6.3 NORMALIZATION

This step is used to manipulate the scans into a standard space. SPM-12 uses
a unified segmentation procedure. It combines three steps.
1. Segmentation
2. Bias correction
3. Spatial normalization8
Spatial Preprocessing in Segmentation of Brain MRI 31

Segmentation is the process of separating different tissue classes. Bias

correction is the process of smoothly removing varying intensity differences
among images. Spatial normalization is achieved by using deformation fields.
Deformation fields are nothing but images used to quantify the displacement
at each location in 3-D space. The bright color indicates the position needs to
be shifted right, and the dark color indicates the location needs to be shifted
left.

2.6.4 SMOOTHING

Smoothing is obtained by taking the average of every voxel with a weighted

sum of its neighbors. The weighted sum is defined by the Gaussian kernel.
The size of the Gaussian kernel is given by the full width half maximum
(FWHM); the larger the FWHM, the more smoothing we can get. For MRI,
we use Gaussian with a FWHM that is twice the voxel size. The reason to
do smoothing is to correct slight functional anatomical differences between
subjects. The amount of smoothing should be determined by our application.
If our ROI is a very small area, then you cannot apply much smoothing
because it will affect resolution.

2.7 SEGMENTATION

Segmentation5 is a very important process in medical image analysis.

Even though many segmentation algorithms are available, segmentation
using SPM-127 software is very suitable for brain MRI segmentation.
Segmentation based on tissue classes is used here. It will give proper
segmentation of gray matter, white matter, CSF, the skull, and soft tissues
separately.

2.8 RESULTS AND DISCUSSIONS

If 49 slices of brain MRI in DICOM format of a 6-day-old baby are given to

the SPM-12 software as input, it will convert the images into two images in
Nifti format. One corresponds to T1 images, and the second one corresponds
to T2 images.
32 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.8 T1 image in DICOM format.

FIGURE 2.9 T2 image in DICOM format.

FIGURE 2.10 Nifti image from T1 images.

Spatial Preprocessing in Segmentation of Brain MRI 33

FIGURE 2.11 Nifti image from T2 images.

The input T1 and T2 images in DICOM format are given in Figures 2.8
and 2.9, and the input images in Nifti format are given in Figures 2.10 and
2.11.

FIGURE 2.12 Realign estimate.

34 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.13 Joint histogram.

FIGURE 2.14 Deformation field from T1 image.

Spatial Preprocessing in Segmentation of Brain MRI 35

FIGURE 2.15 Deformation field.

FIGURE 2.16 Smoothed Nifti image from T1 image.

36 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.17 Smoothed Nifti image from T1 and T2 images.

The realign estimate output is as given in Figure 2.12. The joint histogram
obtained during coregistration of T1 and T2 images is given in Figure 2.13.
Deformation fields generated during the normalization process for T1 and T2
images are given in Figures 2.14 and 2.15. The smoothed images are given
in Figures 2.16 and 2.17.

FIGURE 2.18 (a–e) Segmented brain parts from T1 image without preprocessing.

FIGURE 2.19 (a–e) Segmented brain parts from T2 image without preprocessing.
Spatial Preprocessing in Segmentation of Brain MRI 37

FIGURE 2.20 (a–e) Segmented brain parts from T1 image with preprocessing.

FIGURE 2.21 (a–e) Segmented brain parts from T2 image with preprocessing.

The segmented brain parts for T1 and T2 images without preprocessing

are as shown in Figures 2.18 (a–e) and 2.19 (a–e), and the segmented parts
for T1 and T2 images with preprocessing are given in Figures 2.20 (a–e) and
2.21 (a–e). The outputs are in Nifti format only; for easy display, they are
converted into DICOM using the Nifti to DICOM converter.

2.9 QUALITY METRICS

Image quality metrics3,4 are used to analyze the quality of the image. Quality
metrics are calculated for the segmented images, which are segmented
before and after preprocessing. If quality metrics are maintained well, then
only they are suitable for further analysis. MSE, PSNR, SNR, EPI, and SSIM
are calculated for the brain parts segmented with and without preprocessing
for T1 and T2 images.
The results are plotted as a graph as shown in Figure 2.22 (a–j), from
which we can understand the effect of preprocessing. From the above images,
we know that if we preprocess the MRI brain T1 and T2 images before
segmentation, we can reduce the MSE and increase the PSNR, SNR, and
SSIM, which are very important in the analysis of medical images. The EPI
value is also maintained well. For example, the MSE for white matter in T1
is 653.2879 without preprocessing and 609.7175 with preprocessing. PSNR
for gray matter in T1 is 20.1991 without preprocessing and 20.3984 with
preprocessing. SNR for white matter in T1 is 4.3804 without preprocessing
and 8.5689 with preprocessing. EPI for gray matter in T1 is 0.7253 without
preprocessing and 0.6864 with preprocessing. The SSIM for white matter in
T1 is 0.9784 without preprocessing and 0.9913 with preprocessing.
38 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 2.22 (a–j) Quality metrics comparison for T1 and T2 images with and without
preprocessing.
Spatial Preprocessing in Segmentation of Brain MRI 39

2.10 CONCLUSIONS

In this chapter, T1 and T2 MRI brain images are segmented before and after
preprocessing. Realignment, coregistration, normalization, and smoothing
are the spatial preprocessing tools applied. With and without preprocessing,
MSE, PSNR, SNR, EPI, and SSIM values are calculated for the segmented
brain parts, and the results are analyzed. From the results, we can conclude
that preprocessing is very important before the segmentation of the brain
MRI.

KEYWORDS

• preprocessing
• segmentation
• MSE
• SNR
• PSNR
• EPI
• SSIM

REFERENCES

1. Gonzalez, R. G.; Woods, R. E. Third Edition Digital Image Processing; Pearson: South
Asia, 2014.
2. Sridhar, S. Digital Image Processing; Oxford: India, 2011.
3. Najarian, K.; Splinter, R. Biomedical Signal and Image Processing; Taylor & Francis:
New York, 2006.
4. Demirkaya, O.; Asyali, M. H.; Sahoo, P. K. Image Processing with MATLAB®
Applications in Medicine and Biology; CRC Press: New York, 2009.
5. Semmlow, J. L. Biosignal and Medical Image Processing; CRC Press: New York, 2009.
6. Josien, P. W. P.; Antoine Maintz, J. B.; Viergever, M. A. Mutual-Information-Based
Registration of Medical Images: A Survey. IEEE Trans. Med. Imaging 2003, 22 (8),
986–1004.
7. Eklund, A.; Nichols, T.; Andersson, M.; Knutsson, H. Empirically Investigating the
Statistical Validity of SPM, FSL and AFNI for Single Subject FMRI Analysis. IEEE
2015, 1376–1380.
8. Muro, A.; Zapirain, B. G.; Méndez, A.; Ruiz, I.; FMRI Processing Tool for the Analysis,
Parametrisation and Comparison of Preprocessed SPM Images. In 18th European Signal
Processing Conference 2010; pp 1335–1339.
CHAPTER 3

COMPARATIVE VOLUME ANALYSIS OF

PEDIATRIC BRAIN WITH ADULT BRAIN
USING T1 MRI IMAGES
S. JACILY JEMILA and A. BRINTHA THERESE
VIT University, Chennai, India

ABSTRACT

Segmentation of the pediatric brain is very useful in medical image analysis

to supervise the brain’s growth and development in infants. Atlas-based
methods are used for automatic segmentation of a brain MRI. In this chapter,
automatic segmentation of four pediatric and three adult brains is done using
Brainsuite 19a brain imaging software, and the volumes of different brain
parts are computed.

3.1 INTRODUCTION

Automatic segmentation8 is a very tough task in pediatric brains due to the

low SNR, partial volume effects, and large changes in the shape and appear­
ance of the brain. Manual segmentation requires more time and cost, and the
probability of error is also higher due to the variation of results from person
to person. So there is a need for an accurate automatic segmentation tool.
Brainsuite1319a is such a type of automatic segmentation software that can
segment the brain into 170 parts.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
42 Computational Imaging and Analytics in Biomedical Engineering

3.2 MATERIALS AND METHODS

3.2.1 DATASETS

Real-time datasets of seven different patients are considered for this analysis.
Among the seven, four are babies and three are adults. Real-time data that
are in DICOM format are converted to Nifty format by using SPM-12, a
neuroimaging software. Real-time datasets at the following ages are consid­
ered for volume analysis:
1. 6-day-old baby
2. 2-month-old baby 1
3. 2-month-old baby 2
4. 2 months, 20 days old baby
5. 42-year old adult
6. 46 years, 7 months 13 days old adult
7. 47-year-old adult

3.2.2 HUMAN BRAIN’S IMPORTANT PARTS AND FUNCTIONS

The surface area of the brain is around 233–465 square inches (1500–2000
cm2). To fit this surface area to the skull, the cortex is folded, and sulci
(grooves) and gyri (folds) are formed. The cerebral cortex, or cortical
surface, is divided into four major lobes. They are the frontal lobe, parietal
lobe, occipital lobe, and temporal lobe. Each lobe has a different function.
The brain is divided into left and right halves using the interhemispheric
fissure, which is a large groove. The corpus callosum is used to facilitate
communication between these two halves. Also, the right and left temporal
lobes communicate through the anterior commissure, which is a tract of fiber.
The cortical area above the corpus callosum is divided by a groove called the
cingulate sulcus. The area between the groove and the corpus callosum is
called the cingulate gyrus.
Some important areas of the brain and their functions:
• Parietal lobe—it will receive and respond to somatosensory input
(pain and touch).
• Frontal lobe—it is involved in motor skills (including speech) and
cognitive functions.
• Occipital lobe—it receives and processes visual information directly
from the eyes.
Comparative Volume Analysis of Pediatric Brain 43

• Temporal lobe—it processes auditory information from the ears.

• Basal ganglia—located within the temporal lobe, coordinate fine
motions, such as fingertip movements.
• Limbic system—important in emotional behavior and controlling
movements of muscles of the digestive tract and body cavities.
• Hippocampus—it is important for short-term memory.
Amygdala—controls social and sexual behavior and other emotions.
Insula—influences automatic functions of the brainstem, processes taste
information, and separates the temporal and frontal lobes.

3.2.3 CORTICAL SURFACE EXTRACTION

Cortical surface extraction13 is the first step in volume calculation using

Brainsuite 19a. The cortical surface is the exterior surface of the brain. Some
important steps in cortical surface extraction are skull stripping, nonunifor­
mity correction, tissue classification, cerebrum labeling, topology correction,
and wisp removal.

3.2.3.1 SKULL STRIPPING

The first step in many of the MRI analysis sequences is the skull stripping.9
In T1 images, the edges between the skull and brain regions are well-defined.
The brain surface extractor (BSE) is used to extract the brain region in
Brainsuite 19a. It performs the following steps: anisotropic diffusion, edge
detection, and morphological operations for brain surface extraction. Aniso­
tropic diffusion applies smoothing to lower contrast edges while retaining
the higher contrast edges. For edge detection, it uses the Marr–Hildreth
edge detection operator, which applies Gaussian blur followed by a Lapla­
cian operator. In morphological operations, first erosion is applied and then
dilation is applied; these are used to eliminate the noise-related connections
between the scalp and brain.

3.2.3.2 NONUNIFORMITY CORRECTION

The intensity of nonuniformity is always present in most of the MRI images

due to variation in the magnetic field. To correct this problem, bias field
44 Computational Imaging and Analytics in Biomedical Engineering

correction (BFC) is applied. BFC estimates the correction field based on the
tissue gain variation.

3.2.3.3 TISSUE CLASSIFICATION

This is done by a partial volume classifier (PVC). PVC assigns integer

values for every voxel based on the tissue type. It considers the following
tissue types: background voxels, cerebrospinal fluid, gray matter, and white
matter. PVC models the brain as a set of piecewise continuous regions of
single tissue types bounded by partial volume contributions. In this method,
the neighborhood of voxels should be similar.

3.2.3.4 CEREBRUM LABELING

Once the brain region is extracted and classified, then it is separated into
cerebellum, cerebrum, and other structures to enable identification of the
cerebral cortex. To achieve this, volumetric registration is needed for an MRI
image.

3.2.3.5 TOPOLOGY CORRECTION

The boundary of the cortex should be topologically a sphere. It should not

have any holes or handles. But because of some segmentation errors, there
may be some holes or handles. A graph-based algorithm is applied to force
the segmented voxels to have a spherical topology.

3.2.3.6 WISP REMOVAL

The wisp removal step is used to remove the segmentation errors in the brain
extraction process by removing thin, wispy structures. It decomposes the
binary mask into a graph, separating weakly connected components.

3.2.4 SURFACE/VOLUME REGISTRATION

Surface and volume registration14 (SvReg) is used to coregister our brain

image with a standard atlas. For coregistration, it uses anatomical information
Comparative Volume Analysis of Pediatric Brain 45

present in the image. SvReg is very helpful for regional analysis. SvReg uses
a series of multistep registration and refinement processes. We use Brainsuit
atlas1 for SvReg. It is based on the Colin 27 atlas. It is an average of 27 scans
of an individual.

3.2.5 VOLUME ESTIMATION

SvReg of Brainsuite 19a gives a cortical surface with segmented regions.

Each segmented region is differentiated by a different color. We can check
the volume for every segmented region by clicking on the ROI details.

3.3 RESULTS AND DISCUSSION

FIGURE 3.1 Cortical surface extraction output for an adult.

FIGURE 3.2 Cortical surface extraction output for a baby.

46 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 3.3 SvReg output of a 6-day-old baby: (a) front view and (b) back view.

FIGURE 3.4 SvReg output of a 2-month-old baby 1: (a) front view and (b) back view.

FIGURE 3.5 SvReg output of a 2-month-old baby 2: (a) front view and (b) back view.
Comparative Volume Analysis of Pediatric Brain 47

FIGURE 3.6 SvReg output of a 2 months 20 days old baby: (a) front view and (b) back
view.

FIGURE 3.7 SvReg output of a 42 years adult: (a) front view and (b) back view.

FIGURE 3.8 SvReg output of a 46 years, 7 months 13 days adult: (a) front view and (b)
back view.
48 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 3.9 SvReg output of a 47 years adult: (a) front view and (b) back view.

The outputs of cortical surface extraction for an adult and a baby are
given in Figures 3.1 and 3.2. When compared to a baby, there are some
curves on the surface. The outputs for SvReg are given in Figures 3.3–3.9
for different datasets. Different colors in the image indicate different regions
of the brain. Volumes corresponding to these different regions are given in
Table 3.1.

TABLE 3.1 Volumes Corresponding to Different Brain Regions for Different Datasets.
Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47
no baby1 months days baby months 13 years years
baby 2 days adult adult adult
1 Background 2099.84814 3589.632 3745.7 2766.4321 2622.3 2784.9 2601.41
2 R. caudate 7.79200029 1.568 0.648 10.72 3.176 7.416 2.304
nucleus
3 L. caudate 13.3520002 1.6 1.352 10.68 1.088 1.528 5.808
nucleus
4 R. putamen 7.46400023 4.648 2.992 6.1040001 2.896 3.688 4.736
5 L. putamen 5.83200026 3.528 2.616 4.48 2.904 2.416 4.96
6 R. globus 9.54400063 3.2 4.12 6.8480005 1.192 2.464 3.056
pallidus
7 L. globus 8.03200054 2.192 4.16 5 1.2 1.528 3.384
pallidus
8 R. nucleus 0.712000012 0.248 0.272 0.288 0.224 0.112 0.16
accumbens
9 L. nucleus 0.624000013 0.28 0.192 0.184 0.128 0.192 0.344
accumbens
Comparative Volume Analysis of Pediatric Brain 49

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
10 R. 0.880000055 0.456 0.32 0.7920001 1.256 0.512 0.688
claustrum
11 L. claustrum 1.39200008 0.624 0.472 1.0400001 0.608 0.888 1.016
12 R. thalamus 76.776001 37.464 28.296 64.632004 18.048 14.944 14.088
13 L. thalamus 85.7360077 36.048 27.4 57.816002 19.656 23.888 19.856
14 R. basal 4.71200037 1.848 1.584 1.7600001 1.656 1.192 1.872
forebrain
15 L. basal 4.98400021 3.024 1.592 1.7360001 1.616 1.464 2.016
forebrain
16 R. lateral 0.296000004 0.064 0.12 0.168 0.328 0.104 0.2
geniculate
nucleus
17 L. lateral 0.168000013 0.152 0.12 0.128 0.224 0.208 0.072
geniculate
nucleus
18 R. medial 0.248000011 0.136 0.144 0.176 0.208 0.04 0.232
geniculate
nucleus
19 L. medial 0.280000001 0.064 0.12 0.064 0.04 0.152 0.104
geniculate
nucleus
20 R. superior 0.440000027 0.048 0.048 0.032 0.408 0.216 0.12
colliculus
21 L. superior 0.192000002 0.04 0.088 0.144 0.28 0.328 0.08
colliculus
22 R. inferior 0.320000023 0.016 0.032 0.024 0.264 0.232 0.08
colliculus
23 L. inferior 0.104000002 0.024 0.032 0.008 0.408 0.376 0.208
colliculus
24 R. mamil­ 0.256000012 0.048 0 0.016 0.016 0.408 0.056
lary body
25 L. mamil­ 0 0 0 0 0.024 0.136 0
lary body
26 L. 4.11199999 0 0 3.6240001 0 0 0
ventricular
system
27 R. lateral 1.39200008 0.04 0 2.296 2.128 5.776 3.4
ventricle
28 Third 3.57600021 2.168 0.872 1.9840001 3.264 4.424 6.312
ventricle
50 Computational Imaging and Analytics in Biomedical Engineering

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
29 Brainstem 49.4880028 23.776 39.648 31.504002 49.424 49.376 56.256
30 Cerebellum 171.544006 83.464 207.01 83.952004 171.64 156.66 268.704
31 R. superior 22.2480011 14.048 6.144 47.392002 34.928 34.24 11.592
frontal
gyrus (gm)
32 L. superior 24.3520012 13.328 6.184 45.016003 59.192 61.568 89.08
frontal
gyrus (gm)
33 R. middle 19.8800011 6.344 6.36 21.912001 23.688 12.072 5.568
frontal
gyrus (gm)
34 L. middle 24.2640018 9.200001 1.968 15.376 41.744 37.96 32.088
frontal
gyrus (gm)
35 R. pars 8.27200031 3.896 2.28 12.664001 15.024 4 4.624
opercularis
(gm)
36 L. pars 5.50400019 4.4 1.056 6.3360004 3.704 4.904 10.856
opercularis
(gm)
37 R. pars 15.9040003 5.992 6.208 17.584002 9.704 0.856 2.328
triangularis
(gm)
38 L. pars 10.2160006 3.288 4.008 20.968 6.968 24.456 21.584
triangularis
(gm)
39 R. pars 2.08800006 0 2.36 0 0.288 0 0.016
orbitalis
(gm)
40 L. pars 2.00800014 0 0.912 0.784 0.744 1.496 1.424
orbitalis
(gm)
41 R. 16.2880001 12.768 5.24 20.576 81.44 21.224 13.16
precentral
gyrus (gm)
42 L. precentral 16.9280014 13.544 9.736 20.808001 78.952 22.672 31.552
gyrus (gm)
43 R. 0.30400002 0 0.344 1.6 0.976 0.064 0.008
transverse
frontal
gyrus (gm)
Comparative Volume Analysis of Pediatric Brain 51

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
44 L. 0.552000046 0 0.136 0.424 0.392 0.928 1.136
transverse
frontal
gyrus (gm)
45 R. gyrus 5.74400043 0.112 2.768 2.9200001 1.776 1.056 1.712
rectus (gm)
46 L. gyrus 13.8160009 0.4 2.088 2.9360001 2.368 2.08 7.016
rectus (gm)
47 R. middle 2.00800014 0 0.112 0.136 0.064 0.056 0
orbitofrontal
gyrus (gm)
48 L. middle 2.72800016 0.056 1.032 0.536 0.152 0.688 3.224
orbitofrontal
gyrus (gm)
49 R. anterior 0.856000066 0 0.08 0.312 0.608 0.016 0.248
orbitofrontal
gyrus (gm)
50 L. anterior 1.1680001 0 0.224 0.096 0.088 0.296 1.656
orbitofrontal
gyrus (gm)
51 R. posterior 1.46400011 0 2.152 1.192 0.472 0.264 1.616
orbitofrontal
gyrus (gm)
52 L. posterior 2.28000021 0.152 3.344 2.4160001 0.176 2.008 4.768
orbitofrontal
gyrus (gm)
53 R. lateral 0 0 0 0 0.344 0 0
orbitofrontal
gyrus (gm)
54 L. lateral 3.16800022 0.008 1.784 0.016 0.088 0.168 0.824
orbitofrontal
gyrus (gm)
55 R. 0 0.504 0.52 0.7040001 0.816 1.448 5.032
paracentral
lobule (gm)
56 L. 2.53600001 0.816 6.856 0 0.184 1.24 1.272
paracentral
lobule (gm)
57 R. cingulate 16.3439999 9.072001 1.76 34.816002 60.432 31.136 39.888
gyrus (gm)
58 L. cingulate 34.8160019 16.592 4.384 35.400002 21.936 99.088 70.808
gyrus (gm)
52 Computational Imaging and Analytics in Biomedical Engineering

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
59 R. subcal­ 2.57600021 0.552 1.064 1.9680001 0.768 0 0
losal gyrus
(gm)
60 L. subcal­ 2.83200002 0.072 0.184 0.368 0.056 0.552 0
losal gyrus
(gm)
61 R. 18.6880016 11.048 3.928 17.856001 8.056 20.696 31.544
postcentral
gyrus (gm)
62 L. 20.2000008 12.728 8.264 28.672001 41.288 70.304 52.088
postcentral
gyrus (gm)
63 R. 13.5360003 12 6.32 16.296001 87.256 32.552 19.304
supramar­
ginal gyrus
(gm)
64 L. supra­ 20.4160004 9.712001 8.52 11.816001 23.104 78.728 57.832
marginal
gyrus (gm)
65 R. angular 28.5040016 12.784 5.808 18.240002 20.248 23.864 3.112
gyrus (gm)
66 L. angular 10.9360008 4.576 4.76 7.5360003 47.304 8.6 40.72
gyrus (gm)
67 R. superior 14.3840008 5.08 3.192 8.1280003 9.88 32.848 51.376
parietal
gyrus (gm)
68 L. superior 20.0400009 8.496 3.472 9.0880003 41.528 5.272 9.896
parietal
gyrus (gm)
69 R. 0.024 0.816 0.952 6.9440002 6.728 1.128 1.072
precuneus
(gm)
70 L. 0.448000014 0.368 0.04 1.184 2.624 1.408 2.328
precuneus
(gm)
71 R. temporal 6.19200039 0 1.072 1.368 28.688 2.16 8.76
pole (gm)
72 L. temporal 2.84800005 0 1.04 1.2160001 0.392 0.096 1.416
pole (gm)
73 R. superior 12.6560001 5.056 8.24 11.568001 38.688 18.944 13.832
temporal
gyrus (gm)
Comparative Volume Analysis of Pediatric Brain 53

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
74 L. superior 12.368001 3.92 3 4.5360003 13.432 8.84 23.16
temporal
gyrus (gm)
75 R. 2 0.568 0.384 1.016 2.096 5.88 1.584
transverse
temporal
gyrus (gm)
76 L. 1.50400007 0.968 0.512 0.776 1.352 1.368 10.096
transverse
temporal
gyrus (gm)
77 R. middle 30.1680012 3.888 11.584 9.1440001 30.488 12.872 15.2
temporal
gyrus (gm)
78 L. middle 33.0320015 5.44 8.248 10.024 34.848 31.712 59.792
temporal
gyrus (gm)
79 R. inferior 22.0480003 0.44 6.32 0.488 26.632 26.184 8.096
temporal
gyrus (gm)
80 L. inferior 23.6400013 0.28 7.896 1.0320001 36.888 27.368 25.792
temporal
gyrus (gm)
81 R. fusiforme 2.15200019 0 0 0.032 0.688 5.256 6.448
gyrus (gm)
82 L. fusiforme 3.0480001 0 0.032 0.024 0.824 2.384 8.112
gyrus (gm)
83 R. parahip­ 0 0 0 0 0.032 0.192 0.368
pocampal
gyrus (gm)
84 L. parahip­ 0 0 0 0 0.408 0.968 0.592
pocampal
gyrus (gm)
85 R. hippo- 0.472000033 2.232 0.392 0.168 1.464 1.28 1.648
campus
(gm)
86 L. hippo- 0.248000011 1.248 0 0.44 1.56 1.576 0.752
campus
(gm)
87 R. amygdala 0.344000012 0.712 0.352 0.328 0.176 0.664 0.072
(gm)
88 L. amygdala 0.527999997 1.12 0.048 0.272 0.344 0.488 0.912
(gm)
54 Computational Imaging and Analytics in Biomedical Engineering

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
89 R. superior 2.28000021 0.688 0.568 3.3200002 0.336 0.64 0.92
occipital
gyrus (gm)
90 L. superior 0.104000002 0.352 0.096 0.6160001 0.672 0.864 0.928
occipital
gyrus (gm)
91 R. middle 5.18400002 3.016 0.392 6.5120001 2.592 3.184 0.744
occipital
gyrus (gm)
92 L. middle 3.88800025 1.68 0.464 4.1680002 11.888 3.752 6.824
occipital
gyrus (gm)
93 R. inferior 3.08000016 0.808 0.336 1.6160001 21.608 10.32 1.336
occipital
gyrus (gm)
94 L. inferior 0.088000007 0.016 0.08 0.288 3.776 3.968 1.888
occipital
gyrus (gm)
95 R. lingual 2.08800006 0.624 0.4 1.5520001 6.576 1.008 12.064
gyrus (gm)
96 L. lingual 0.056000002 0.128 0.312 0.152 3.76 2.088 6.848
gyrus (gm)
97 R. cuneus 2.60000014 3.008 0.728 1.192 2.992 11.656 7.744
(gm)
98 L. cuneus 0.232000008 0.744 1.208 0.44 0.904 7.528 5.072
(gm)
99 R. insula 9.00800037 1.488 6.416 10.512 5.632 9.696 7.128
(gm)
100 L. insula 9.84000015 1.992 4.152 10.8 10.144 9.352 6.568
(gm)
101 White 1082.71204 590.944 384.35 753.28803 300.67 378.38 363.76
matter
(cerebrum)
102 R. superior 8.40799999 5.192 1.984 37.224003 12.176 13.624 9.24
frontal
gyrus (wm)
103 L. superior 8.3760004 4.12 2.472 48.224003 23.376 39.424 53.192
frontal
gyrus (wm)
104 R. middle 5.96000051 2.704 2.952 8.1920004 10.864 7.016 2.816
frontal
gyrus (wm)
Comparative Volume Analysis of Pediatric Brain 55

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
105 L. middle 5.67200041 3.24 1.648 8.5520001 18.184 20.752 15.744
frontal
gyrus (wm)
106 R. pars 2.28800011 1.328 1.432 3.1120002 7.928 0.8 0.752
opercularis
(wm)
107 L. pars 2.3360002 1.44 2.336 2.664 3.496 1.344 4.696
opercularis
(wm)
108 R. pars 5.44000006 1.648 3.704 6.2960005 0.936 1.144 1.056
triangularis
(wm)
109 L. pars 3.03200006 0.768 1.304 5.0640001 3.912 14.76 8.816
triangularis
(wm)
110 R. pars 0.088000007 0 0.624 0 0 0 0
orbitalis
(wm)
111 L. pars 0.256000012 0 0.048 0.112 0 0.264 0.272
orbitalis
(wm)
112 R. 9.20800018 6.208 2.48 13.696001 41.288 8.464 7.648
precentral
gyrus (wm)
113 L. precentral 12.1760006 6.72 2.528 21.224001 54.16 6.008 23.184
gyrus (wm)
114 R. 0.048 0.04 0.272 0.48 0.32 0 0
transverse
frontal
gyrus (wm)
115 L. 0.184000015 0 0.16 0.064 0.096 0.112 0.296
transverse
frontal
gyrus (wm)
116 R. gyrus 3.72000027 0.04 0.408 0.448 0.184 0.008 0.048
rectus (wm)
117 L. gyrus 5.61600018 0.216 0.16 0.536 0.248 0.768 1.192
rectus (wm)
118 R. middle 0.576000035 0 0 0 0 0 0
orbitofrontal
gyrus (wm)
56 Computational Imaging and Analytics in Biomedical Engineering

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
119 L. middle 1.44800007 0.016 0.312 0.032 0.016 0.112 0.616
orbitofrontal
gyrus (wm)
120 R. anterior 0.064000003 0.016 0.32 0 0.176 0 0
orbitofrontal
gyrus (wm)
121 L. anterior 0.368000031 0 0.032 0 0 0.016 0.144
orbitofrontal
gyrus (wm)
122 R. posterior 0.160000011 0 0.144 0.04 0 0 0
orbitofrontal
gyrus (wm)
123 L. posterior 0.45600003 0 0.072 0.072 0 0.016 0.352
orbitofrontal
gyrus (wm)
124 R. lateral 0.30400002 0.008 0.16 0 0.336 0 0
orbitofrontal
gyrus (wm)
125 L. lateral 0.496000022 0.008 0.096 0 0.024 0.096 0.216
orbitofrontal
gyrus (wm)
126 R. 2.18400002 0.296 0.976 2.3760002 1.064 0.48 3.176
paracentral
lobule (wm)
127 L. 10.9520006 0.92 0.224 0.936 0.216 0.952 1.888
paracentral
lobule (wm)
128 R. cingulate 6.64800024 3.968 0 37.512001 28.912 13.512 19
gyrus (wm)
129 L. cingulate 93.3280029 8.848001 2.112 62.952004 15.808 62.272 43.568
gyrus (wm)
130 R. subcal­ 0.84800005 0.336 0.496 0.8800001 0.112 0 0
losal gyrus
(wm)
131 L. subcal­ 1.24800003 0.056 0.072 0.592 0.008 0.12 0
losal gyrus
(wm)
132 R. 12.2720003 5.688 1.432 13.384001 6.136 11.08 20.376
postcentral
gyrus (wm)
133 L. 17.8400002 8.280001 6.776 21.192001 26.128 46.936 37.944
postcentral
gyrus (wm)
Comparative Volume Analysis of Pediatric Brain 57

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
134 R. 9.14400005 6.976 3.592 10.264001 42.072 13.144 11.456
supramar­
ginal gyrus
(wm)
135 L. 25.6240005 6.528 5.624 8.3440008 10.76 52.856 42.232
supramar­
ginal gyrus
(wm)
136 R. angular 20.8480015 4.192 1.32 2.7840002 14.216 9.648 1.368
gyrus (wm)
137 L. angular 18.960001 1.608 0.536 1.8160001 17.928 4.312 25.992
gyrus (wm)
138 R. superior 20.6640015 4.36 2.456 11.152 5.088 17.888 23.824
parietal
gyrus (wm)
139 L. superior 47.5600014 4.984 1.816 12.504001 24.728 1.528 5.424
parietal
gyrus (wm)
140 R. 3.17600012 1.36 0.464 7.3520002 1.064 0.856 1.088
precuneus
(wm)
141 L. 1.37600005 0.656 0.04 1.4080001 3.024 1.04 1.128
precuneus
(wm)
142 R. temporal 0.224000007 0 0.072 0.5840001 9.048 0.32 0.384
pole (wm)
143 L. temporal 0.184000015 0 0.456 0.48 0.016 0 0.264
pole (wm)
144 R. superior 6.54400015 1.656 2.848 2.6960001 17.456 8.168 7.224
temporal
gyrus (wm)
145 L. superior 6.6880002 1.584 1.16 1.0400001 5.336 4.984 16.664
temporal
gyrus (wm)
146 R. 0.968000054 0.256 0.576 0.24 1.152 1.304 0.688
transverse
temporal
gyrus (wm)
147 L. 1.2240001 0.528 0.328 0.216 0.152 0.984 7.656
transverse
temporal
gyrus (wm)
58 Computational Imaging and Analytics in Biomedical Engineering

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
148 R. middle 5.87200022 2.104 3.688 4.5520001 11.144 1.712 4.824
temporal
gyrus (wm)
149 L. middle 7.14400053 1.76 2.68 3.0640001 6.624 16.08 28.4
temporal
gyrus (wm)
150 R. inferior 8.68000031 1.256 2.816 3.3760002 10.048 7.856 3.328
temporal
gyrus (wm)
151 L. inferior 4.60000038 0.688 3.352 2.296 10.928 8.928 7.664
temporal
gyrus (wm)
152 R. fusiforme 7.96800041 0.456 2.8 2.1920002 2.048 5.328 2.568
gyrus (wm)
153 L. fusiforme 4.4000001 0.592 2.304 1.8480001 3.328 4.272 4.912
gyrus (wm)
154 R. parahip­ 0.504000008 0 0.072 0.016 0.248 0 2.456
pocampal
gyrus (wm)
155 L. parahip­ 0.496000022 0.136 0.456 0.064 1.768 0.312 1.144
pocampal
gyrus (wm)
156 R. hippo­ 1.12800002 0.464 1.16 1.6 0.512 1.2 4.24
campus
(wm)
157 L. hippo­ 1.91200006 2.152 0.4 0.896 0.616 0.752 1.136
campus
(wm)
158 R. amygdala 0.600000024 0.256 0.232 0.7600001 0 0.488 0.568
(wm)
159 L. amygdala 0.520000041 0.128 0.48 0.92 0 0.008 0.248
(wm)
160 R. superior 4.96000004 0.944 1.008 3.0240002 0.088 0.472 2.16
occipital
gyrus (wm)
161 L. superior 1.48800004 0.336 0.24 0.32 1.424 0.608 1.096
occipital
gyrus (wm)
162 R. middle 10.0400009 1.656 1.192 3.0480001 6.048 3.432 0.808
occipital
gyrus (wm)
Comparative Volume Analysis of Pediatric Brain 59

TABLE 3.1 (Continued)

Sl. Brain parts 6 days baby 2 months 2 2 month 20 46 years, 7 42 47

no baby1 months days baby months 13 years years
baby 2 days adult adult adult
163 L. middle 3.84000015 1.512 0.376 3.2640002 12.752 1.84 5.4
occipital
gyrus (wm)
164 R. inferior 7.04800034 0.856 1.024 2.776 6.048 4.056 0.288
occipital
gyrus (wm)
165 L. inferior 1.82400012 1.144 0.56 0.8800001 1.288 1.984 1.496
occipital
gyrus (wm)
166 R. lingual 16.6160011 2.184 4.656 4.3840003 12.088 4.208 11.84
gyrus (wm)
167 L. lingual 11.000001 1.752 3.024 2.424 6.544 3.272 5.632
gyrus (wm)
168 R. cuneus 6.46400023 3.504 2.256 1.784 2.448 4.176 3.64
(wm)
169 L. cuneus 1.77600014 0.576 2.648 0.536 1.424 2.168 2.848
(wm)
170 R. insula 1.1680001 0.632 0.832 0.744 1.96 2.08 1
(wm)
171 L. insula 1.64800012 0.768 0.808 0.648 2.408 2.632 1.512
(wm)

In Table 3.1, “wm” represents white matter regions, and “gm” represents
gray matter regions. From the above table, we noticed that the left mamil­
lary body, right parahippocampal gyrus (gm), and left parahippocampal
gyrus (gm) are not present in babies. Left ventricular system, right lateral
orbitofrontal gyrus, right middle orbitofrontal gyrus (wm), left anterior
orbitofrontal gyrus (wm), right pars orbitalis (wm), and left pars orbitalis
(wm) are not present in most of the babies and adults. Right mamillary body,
right lateral ventricle, right pars orbitalis (gm), left pars orbitalis (gm), right
transverse frontal gyrus (gm), left transverse frontal gyrus (gm), right middle
orbitofrontal gyrus (gm), right anterior orbitofrontal gyrus (gm), left anterior
orbitofrontal gyrus (gm), right posterior orbitofrontal gyrus (gm), right para-
central lobule (gm), right temporal pole (gm), left temporal pole (gm), right
fusiforme gyrus (gm), left fusiforme gyrus (gm), left transverse frontal gyrus
(wm), left lateral orbitofrontal gyrus (wm), right cingulate gyrus (wm), right
temporal pole (wm), left temporal pole (wm), and right parahippocampal
gyrus (wm) are missing in some babies.
60 Computational Imaging and Analytics in Biomedical Engineering

3.4 CONCLUSION

In this chapter, automatic segmentation of four pediatric and three adult

brains is done using Brainsuite 19a brain imaging software, and the volumes
of different brain parts (170) are computed.

KEYWORDS

• segmentation
• pediatric brain
• atlas-based method
• volume
• MRI

REFERENCES

1. Gonzalez, R. G.; Woods, R. E. Third Edition Digital Image Processing; Pearson: South
Asia, 2014.
2. Sridhar, S. Digital Image Processing; Oxford, India, 2011.
3. Najarian, K.; Splinter, R. Biomedical Signal and Image Processing; Taylor & Francis:
New York, 2006.
4. Demirkaya, O.; Asyali, M. H.; Sahoo, P. K. Image Processing with MATLAB®
Applications in Medicine and Biology; CRC Press: New York, 2009.
5. Semmlow, J. L. Biosignal and Medical Image Processing; CRC Press: New York, 2009.
6. Gousiasa,I. S.; Hammersa, A.; Heckemanna, R. A.; Counsella, S. J.; Dyeta, L.
E.; Boardmana, J. P.; David Edwardsa, A. Atlas Selection Strategy for Automatic
Segmentation of Pediatric Brain MRIs into 83 ROI. IEEE, 2010.
7. Klimont, M. B.; Flieger, M.; Rzeszutek, J.; Stachera, J.; Zakrzewska, A.; JoNczyk-
Potoczna, K. Automated Ventricular System Segmentation in Paediatric Patients Treated
for Hydrocephalus Using Deep Learning Methods. BioMed Res. Int. 2019, 1–9.
8. Makropoulos, A.; Gousias, I. S.; Ledig, C.; Aljabar, P.; Serag, A.; Hajnal, J. V.; David
Edwards, A.; Counsell, S.J.; Rueckert, D. Automatic Whole Brain MRI Segmentation
of the Developing Neonatal Brain. IEEE Trans. Med. Imaging 2014, 33 (9), 1818–1831.
9. Devi, C. N.; Sundararaman, V. K.; Chandrasekharan, A.; Alex, Z. A. Automatic Brain
Segmentation in Pediatric MRI Without the Use of Atlas Priors. IEEE WiSPNET 2016
Conference, 2016; pp 1481–1484.
10. Gousias, I. S.; Hammers, A.; Counsell, S. J.; David Edwards, A.; Rueckert, D. Automatic
Segmentation of Pediatric Brain MRIs Using a Maximum Probability Pediatric Atlas.
IEEE 2012.
Comparative Volume Analysis of Pediatric Brain 61

11. Gao, Y.; Zhang, M.; Grewen, K.; Thomas Fletcher, P.; Gerig, G. Image Registration and
Segmentation in Longitudinal MRI Using Temporal Appearance Modeling. IEEE 2016,
629–632.
12. Levman, J.; Vasung, L.; MacDonald, P.; Rowley, S.; Stewart, N.; Lim, A.; Ewenson, B.;
Albert Galaburdad, E. Regional Volumetric Abnormalities in Pediatric Autism Revealed
by Structural Magnetic Resonance Imaging. Int. J. Dev. Neurosci. 2018, 34–45.
13. Kasiri, K.; Dehghani, M. J.; Kazemi, K.; Helfroush, M. S.; Kafshgari, S. Comparison
Evaluation of Three Brain MRI Segmentation Methods in Software Tools. 17th Iran.
Conf. Biomed. Eng. 2010.
14. Gibson, E.; Young, M.; Sarunic, M. V.; Beg, M. F. Optic Nerve Head Registration Via
Hemispherical Surface and Volume Registration. IEEE Trans. Biomed. Eng. 2010, 57
(10), 2592–2595.
CHAPTER 4

COMPARISON OF REGION OF
INTEREST AND CORTICAL AREA
THICKNESS OF SEIZURE AND
HEMOSIDERIN-AFFECTED BRAIN
IMAGES
VAJIRAM JAYANTHI1, SIVAKUMAR SHANMUGASUNDARAM1, and
C. VIJAYAKUMARAN2
1
SENSE, Vellore Institute of Technology, Chennai, India
2
Department of CSE, SRM Institute of Science and Technology,
Kattankulathur, Chennai, India

ABSTRACT

Brain disorders and tumors are caused by severe neurological abnormalities

in their functions. Hemosiderin induces changes in the characteristic func­
tional magnetic field and can be detected using susceptibility-weighted T1
images. Clinical syndromes and their relevance are unclear. Hemosiderin and
epilepsy brain images are preprocessed with available software tools, based
on the region of interest, mean thickness area, and cortical thickness area of
the different parts of the brain. The Brainsuite tool analyzes and processes
the fMRI images that produce a spherical topology with cortical surface
representations. The functional magnetic resonance image will calculate the
lesion or scar-affected region based on the variation of the neural activity.
The single package is used for segmentation and analyzing accurate brain
images for diagnostic and clinical purposes.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
64 Computational Imaging and Analytics in Biomedical Engineering

4.1 INTRODUCTION

Hemosiderin is a spinal tumor and is associated with ependymomas of

the spinal cord.1 It may also be visible in paragangliomas, hemangio­
blastomas, and larger cavernous spine malformations; it may develop a
hemosiderin cap and appear and produce a lack of a complete hemosiderin
ring where they reach the pial surface. In brain tumors, hemorrhage is
common. The seizure initiation mechanism of hemorrhage causes changes
in the blood metabolism, like hemosiderin and focal cerebral irritation,
in subarachnoid hemorrhage and into the basal cisterns, which directly
contact the temporal and frontal lobes. Patients with an intraparenchymal
component to the hemorrhage also have a subarachnoid hemorrhage.2–6
The natural course of tumors is caused by the new onset of seizures,
and this has to be addressed in statistical studies of larger patients. The
neurological disorder of seizure leads to fits. Due to the unavailability
of patient data documenting effectiveness, most of the patients suffer
from subarachnoid hemorrhage or intracerebral disease, which leads to
short-term antiepileptic prophylaxis.

4.2 PREPROCESSING OF IMAGE

Image preprocessing is the primary step in the processing of the images.

Before model training and inference of images, they should be formatted
based on the processing techniques. This includes resizing, orienting,
and color corrections. Image restoration involves the correction of
degradation, noise, and atmosphere deflections,7,8 which occur during
the processing of images. The output produces a corrected image that is
similar to the original image’s characteristics. The purpose of arbitrarily
manipulating an image is to support a preferred reality and achieve
an aesthetic standard. The processing of image translation takes place
between the digital imaging devices and the human visual system,9 and
the image restoration involves taking a noisy image and estimating the
clean image from the original image. Image enhancement techniques
like deblurring or contrast stretching by a nearest neighbor procedure,
provided by imaging packages,10 and the image preprocessing methods
are discussed in Figure 4.1.
Comparison of Region of Interest and Cortical Area Thickness of Seizure 65

FIGURE 4.1 fMRI and brain image preprocessing methods.

4.2.1 MOTION AND SLICE TIME CORRECTION

Motion correction has a number of correction methods that modify the MR

k-space or image data during reconstruction or after acquisition, independent
of the potential motion performed by acquisition.11 The nature of fMRI slices
and acquisition protocols is that they are temporally misaligned from each
other and cannot be acquired simultaneously.12 This misalignment is to be
corrected by slice timing correction (STC) and preprocessing pipelines.
Slice acquisition parameters mainly depend on the slice timing correction,
interact in a nonlinear fashion with head movement, and change the other
preprocessing steps of fMRI acquisition13 parameters and design.

4.2.2 INTENSITY NORMALIZATION

Normalization is a subject to ensure comparability across images; full brain

normalization may work better than white strip interpretation, subsequent
lesion analysis, and segmentation algorithms. This method is used for inten­
sities14 that are comparable across all tissues.

4.2.3 REGISTRATION

The registration process overlays two or more images from various sensors
or imaging equipment taken at different angles and times, or from the
66 Computational Imaging and Analytics in Biomedical Engineering

geometrically aligned same scene of images for analysis (Zitová and Flusser,
2003), transforming the different sets of data into one coordinate system.
The military’s automatic target recognition, medical imaging, and analysis
of image data are obtained from satellites. The MRI, CT, SPECT, or PET
enables the combination of data from multiple modalities to get complete
information about the patient. It helps to facilitate treatment verification,
improve interventions, monitor tumor growth,15 or compare patient data to
anatomical atlases.

4.2.4 REGRESSION

The regression of the locality-context trade-off is highly task-dependent.

Several deep convolutional neural networks (DCNNs) are used for image-to­
image regression tasks such as denoising, relighting, colorization, etc. These
approaches typically involve specific architectures coupled with fine-tuned
ancillary postprocessing methods that perform more tasks than classification
tasks. The locality-context is coupled with the habitual trend of incorporating
VGG and ResNet architectures for nonclassification16 tasks.

4.2.5 SMOOTHING SPATIAL FILTER

Noise reduction and blurring of images are done by smoothing the spatial
filter. Blurring is a preprocess of removing noise, and smoothing filters are
used to reduce image noise, sharpen edges of image structures, and enhance
an image with the help of spatial domain operations or filtering of images.
The software tool analysis is covered by the Brainsuite and its step-by-step
procedure of inhomogeneity correction, brain extraction-registration, tissue
segmentation, intensity normalization, and cortical surface extraction. After
that, surface volume registration, region of interest (ROI), mean thickness area,
and cortical surface area thickness were calculated as shown in Figure 4.2.

FIGURE 4.2 Brainsuite 21a functions of processing and interfacing.

Comparison of Region of Interest and Cortical Area Thickness of Seizure 67

4.3 MATERIALS AND METHODS

The parameters of the fMRI WRE CON images of T1AXPOST and Hemo­
siderin are preprocessed by the Brainsuite tool. The tool includes skull and
scalp removal, image nonuniformity compensation, voxel-based tissue
classification, topological correction, rendering, and editing functions. The
purpose of preprocessing image acquisition is to eliminate unwanted infor­
mation or noise from ultrasound images without evading vital information.
These variations follow further image analysis steps. The analysis of tissue
types, pathological regions, and anatomical structures is carried out by the
segmentation process. The first task is to find a ROI, eliminating unwanted
regions from processing, and the second task is to segment the disease.
The boundary estimation, classification, and categorization of diseases are
done by segmentation to maintain the accuracy and sensitivity of detecting
lesions (Guo and Şengür, 2013). The features technique is used to eliminate
the false detection rate and improve diagnosis after the segmentation of the
disease. The suspicious areas can be categorized as benign or malignant on
the basis of selected features using different classification techniques. The
distinct histopathological components and malignant modules with vague
boundaries are often fused with adjoining tissues to produce the delineation
of tissue.17 To improve the accuracy, decrease the misdiagnosis from earlier
detection, and improve the correct diagnosis, all required a fully computer­
ized system (Nguyen et al., 2019).
The tool is designed to produce cortical representations, and each stage of
the cortical surface identification process was captured, and ROI, mean area
thickness, and cortical area thickness were extracted. The present results are
compared with normal brain image values for further diagnosis. The average
thickness of the full brain is around 2.5–3 mm, and the individual brains vary
from about 2 mm at their thinnest in the calcarine cortex up to 4 mm and
over in the thicker regions of the precentral gyrus, superior temporal lobes,
and superior frontal lobes (Zilles, 1990). The cerebral cortex area thickness
lies between 1 and 4.5 mm, and it is a highly folded neuron sheet, with an
average of approximately 2.5 mm, which is compared with the measured
values to detect the abnormalities.

4.4 RESULTS AND DISCUSSION

The tool is used for surface and volume registration, data diffusion, tensor
fitting, and structural coregistration. The brain image analysis is used to
68 Computational Imaging and Analytics in Biomedical Engineering

extract label, volume, and cortical surface anatomy and diffusion coregister
to structural images to allow modeling of labeled brain anatomy. Functional
magnetic resonance imaging (fMRI) measures the small changes in brain
activity occurring due to blood flow. It can detect abnormalities in the
brain that cannot be found with other imaging processes. Hemosiderin is an
iron-storage complex that is composed of lysosomes and partially digested
ferritin. The breakdown of the hemisphere gives rise to biliverdin and iron.
Then the body traps and releases the iron and stores it in hemosiderin tissues.
Hemosiderin (golden-brown granules, iron-positive) is usually found in
macrophages of red pulp and also in the marginal zone to a lesser degree.
The preprocessing steps of hemosiderin are shown in Figures 4.3 and 4.4,
and the preprocessing methods and extraction of surface area and volume
are shown in Figure 4.5.

FIGURE 4.3 Stripping of skull.

FIGURE 4.4 Surface extraction of cortical area.

Figures 4.3 and 4.4 fMRI brain Wrecon images A to B [hemosiderin]

preprocessing result obtained from the Brainsuite tool.

FIGURE A Stripping of skull.

Comparison of Region of Interest and Cortical Area Thickness of Seizure 69

FIGURE B Scalp and skull processing.

FIGURE C Classifications of tissue.

FIGURE D Cerebrum labeling.

IGURE E Identification of cortex.

FIGURE F Scrub mask.

FIGURE G Correction of topology.

70 Computational Imaging and Analytics in Biomedical Engineering

FIGURE H Wisp and Dewisp process.

FIGURE I Surface correction of Pial area.

FIGURE J Splitting of hemisphere.

FIGURE K Cortical thickness estimation.

FIGURE L Surface thickness of cortical area.

FIGURE M Surface volume registration 1.

FIGURE N Surface volume registration 2.

FIGURE 4.5 fMRI brain Wrecon image [Image of Seizure Brain—T1 AX POST], from
Figure A to N are the preprocessing results of seizure-affected MRI image of T1 axial post
from Brainsuite tool.
Comparison of Region of Interest and Cortical Area Thickness of Seizure 71

TABLE 4.1 Region of Interest, Mean Thickness, and Cortical Area Calculated Based on
Various Brain Parts of fMRI Brain Seizure Images (Brain T1 AX POST) and fMRI Brain
Images (hemosiderin).
ROI_ Brain parts Mean thick- Cortical area Cortical area
ID ness (mm) (mm2) fMRI (mm2) fMRI
fMRI brain brain seizure brain images
seizure images images (brain (hemosiderin)
(brain T1 AX T1 AX post)
post)
120 R. superior frontal gyrus 0.610544 657.278228 3510.586298
121 L. superior frontal gyrus 1.304945 1655.557214 6183.092200
130 R. middle frontal gyrus 1.153275 337.825409 4008.851409
131 L. middle frontal gyrus 0.869990 718.740569 12,975.890310
142 R. pars opercularis 1.088048 15.596868 1286.285113
143 L. pars opercularis 0.585431 443.580352 1149.394662
144 R. pars triangularis 1.355878 142.183448 11,051.325879
145 L. pars triangularis 1.600054 927.154561 1144.524778
146 R. pars orbitalis 2.367167 1.241634 126.537966
147 L. pars orbitalis 6.727943 77.504858 184.713848
150 R. precentral gyrus 0.153721 66.350107 5353.633353
151 L. precentral gyrus 0.355621 1141.159916 5750.413625
162 R. transverse frontal gyrus 3.707373 0.000000 84.850144
163 L. transverse frontal gyrus 4.049633 136.511933 139.778236
164 R. gyrus rectus 0.000074 3.663462 557.282092
165 L. gyrus rectus 3.778476 346.149674 430.304697
166 R. middle orbitofrontal 6.180444 2.469826 216.428357
gyrus
167 L. middle orbitofrontal gyrus 0.997979 63.335274 156.230785
168 R. anterior orbitofrontal 4.456243 4.495258 153.023251
gyrus
169 L. anterior orbitofrontal 0.989992 39.046137 169.868460
gyrus
170 R. posterior orbitofrontal 5.490447 0.764592 559.311023
gyrus
171 L. posterior orbitofrontal 1.493429 108.787511 436.082553
gyrus
172 R. lateral orbitofrontal gyrus 6.009783 3.741775 148.448843
72 Computational Imaging and Analytics in Biomedical Engineering

TABLE 4.1 (Continued)

ROI_ Brain parts Mean thick- Cortical area Cortical area

ID ness (mm) (mm2) fMRI (mm2) fMRI
fMRI brain brain seizure brain images
seizure images images (brain (hemosiderin)
(brain T1 AX T1 AX post)
post)
173 L. lateral orbitofrontal gyrus 1.895485 57.798568 163.160877
182 R. paracentral lobule 0.001593 21.024277 1002.323518
183 L. paracentral lobule 0.519313 79.186887 802.989708
184 R. cingulate gyrus 0.299899 338.531295 10,883.008641
185 L. cingulate gyrus 1.069448 1765.784771 7928.537898
186 R. subcallosal gyrus 0.001817 5.363604 159.993860
187 L. subcallosal gyrus 2.490743 65.046220 107.299517
222 R. postcentral gyrus 0.064345 159.239118 7476.025936
223 L. postcentral gyrus 0.447987 1271.812278 8852.655389
224 R. supramarginal gyrus 0.069290 87.657322 11,284.485650
225 L. supramarginal gyrus 0.283713 719.854933 1828.480007
226 R. angular gyrus 0.000081 3.426407 1638.347003
227 L. angular gyrus 0.929038 1175.327788 839.668871
228 R. superior parietal gyrus 0.005154 49.395522 4067.667996
229 L. superior parietal gyrus 0.754984 1064.805629 3984.797039
242 R. precuneus 0.655383 128.855513 770.375863
243 L. precuneus 0.884500 86.900164 873.517733
310 R. temporal pole 5.024895 0.903588 122.678009
311 L. temporal pole 1.080730 881.510760 90.360559
322 R. superior temporal gyrus 0.033416 11.524551 216.700062
323 L. superior temporal gyrus 1.243263 1002.493071 494.096565
324 R. transverse temporal gyrus 0.000071 2.074070 27.874815
325 L. transverse temporal gyrus 0.698706 42.794973 22.813240
326 R. middle temporal gyrus 0.925798 4.868301 2129.113825
327 L. middle temporal gyrus 0.394457 902.492128 2133.666365
328 R. inferior temporal gyrus 0.035113 15.863507 1325.005013
329 L. inferior temporal gyrus 0.531635 2871.989764 1535.948755
Comparison of Region of Interest and Cortical Area Thickness of Seizure 73

TABLE 4.1 (Continued)

ROI_ Brain parts Mean thick- Cortical area Cortical area

ID ness (mm) (mm2) fMRI (mm2) fMRI
fMRI brain brain seizure brain images
seizure images images (brain (hemosiderin)
(brain T1 AX T1 AX post)
post)
330 R. fusiforme gyrus 0.092988 21.762781 1332.834595
331 L. fusiforme gyrus 0.728324 1062.793859 921.583444
342 R. parahippocampal gyrus 0.112606 0.472803 76.263671
343 L. parahippocampal gyrus 0.253237 343.688600 23.527293
422 R. superior occipital gyrus 0.000094 1.409307 98.363716
423 L. superior occipital gyrus 0.164753 330.183458 149.840778
424 R. middle occipital gyrus 0.000089 0.819907 723.966451
425 L. middle occipital gyrus 0.184693 3468.862603 704.513176
442 R. inferior occipital gyrus 0.027703 33.959151 1224.301781
443 L. inferior occipital gyrus 1.139114 515.485582 65.896897
444 R. lingual gyrus 0.019824 141.372045 990.610529
445 L. lingual gyrus 0.734774 322.399934 452.486295
446 R. cuneus 0.000096 1.359830 920.563079
447 L. cuneus 0.308793 534.099470 764.495928
500 R. insula 0.363296 47.627683 520.052231
501 L. insula 2.614143 735.265075 365.875768

Table 4.1 shows the various regions of the brain and their mean area
thickness and cortical area thickness in both the seizure-affected brain
and the hemosiderin brain. The values are compared with normal, healthy
people’s brain area values, and the severity of disease is calculated for further
diagnosis and clinical analysis with the purpose of treatment.
The medical care system follows various methods to detect and analyze
epileptic seizures. To identify the seizure type and etiology, diagnostic tools
like electroencephalogram (EEG), magnetic resonance imaging (MRI),
magnetoencephalogram (MEG), single photon emission computed tomog­
raphy (SPECT), neuropsychiatric testing, and positron emission tomography
(PET) are used. For identifying specific seizure types, the EEG is critical.
For newly diagnosed patients, a CT scan will be used, but always an MRI
is preferred for the brain analysis. It can help determine the proper seizure
74 Computational Imaging and Analytics in Biomedical Engineering

type and syndrome. MRI may locate brain lesions like scars or anatomic
defects that are not detected by CT scans or conventional radiographs.
All data are fetched into any one of the analysis techniques of statistical
parametric mapping (SPM), MEG, or Curry analysis software to find and
detect the stages of seizures and where they start, and the requirements of
either medication or surgery are advised by medical practitioners. Machine
learning algorithms are used to detect and analyze datasets by using different
learning, classifier, and statistical measurement methods. Recent work is
focused on fMRI to find the correlation between epileptic seizures and cere­
bral hemodynamic changes.18 Seizures can be detected through the observa­
tion of the brain, heart rate, oxygen level, muscle activities, and artificial
sounds or visual signatures like MRI, motion, and audio or video recording
of the head and body of the person.19 Measuring asymmetric interactions
in the resting state of brain networks is discussed in Ref. [20]; correcting
inhomogeneity-induced distortion in fMRI using nonrigid registration in
Ref. [21]; the linear spherical deconvolution and model-free linear trans­
form methods for diffusion MRI in Ref. [22]; and the identification of partial
correlation-based networks compared with cortical thickness data in Ref.
[23]. The Gaussian models are used to compare the nonrigid image registra­
tion.24 The automatic cortical surface registration and labeling are analyzed
in Ref. [25]. The geodesic curvature flow on surfaces for automatic sulcal
delineation is compared with the cortical surface area.26 An invariant shape
representation using the anisotropic Helmholtz equation for the human brain
is shown in Ref. [27]. The Fourier 2-sphere linear transforms of diffusion
MRI are compared with different regions of the cortical surface.28 Correcting
the susceptibility-induced distortion in diffusion-weighted MRI using
constrained nonrigid registration is also discussed in Ref. [29].

4.5 CONCLUSIONS

The present research is based on neutrosophy-based methods to solve various

image-processing problems based on their handling capability of indetermi­
nate information. This chapter covers the neuroanatomical abnormalities in
two fMRI images, in which the cortical thickness, volume, and area were
extracted from the software tool and compared between patients and healthy
controls. The cortical area thickness is used to diagnose cortical abnormali­
ties in different regions of the brain. More studies are required to analyze the
brain and its abnormalities.
Comparison of Region of Interest and Cortical Area Thickness of Seizure 75

KEYWORDS

• seizure
• hemosiderin
• fMRI (functional magnetic resonance image)
• software tool
• brain

REFERENCES

1. Kondziolka, D.; Bernstein, M.; Resch, L.; Tator, C. H.; Fleming, J. F.; Vanderlinden,
R. G.; Schutz, H. Significance of Hemorrhage into Brain Tumors: Clinicopathological
Study. J. Neurosurg. 1987, 67, 852–857.
2. Rosen, A. D.; Frumin, N. V. Focal Epileptogenesis After Intracortical Hemoglobin
Injection. Exp. Neurol. 1979, 66, 277–284.
3. Ueda, Y.; Willmore, L. J.; Triggs, W. Amygdalar Injection of FeCl3 Causes Spontaneous
Recurrent Seizures. Exp. Neurol. 1998, 153, 123–127.
4. Moran, N. F.; Fish, D. R.; Kitchen, N.; Shorvon, S.; Kendall, B. E.; Stevens, J. M.
Supratentorial Cavernous Haemangiomas and Epilepsy: A Review of the Literature and
Case Series. J. Neurol. Neurosurg. Psychiatry 1999, 66, 561–568.
5. van Breemen, M. S.; Wilms, E. B.; Vecht, C. J. Epilepsy in Patients with Brain Tumours:
Epidemiology, Mechanisms, and Management. Lancet Neurol. 2007, 6, 421–430.
6. Kucukkaya, B.; Aker, R.; Yuksel, M.; Onat, F.; Yalcin, A. S. Low Dose MK-801 Protects
Against Iron-Induced Oxidative Changes in a Rat Model of Focal Epilepsy. Brain Res.
1998, 788, 133–136.
7. Robinson, R. J.; Bhuta, S. Susceptibility-Weighted Imaging of the Brain: Current Utility
and Potential Applications. J. Neuroimaging. 2011, 21 (4), e189–e204.
8. Park, M. J.; Kim, H. S.; Jahng, G. H.; Ryu, C. W.; Park, S. M.; Kim, S. Y. Semi
Quantitative Assessment of Intra-Tumoral Susceptibility Signals Using Non-Contrast-
Enhanced High-Field High-Resolution Susceptibility-Weighted Imaging in Patients
with Gliomas: Comparison with MR Perfusion Imaging. AJNR Am. J. Neuroradiol.
2009, 30, 1402–1408.
9. Wen, P. Y.; Macdonald, D. R.; Reardon, D. A. Updated Response Assessment Criteria
for High-Grade Gliomas: Response Assessment in Neuro-Oncology Working Group. J.
Clin. Oncol. 2010, 28, 1963–1972.
10. Sato, J. R.; Hoexter, M. Q.; Fujita, A.; Rohde, L. A. Evaluation of Pattern Recognition
and Feature Extraction Methods in ADHD Prediction. Front. Syst. Neurosci. 2012.
11. Shinohar, R., T. et al. Statistical Normalization Techniques for Magnetic Resonance
Imaging. NeuroImage: Clin. 2014, 6, 9–19.
12. Godenschweger, F.; Kägebein, U.; Stucht, D.; Yarach, U.; Sciarra, A.; Yakupov, R.;
Lüsebrink, F. et al. Motion Correction in MRI of the Brain. Phys. Med. Biol. 2016, 61
(5), R32–R56.
76 Computational Imaging and Analytics in Biomedical Engineering

13. Parker, D. B. et al. The Benefit of Slice Timing Correction in Common fMRI
Preprocessing Pipelines. Front. Neurosci. 2019, 13, 821–830.
14. Fortin, J-P. et al. Removing Inter-Subject Technical Variability in Magnetic Resonance
Imaging Studies. Neuro Image 2016, 132, 198–212.
15. Zitova, B. In Encyclopedia of Biomedical Engineering, Acquisition Methods, Methods
and Modeling 2019.
16. Santhanam, V. et al. Generalized Deep Image to Image Regression, 2017.
17. Koundal, D.; Sharma, B. In Neutrosophic Set in Medical Image Analysis, 2019.
18. Fergus, P.; Hussain, A.; Hignett, David; Al-Jumeily, D.; Khaled, A-A.; Hani, H. A
Machine Learning System for Automated Whole- Brain Seizure Detection. Appl.
Comput. Inf. 2015.
19. Fergus, P.; Hignett, D.; Hussain, A.; Al-Jumeily, D.; Khaled, A-A. Automatic Epileptic
Seizure Detection Using Scalp EEG and Advanced Artificial Intelligence Techniques.
BioMed Res. Int. 2015, 1–17.
20. Joshi, A. A.; Salloum, R.; Bhushan, C.; Leahy, R. M. Measuring Asymmetric Interactions
in Resting State Brain Networks. Inf. Process Med. Imaging 2015, 24, 399–410.
21. Chambers, M. C.; Bhushan, C.; Haldar, J. P.; Leahy, R. M.; Shattuck, D. W. Correcting
Inhomogeneity-Induced Distortion in FMRI Using Non-Rigid Registration. Proc. IEEE
Int. Symp. Biomed. Imaging 2015, 1364–1367.
22. Haldar, J. P.; Leahy, R. M. The Equivalence of Linear Spherical Deconvolution and
Model-Free Linear Transform Methods for Diffusion MRI. IEEE 10th Int. Symp.
Biomed. Imaging (ISBI), 2013, 508–511.
23. Wheland, D.; Joshi, A.; McMahon, K.; Hansell, N.; Martin, N.; Wright, M.; Thompson,
P.; Shattuck, D.; Leahy, R. R. Identification of Partial-Correlation Based Networks
with Applications to Cortical Thickness Data. In 9th IEEE International Symposium on
Biomedical Imaging (ISBI), 2012; pp 2–5.
24. Somayajula, S.; Joshi, A. A.; Leahy, R. M. Non-Rigid Image Registration Using Gaussian
Mixture Mode. In Biomedical Image Registration. Springer: Berlin Heidelberg, 2012;
pp 286–295.
25. Joshi, A. A.; Shattuck, D. W.; Leahy, R. M. A Method for Automated Cortical Surface
Registration and Labeling. Biomed. Image Regist. Proc. 2012, 7359, 180–189.
26. Joshi, A. A.; Shattuck, D. W.; Damasio, H.; Leahy, R. M. Geodesic Curvature Flow on
Surfaces for Automatic Sulcal Delineation. In 9th IEEE International Symposium on
Biomedical Imaging (ISBI), 2012.
27. Ashrafulla, S.; Shattuck, D. W.; Damasio, H.; Leahy, R. M. An Invariant Shape
Representation Using the Anisotropic Helmholtz Equation. Med. Image Comput.
Comput. Assist. Interv. 2012, 15 (3), 607–614.
28. Haldar, J. P.; Leahy, R. M. New Linear Transforms for Data on a Fourier 2-Sphere with
Application to Diffusion MRI. In 9th IEEE International Symposium on Biomedical
Imaging (ISBI), 2012; pp 402–405.
29. Bhushan, C.; Haldar, J. P.; Joshi, A. A.; Leahy, R. M. Correcting Induced Distortion
in Diffusion-Weighted MRI Using Constrained Non-Rigid Registration. In Signal &
Information Processing Association Annual Summit and Conference (APSIPA ASC),
2012; pp 1–19.
CHAPTER 5

DESIGN AND ANALYSIS OF CLASSIFIER

FOR ATRIAL FIBRILLATION AND DEEP
NEURAL NETWORKS WITH ECG
M. SIVAKUMAR,1 S. OMKUMAR,2 B. ARPUTHAMARY,3 and
YOHANNES BEKUMA BAKARE4
1
Department of ECE, Mohamed Sathak A.J. College of Engineering,
Chennai, India
2
Department of ECE, SCSVMV (Deemed University), Kanchipuram,
India
3
Department of Computer Applications, Bishop Heber College,
Tiruchirappalli, India
4
Department of Electrical and Computer Engineering ,College of
Engineering and Technology, Wollega University, Ethiopia

ABSTRACT

Atrial fibrillation is the popular factor of risk in evaluating the coronary and
heart diseases. Deep learning is a quite common area of interest in various
medical applications connected with classification of heartbeats with ECG
signals. ECG can be employed for the detection of AF. This paper describes
constructing a classifier for the detection of atrial fibrillation in signals of
ECG using deep neural networks. Generally, neural network structures of
convolution (CNN) and recurrent types (RNN) are employed for classifi­
cation of heartbeats. The proposed method in this paper employs memory
networks and analysis using both time and frequency. If raw signals are
trained with the LSTM network, the classification accuracy obtained is

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
78 Computational Imaging and Analytics in Biomedical Engineering

poor. Instead if network is trained with features of time-frequency-moment

for each signal, the classification performance gets improved and time for
training will be less. The simulation results show significant improvement
in sensitivity and efficiency for heartbeat classifications. It is very much
clear from observations that models of LSTM DL’s are suitable for heartbeat
classifications.

5.1 INTRODUCTION

The behavior of heartbeat of a person can be observed generally using ECG.

ECGs will indicate the heartbeat of a person to be either normal or not. AF
can be defined as inappropriate mannerism of heartbeat which prevails if
the upper chambers of heart (artia) mismatch when compared with that of
chambers of lower region. The data obtained from ECG are available from
challenge of PhysioNet1–3 for the proposed work. The data are sampled at
a frequency of 300 Hz and categorized into “4” groups such as Normal,
Atrial Fibrillation, Rhythm, and Others. This illustration indicates how
the process of classification can be automated using the concept of deep
learning. A binary classifier is employed in the procedure to differentiate
normal and signed signals. The concept of long short-term memory (LSTM)
networks is discussed in this paper to observe the sequence and series of data
with respect to time parameter. A bidirectional LSTM layer can monitor and
observe sequence of time in both directions—forward and backward.

5.2 MATERIALS AND METHODS

(i) ECG Data Description

The data obtained from the PhysioNet database comprises the classes of
people with cardiac arrhythmia (ARR), failure of heart (CHF), and sinus
(NSR). The database collected comprises 96, 30, and 36 observations from
the patients with ARR, CHF, and NSR, respectively. The MAT-file comprises
test ECG data in the format of time series. An image classifier is employed to
differentiate between the above “3” groups in this example.
Figure 5.1 shows the DL Simulink model. A MATLAB Function block
inside the pre-processing system is used to execute CWT for getting ECG
signal scalogram and finally the image is obtained from it. The image clas­
sifier will do prediction for classification of images with respect to deep
learning CNN.
Design and Analysis of Classifier for Atrial Fibrillation 79

FIGURE 5.1 ECG deep learning simulink model.

A function named ecg_to_scalogram is defined to execute the following

activities:
• The input given is 65,536 samples of ECG data with double precision.
• Input representation is available from ECG if Wavelet transform is
applied.
• The wavelet coefficients give the scalogram.
• The scalogram is then converted to image (227 × 227 × 3).
The ECG post processing MATLAB function illustrates the corresponding
image function to identify label with respect to highest score obtained from
classifier of images.
(a) Loading of Data and Observation: A MAT file is generated from
the PhysioNet which comprises ECG signals in a proper format.
(b) Add Signals and Labels to Workspace: A cell array named
signal is used to include the signals from ECG. A categorical array
named label is used to include the ground-truth labels of the signals
accordingly.
(c) Readings: The number of AFib and normal signals composite in the
data can be monitored by the Matlab summary function.
(d) Generation of Histograms: Majority of the data signals are 9000
samples long and shown in Figure 5.2.
(e) One signal segment can be monitored from each and every class.
80 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 5.2 Histogram plot of ECG signal lengths about 9000 samples.

FIGURE 5.3 Normal signal plot—P wave and QRS complex observation.
Design and Analysis of Classifier for Atrial Fibrillation 81

In Figure 5.3, the irregular and regular intervals of time correspond

to AFib heartbeats and normal heartbeats respectively. Generally,
the P wave is absent in AFib heartbeats but projects before the QRS
complex in case of normal heartbeats.
(f) Classifier Training with Raw Data: The raw signals produced
earlier can be employed to design the classifier. The set of readings
can be used to train and test the classifier accuracy with respect to
new data. Finally, the summary function can be applied to indicate
ratio of two signals to 1:7.
(ii) Training of LSTM Network Architecture using ADAM Solver
The Matlab trainNetwork function can be used to train the LSTM network
with various specified options of training and layer architecture. The total
process of training will consume some time due to bigger training set. The
network training is provided by the ADAM solver.

FIGURE 5.4 Plot of training progress.

The accuracy of classification on each batch is represented by training

accuracy as shown in Figure 5.4. On successful progress of training, this
value reaches 100% at maximum. On successful progress of training, the
cross entropy loss value reaches a minimum value of zero. Figure 5.5
illustrates accuracy of classifier’s training oscillating between 50% and
60%.
82 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 5.5 Confusion matrix chart.

(iii) Proposed Classifier Performance Enhancement with Extraction of

Features
The parameters from given data will enhance the classifier accuracies with
respect to both training and testing. To conclude on the extraction of features,
this paper implements a methodology which computes time/frequency-based
images which can be utilized for training the neural networks.

FIGURE 5.6 Spectrogram plot of normal ECG vs AFib ECG.

The LSTM technique is applied instead of CNN and more applicable

for signals of single dimension. The information from spectrograms can be
Design and Analysis of Classifier for Atrial Fibrillation 83

extracted by applying time-frequency based (TF) moments. These moments

are actual inputs to LSTM. Then it computed by applying Fourier transforms
of short time over time windows as shown in Figure 5.6.

FIGURE 5.7 Spectrograms for TF moments.

The spectrum of a signal either high or low can be determined by the

spectral entropy. If the signal has a spectrum of spiky type, then it has a low
value of spectral entropy. On the other hand, if the signal has a spectrum of
flat type, then it has a high value of spectral entropy. The spectral entropy is
then estimated by Matlab pentropy function. The number of time windows
used is 255 in this context. The centers of time windows and function time
outputs correspond to each other. Figure 5.7 shows the TF moments extracted
from the spectrograms.
(iv) Training the LSTM Network with Time-Frequency Features
It is essential to change the architecture of LSTM for two-dimensional
signals by mentioning the size of input sequence equal to 2.
(a) The LSTM layer must be bidirectional with output size equal to 100.
(b) Two classes must be specified by involving 2 layers—softmax and
classification layers.
(c) The options of training must be specified by allotting the number of
epochs to a maximum value of 30.
(d) The Matlab trainNetwork. Function can be used for training the
LSTM network overall.
84 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 5.8 Performance improved training-progress plot of raw ECG sequences.

Figure 5.8 indicates the enhancement in accuracy for training. The

value of cross-entropy loss moves toward the value of 0. Due to shorter
TF moments, the training time decreases when compared to that of raw
sequences. The confusion chart for LSTM, as shown in Figure 5.9, can be
employed to have a clear picture on various concepts such as accuracy in
testing, quality of training, etc.

FIGURE 5.9 Visualize the improved training and testing accuracy using confusion matrix
chart.
Design and Analysis of Classifier for Atrial Fibrillation 85

5.3 CONCLUSION

This paper illustrates the construction of a classifier for observing atrial

fibrillations in ECG signals by employing the network of LSTM. The
methodology applies the concept of oversampling to override the bias in
classification which prevails in regions where more number of patients with
good health are available. Instead of training LSTM network with data of
raw signals, training can be provided using features of TF moments for each
signal. Thereby, performance of classification is improved and the training
time is decreased.

KEYWORDS

• atrial fibrillation
• deep learning
• heartbeat
• dataset
• ECG
• heart disease

REFERENCES

1. AF Classification from a Short Single Lead ECG Recording: The PhysioNet/Computing

in Cardiology Challenge, 2017 [Online]. https://physionet.org/challenge/2017/
2. Clifford, G.; Liu, C.; Moody, B.; Lehman, L. W.; Silva, I.; Li, Q.; Johnson, A.; Mark,
R. G. AF Classification from a Short Single Lead ECG Recording: The PhysioNet
Computing in Cardiology Challenge 2017. IEEE Comput. Cardiol. 2017, 44, 1–4.
3. Goldberger, A. L., Amaral, L. A.; Glass, L.; Hausdorff, J. M.; Ivanov, P. C.; Mark, R. G.;
Mietus, J. E.; Moody, G. B.; Peng, C. K.; Stanley, H. E. PhysioBank, PhysioToolkit, and
PhysioNet: Components of a New Research Resource for Complex Physiologic Signals.
Circulation 2000, 101 (23), e215–e220. http://circ.ahajournals.org/content/101/23/
e215.full
4. Pons, J.; Lidy, T.; Serra, X. In Experimenting with Musically Motivated Convolutional
Neural Networks, 14th International Workshop on Content-Based Multimedia Indexing
(CBMI), June 2016.
5. Wang, D. Deep Learning Reinvents the Hearing Aid. IEEE Spectr. 2017, 54 (3), 32–37.
DOI: 10.1109/MSPEC.2017.7864754.
86 Computational Imaging and Analytics in Biomedical Engineering

6. Brownlee, J. How to Scale Data for Long Short-Term Memory Networks

in Python [Online] July 7, 2017. https://machinelearningmastery.com/
how-to-scale-data-for-long-short-term-memory-networks-in-python/.
7. Shen, M.; Zhang, L.; Luo, X.; Xu, J. In Atrial Fibrillation Detection Algorithm Based
on Manual Extraction Features and Automatic Extraction Features, Proceedings of the
IOP Conference Series: Earth and Environmental Science; Hulun Buir, China; August
28–30, 2020.
8. Wang, J.; Wang, P.; Wang, S. Automated Detection of Atrial Fbrillation in ECG Signals
Based on Wavelet Packet Transform and Correlation Function of Random Process.
Biomed. Signal Process. Control 2020, 55, 101662.
9. Shankar, M. G.; Babu, C. G. An Exploration of ECG Signal Feature Selection and
Classification Using Machine Learning Techniques. Int. J. Innov. Technol. Explor. Eng.
Regul. 2020, 9, 797–804.
10. Shi, H.; Wang, H.; Qin, C.; Zhao, L.; Liu, C. An Incremental Learning System for
Atrial Fibrillation Detection Based on Transfer Learning and Active Learning. Comput.
Methods Prog. Biomed. 2020, 187, 105219.
11. Hurnanen, T.; Lehtonen, E.; Tadi, M. J.; Kuusela, T.; Kiviniemi, T.; Saraste, A.; Vasankari,
T.; Airaksinen, J.; Koivisto, T.; Pänkäälä, M. Automated Detection of Atrial Fibrillation
Based on Time–Frequency Analysis of Seismocardiograms. IEEE J. Biomed. Health
Inf. 2016, 21, 1233–1241.
12. Andersen, R. S.; Peimankar, A.; Puthusserypady, S. A Deep Learning Approach for
Real-Time Detection of Atrial Fibrillation. Expert Syst. Appl. 2019, 115, 465–473.
13. Muthukumaran, D.; Sivakumar, M. Medical Image Registration: A Matlab Based
Approach. Int. J. Sci. Res. Comput. Sci. Eng. Inf. Technol. 2017, 2, 29–34.
14. Ahalya, S.; Umapathy, K.; Sivakumar, M. Image Segmentation–A MATLAB Based
Approach. Int. J. Eng. Res. Comput. Sci. Eng. 2017, 10–14, ISSN: 2394–2320.
15. Lehari, M. N. S.; Umapathy, K.;Sivakumar, K. M. SVD Algorithm for Lossy Image
Compression. Int. J. Eng. Res. Comput. Sci. Eng. 2017, 44–48, ISSN: 2394–2320.
CHAPTER 6

DESIGN AND ANALYSIS OF EFFICIENT

SHORT TIME FOURIER TRANSFORM
BASED FEATURE EXTRACTION FOR
REMOVING EOG ARTIFACTS USING
DEEP LEARNING REGRESSION
M. SIVAKUMAR,1 K. UMAPATHY,2 P. SRINIVASAN,3 and
S. PUNITHA4
Department of ECE, Mohamed Sathak A.J. College of Engineering,
1

Chennai, Tamil Nadu, India

2
Department of ECE, SCSVMV (Deemed University), Kanchipuram,
Tamil Nadu, India
3
Department of CSE, Muthayammal Engineering College, Rasipuram,
Tamil Nadu, India
4
Department of ECE, Muthayammal Engineering College, Rasipuram,
Tamil Nadu, India

ABSTRACT

This paper enunciates how to eliminate electro-oculogram (EOG) noise from

the signals of electroencephalogram (EEG) by employing the benchmark
dataset of EEG DenoiseNet. This work involves EOG-contaminated signals
of EEG signals for training a Regression model based on deep learning to
eliminate the artifacts of EOG. Initially, original input signals were used to

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
88 Computational Imaging and Analytics in Biomedical Engineering

train the model of regression and then transformed signals using short time
Fourier transform (STFT) were employed. The proposed model based on
STFT enhances the total especially at lower values of SNR. The mean square
error between actual EEG and denoised EEG signals is considered as the
metric for performance. The MSE value is also calculated between actual
signal and noisy EEG signals to indicate the worst case MSE if no denoising
is implemented. The results were simulated using MatLab and they indicate
that greater improvement in performance can be achieved by using the
sequences of STFT even at worst values of SNRs.

6.1 INTRODUCTION

Nowadays, deep learning networks are gaining more importance in different

domains including electroencephalography (EEG) signal processing. They
provide enhanced performance when compared to techniques available in
tradition. But currently the constraint for implementing solutions of deep
learning for EEG denoising is nonavailability of standard datasets with
proper benchmark. Hence, benchmark called EEGdenoiseNet is presented
here for training and testing denoising models based on deep learning.
This dataset comprises about 4514 clean segments, 3400 ocular artifacts,
and 5598 muscular artifacts by which contaminated EEG segments can be
synthesized with that of clean EEG segments. By applying the benchmark,
the performance of denoising can be evaluated for different types of classical
networks. Moreover, the methods of deep learning will have more signifi­
cance for denoising of EEG even under contamination of high noise.
There was a consideration for wavelet transformation, but there exists
a compromise in classification accuracy in various time/frequency decom­
posing methods. STFT is a method by which time and phase-based data can
be obtained from a signal. This method shall be employed for training of
EEG signals.

6.2 MATERIALS AND METHODS

(i) Feature Extraction using STFT Implementation

In the place of FFT, STFT can be applied to obtain the structure of frequency
for entire EEG signal in terms of time windows. Generally, applying extracted
features in lieu of original signal data is the traditional strategy for enhancing
Design and Analysis of Efficient Short Time Fourier 89

the performance of a model of deep learning. The extracted features will

give a good means of representation of input data easily accessible and
understandable by the network. This transformation with 64 windows and 63
overlap samples will yield 33 features of complexity with 449 samples each.
The complex features must be segregated into two components of real and
imaginary by placing the real features over the imaginary features to obtain
real features of 66 with each one length of 449 samples.
Figure 6.1 illustrates the complete activities of the process. A small
signal segment with a time window of few ms is taken for consideration
and then tapered to avoid edge artifacts. In this example, the hann window
is applied. FFT transformation is applied on the data segment tapered at the
rate of Nyquist frequency. The spectrum of the particular segment is then
placed in time space corresponding to that of FFT.

FIGURE 6.1 (A) Time segment extracted; (B) signal tapered; (C) FFT on tapered signal;
(D) FFT into time-frequency domain.
90 Computational Imaging and Analytics in Biomedical Engineering

(ii) Removal of EOG Artifacts:

FIGURE 6.2 Block diagram of proposed system.

Figure 6.2 shows the block diagram of the proposed approach for the
removal of EOG artifacts. It includes extraction of features using STFT and
reconstruction of EOG artifact signal. The Matlab transformSTFT helper
function normalizes the input signal and then computes its STFT-based clean
EEG data.

(iii) Implementation Using MatLab:

Training, validation, and testing datasets can be generated by applying
createDataset helper function of MatLab. This helper function produces
pairs of clean and noisy segments of EEG with different values of SNR by
integrating both clean EEG and EOG signals. The following expression can
be used to get a noisy segment with a given value of SNR:
noisyEEG = EEG + λ⋅EOG

The value of λ can be varied to control the artifact power with a specific
value of SNR. The following variables are included in the data segment for
implementation using MatLab.
• EEG for a clean EEG segment
• EOG for an EOG segment
Design and Analysis of Efficient Short Time Fourier 91

• noisyEEG for a noisy EEG segment

• SNR for signal-to-noise ratio of noisy segment
• Fs for sample rate of signal segments
Initially, the training file is read for data and both clean and noisy EEG
signals are plotted. Then call to methods of data store is done. It provides an
array of size of 1 × 2 with the first element and the second element corre­
sponding to noisy EEG segment and clean EEG segment, respectively. The
input and output signals are normalized for improving the performance of
the system.

FIGURE 6.3 Training plot of the clean and EEG with EOG artifact segments.

Figure 6.3 shows the training plot of the clean and EEG with EOG artifact
segments. The purpose of the proposed system is to train a network so that
it can yield STFT denoised signal representations with respect to input of
STFT corresponding to that of noisy signals. Finally, inverse STFT (ISTFT)
is applied to recover the denoised signal as shown in Figure 6.4.

FIGURE 6.4 Recovering the denoised EEG STFT segment for deep learning regression.
92 Computational Imaging and Analytics in Biomedical Engineering

Thus, the enhancement in performance can be obtained by using trans­

formSTFT helper function.

FIGURE 6.5 Resultant plot of STFT-based regression-based signal representation.

Figure 6.5 shows the resultant plot of STFT-based regression-based

signal representation with enhancement in extracted features in lieu of the
original raw signal data. The getRandomEEG helper function obtains EEG
signal of random with a particular value of signal-to-noise ratio from the
dataset in test.

FIGURE 6.6 Resultant plot of EEG denoising noisy signals for different SNRs.
Design and Analysis of Efficient Short Time Fourier 93

FIGURE 6.7 Improved SNR performance plot of average MSE without denoising.

Figures 6.6 and 6.7 show the plot of average value of MSE obtained
without denoising and denoising network trained with raw input signals and
STFT transformed signals, respectively. The performance of the system got
improved by using STFT with respect to decreased values of SNR.

6.3 CONCLUSION

This paper explains how a deep network shall be trained with EEG signals
to perform regression for signal denoising and removing artifacts of EOG
using the feature extraction model. Additionally, a comparison between the
two models trained with raw clean and noisy signals of EEG is also done.
The other technique applied is short-time Fourier transform to train,
validate, and test datastores using the Matlab transformSTFT function. The
complex features of EEG signals are treated as independent real features
by using Matlab implementations. From the simulation results, it is clear
that using STFT-based sequences yields greater performance enhancement
at worst values of SNRs and both approaches converge in performance with
the improvement in SNR.
94 Computational Imaging and Analytics in Biomedical Engineering

KEYWORDS

• regression model
• short-time Fourier transform
• mean-squared error
• denoising
• deep learning networks

REFERENCES

1. Mashhadi, N.; et al. In Deep Learning Denoising for EOG Artifacts Removal from EEG
Signals, IEEE Global Humanitarian Technology Conference (GHTC), 2020.
2. Yang, B.; Duan, K.; Fan, C.; Hu, C.; Wang, J. Automatic Ocular Artifacts Removal in
EEG Using Deep Learning. Biomed. Signal Process. Control 2018, 43, 148–158. DOI:
10.1016/j.bspc.2018.02.021.
3. Zhang, H.; Zhao, M.; Wei, C.; Mantini, D.; Li, Z.; Liu, Q. A Benchmark Dataset for Deep
Learning Solutions of EEG Denoising, [Online] 2019. https://arxiv.org/abs/2009.11662
4. Gandhi, T.; Panigrahi, B. K.; Anand, S. A Comparative Study of Wavelet Families for
EEG Signals Classification. Neurocomputing 2011, 74 (17), 3051–3057.
5. Bruns, A. Fourier-, Hilbert- and Wavelet-Based Signal Analysis: Are They Really
Different Approaches. J. Neurosci. Methods 2004, 137 (2), 321–332.
6. Zeng, H.; Song, A. Removal of EOG Artifacts from EEG Recordings Using Stationary
Subspace Analysis. Hindawi Publishing Corporation. Sci. World J. 2014, 2014.
7. Hussin, S. S.; Sudirman, R. R. EEG Interpretation through Short Time Fourier
Transform for Sensory Response Among Children. Australian J. Basic Appl. Sci. 2014,
8 (5), 417–422.
8. Mowla, Md. R.; Ng, S. C.; Zilany, M. S. A.; Paramesran, R. Artifacts-Matched Blind
Source Separation and Wavelet Transform for Multichannel EEG Denoising. Biomed.
Signal Process. Control. 2015, 22, 111–118. DOI: 10.1016/ j.bspc. 2015.06.009.
9. Nguyen, H. A. T.; et al. EOG Artifact Removal Using a Wavelet Neural Network.
Neurocomputing 2012, 97, 374–389.DOI: 10.1016/j.neucom.2012.04.016.
10. He, C.; Xing, J.; Li, J.; Yang, Q.; Wang, R. A New Wavelet Threshold Determination
Method Considering Interscale Correlation in Signal Denoising. Math. Probl. Eng.
2015, 2015, 280251. DOI: 10.1155/2015/280251.
11. Klados, M. A.; Bamidis, P. D. A Semi-Simulated EEG/EOG Dataset for the Comparison
of EOG Artifact Rejection Techniques. Data Br. 2016, 8, 1004–1006. DOI: 10.1016/j.
dib.2016.06.032.
12. Muthukumaran, D.; Sivakumar, M. Medical Image Registration: A Matlab Based
Approach. Int. J. Sci. Res. Comput. Sci. Eng. Inf. Technol. 2017, 2, 29–34.
13. Ahalya, S.; Umapathy, K.; Sivakumar, M. Image Segmentation–A MATLAB Based
Approach. Int. J. Eng. Res. Comput. Sci. Eng. 2017, 10–14, ISSN: 2394–2320.
14. Lehari, M. N. S. Umapathy, K.; Sivakumar, K. M. SVD Algorithm for Lossy Image
Compression. Int. J. Eng. Res. Comput. Sci. Eng. 2017, 44–48, ISSN: 2394–2320.
CHAPTER 7

MACHINE LEARNING FOR MEDICAL

IMAGES
N. GOPINATH,1 M. SURESH ANAND,1 and M.V. ISHWARYA2
1
Department of Computer Science and Engineering, Sri Sairam
Engineering College, Chennai, India
2
Department of Computer Science and Engineering, Agni College of
Technology, Chennai, India

ABSTRACT

Research in medical imaging is increasingly turning to machine learning

approaches. To recognize patterns in medical images, this method relies on
pattern recognition. To enhance medical imaging applications, these algo­
rithms are now being used to detect faults in the sickness diagnosing system,
which may lead to significantly ambiguous medical therapy. Medical imaging
relies heavily on these algorithms to detect early indications of sickness. To
make predictions, it employs either supervised or unsupervised algorithms
based on a predetermined dataset. Image categorization, object identifica­
tion, pattern recognition, and reasoning are just a few of the principles that
are examined in medical imaging. By extracting meaningful patterns for a
specific condition, they are employed in medical imaging to improve accu­
racy. They also aid in the process of deciding on a course of action. Using
Machine Learning to improve medical imaging is a trend that will continue
in the future. Machine Learning approaches in medical imaging are the focus
of this chapter’s discussion. Machine Learning is a must-know for medical
imaging professionals. We intended to give researchers a framework for

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
96 Computational Imaging and Analytics in Biomedical Engineering

understanding existing medical imaging approaches, highlighting the bene­

fits and shortcomings of these algorithms, and discussing future prospects.

7.1 INTRODUCTION

“It is thrilling to see what machine learning can do for the area of computer
science and engineering. As a subfield of artificial intelligence, it promotes
the extraction of meaningful patterns from instances, which is an essential
component of human intelligence.”1 To educate a computer system to think
like an expert, machine learning methods are essential.2 The goal of machine
learning research is to give computers the ability to learn on their own.3 For
example, a computer may collect patterns from data and then analyze them
on its own for autonomous reasoning in this domain.4 Medical imaging is
a fast expanding study field that is used to identify and cure diseases early.
Based on certain expectations, digital image processing has a major impact
on decision-making procedures. It improves the accuracy and extraction of
features. The process of functional evaluation is complex and contains a
wide range of attributes.5,6 There are several computer systems available that
use digital image processing methods. Image processing technologies must
be validated before they can be used to implement specific operations that
impact their performance. Medical imaging techniques are used for making
choices and taking action. Basic and advanced functionality are available
for image analysis and visualization. There are two subdomains of Artificial
Intelligence: Machine Learning and Deep Learning.8 To achieve Machine
Learning (ML), AI, and Deep Learning (DL) all require the same methodolo­
gies, which in turn are used to achieve AI.9
Medical pictures go through a series of processes before they may be used
to identify an output signal. It initially goes via the machine learning and
deep-learning algorithms. To focus on a certain area, the picture is divided up
into smaller sections. From these segments, attributes may be found utilizing
information retrieval methods. We then filter out any extraneous noise before
selecting the suitable quality. There are many ways to categorize the data
and generate predictions based on it in the end. The following stages are
always followed in a machine learning experiment. The most prevalent types
of machine learning algorithms include supervised, semisupervised, unsu­
pervised, reinforcement, and active learning. As a result, deep learning is a
kind of advanced machine learning that uses neural networks to categorize
and predict more precisely.10,11
Machine Learning for Medical Images 97

“FIGURE 7.1 Artificial intelligence, machine learning, and deep learning domain.
”

7.2 MEDICAL IMAGING AND ITS TYPES

Without any injury to the patient, we would be able to identify and treat
illnesses. Surgery and other invasive treatments are not necessary with
medical imaging, which allows us to observe what is going on within the
body without causing any harm. Possibly, we have taken it for granted at
one point or another. Medical imaging is one of the most powerful tools we
have for efficiently caring for our patients because it may be utilized for both
diagnosis and treatment.
In terms of diagnosis, common imaging types include:
• CT (computer tomography)
• MRI (magnetic resonance imaging)
• Ultrasound
• X-ray
• Nuclear medicine imaging (including positron-emission tomography
(PET)).
• Single photon emission computed tomography (SPECT), etc.
98 Computational Imaging and Analytics in Biomedical Engineering

7.3 HISTORY OF MEDICAL IMAGE ANALYSIS

Low-level photo processing and mathematical modeling were used succes­

sively to construct a rule-based system that could only perform a limited
set of tasks in the early days of medical image analysis. GOFAI (Good Old
Fashioned Artificial Intelligence)12 laws also existed in the field of artificial
intelligence. As supervised approaches became increasingly popular in the
1990s, they became more prevalent in the area of medical image analysis,
notably in the use of training data to train models. Active shape models and
atlas approaches are two examples. With the development of certain creative
ideas, pattern recognition and machine learning have become more popular.
As a result, the move from human-designed systems to computer-trained
systems based on example data has occurred.

7.4 MACHINE LEARNING ARCHITECTURE

The field of machine learning architecture has moved from the realm of
possibility to the realm of evidence. An early machine learning strategy for
identifying patterns established the framework for a future huge AI project.
There are three kinds of machine learning architecture: In addition to the
data collection, data processing, model engineering, exploration, and deploy­
ment that go into this architecture, there are also three types of unsupervised
learning: reinforcement learning, unsupervised learning, and self-paced
learning.13

FIGURE 7.2 Block diagram of decision flow architecture for machine learning
systems ”
Machine Learning for Medical Images 99

7.5 TYPES OF MACHINE LEARNING ARCHITECTS

7.5.1 SUPERVISED LEARNING

It gives a computer system a collection of realistic training situations

with clear goals. This training set system provides accurate responses to
given probable inputs. Supervised Learning is divided into two catego­
ries: regression and classification. Classifying inputs using different clas­
sification algorithms is an important part of a taught system’s operations.
Experiments like this one are known as “multi-labeling.” It is a spam
purification procedure that categorizes emails into one of two catego­
ries: Spam or Not Spam. There are no discrete outcomes in the regres­
sion, which is a supervised approach that returns continuous findings.
Regression predictions are evaluated using the Root Mean Squared Error
(RMSE), while classification predictions are evaluated using accuracy
(RMSE).

7.5.2 UNSUPERVISED LEARNING

Rather than training on a dataset, the system will make the decision on its
own. The system receives no labeling that may be utilized to make predic­
tions. Unsupervised learning may be used to discover the hidden pattern
by learning features from the data. An unsupervised learning technique
known as clustering is used to classify inputs into distinct clusters. These
clusters had previously gone unnoticed. It creates groupings based on
similarities.

7.5.3 SEMISUPERVISED LEARNING

The system is designed to be a training dataset in semisupervised learning.

Using previously trained data, this technique is used to uncover any poten­
tial outcomes that may be missing from the final model. This method is
used to train commitment on data that is not labeled. The semisupervised
learning method was developed using both unlabeled and labeled data,
and it demonstrated characteristics of both unsupervised and supervised
learning.
100 Computational Imaging and Analytics in Biomedical Engineering

7.5.4 ACTIVE LEARNING

It only obtains training tags for a small number of active learning events.
A substance’s optimality is enhanced to achieve the desired tag,. when it
comes to an organization’s budgeting tasks, for instance.

7.5.5 REINFORCEMENT LEARNING

As long as the learner is motivated by the program’s actions, such as driving a

vehicle or playing video games, the learnt data are only supplied in response
to the program’s activities.

7.5.6 EVOLUTIONARY LEARNING

Biological organisms and their progeny’s survival and fatalities are mostly
studied and predicted using this technique in biology. This model may be
used to predict how fitness information can be used to correct the outcome.

7.5.7 DEEP LEARNING

When neural networks are utilized for data learning and prediction, this is the
highest level of machine learning. Various algorithms make up this system.
Systems that can solve any issue and anticipate the result are built using
them. An extensive graph with various processing layers is used, which is
composed of numerous linear and nonlinear conversions.14
There are several factors that go into diagnosing an illness in today’s
medical environment. In order to accurately diagnose patients, doctors must
conduct thorough examinations and assessments. Health care data include
everything from medical assessments to patient comments to treatment
to supplement and prescription use. The healthcare business creates a lot
of data. Poor data management may lead to an association effect in these
reports.15 This is the main worry. If these medical records are to be mined
and processed effectively, this data has to be restructured. Different machine
learning methods may be used to distribute data based on its properties using
classifiers that are tailored to the specific needs of each project. Multiple
categories are available for sorting the data. Data from the medical field is
examined using these kinds of classifiers. It was for recognizing medical data
Machine Learning for Medical Images 101

sets that machine learning technologies were first developed and put to use.
A wide range of options are now available to organize and analyze medical
data using machine learning. Most new hospitals include data collecting and
inspection systems that are used to gather and share data. In the medical
field, it is utilized to rectify numerous diagnoses. There must be an accurate
diagnosis in order for an algorithm to work, and findings may be drawn from
earlier instances that have already been cracked. It is a machine learning
concept that uses patterns in medical pictures to predict and draw inferences
about the health of a patient.16,17

7.6 MEDICAL IMAGING AND MACHINE LEARNING

TECHNIQUES

The use of machine learning algorithms in medical imaging is quite

advantageous in the examination of specific illnesses. In medical image
processing, lesions and organs may be too complicated to depict correctly
by a simple mathematical solution. Image analysis using pixel-based
analysis was employed by the author of Ref. [18]. In medical image
processing, machine learning has been used to analyze pixel values rather
than extracting features from chunks of data. In certain circumstances,
the implementation of this technique may be preferable than the develop­
ment of feature-based classifiers. The lack of contrast between individual
pixels in high contrast photos makes them difficult to analyze. Pixel-based
machine learning does not need feature computation or segmentation,
unlike traditional classifiers that defend against erroneous segmentation
and feature computation problems. Pixel analysis takes a long time to
master because of the high dimensionality of the data (a large number of
pixels in a picture). The author used low-contrast images for analyzing
medical pictures. Histogram Equalization (HE) is the most effective
technique for improving contrast. “Modified Histogram-Based Contrast
Enhancement with Homomorphic Filtering” (MHFIL) is proposed by the
authors. It used a two-phase handling procedure, the first of which boosted
global contrast by modifying the histogram. For picture sharpening,
second-phase homomorphic filtering is also planned. The investigation
looks at 10 medical images of low-contrast chest X-rays. In comparison to
other approaches, the MHFIL has minimum values in all ten photographs.
Radiologists have the highest responsibility for medical image clarifica­
tion, with assignments comprising both high-quality images and their
102 Computational Imaging and Analytics in Biomedical Engineering

analysis. CAD (computer-aided design) has been created for a variety of

purposes. Linear discriminant analysis, support vector machines, decision
trees, and other machine learning methods are used to analyze medical
pictures. The author of Ref. [20] employed machine learning algorithms
to evaluate medical images. They employed local binary patterns widely
among texture descriptors in particular. In addition, a new trial using many
low binary pattern descriptors of biomedical images was investigated.
Beginning with face descriptions, the dataset of neonatal facial photos was
used to categorize pain syndromes. Results on a painstakingly prepared
2D HeLa dataset and the recommended descriptor, which incorporates all
of the different texture descriptors, achieve maximum implementation.
A linear support vector machine classifier is used on the 2D-HeLa and
PAP datasets. Accuracy of 92.4% was the highest among all other dataset
descriptors. Neuronal network techniques are used in medical pictures to
study disease characteristics.21 The neural network groupings are retained
for cancer research. It is used to criticize scenarios in which a cell is
considered to be normal with a high degree of assurance, and when each
single network has just two conceivable outcomes: either a normal cell
or a cancer cell. To integrate the predictions from the network of cells,
a common approach known as plurality voting is employed. There was
a high rate of accuracy and low false-negative analytical value for the
neural network as a whole, according to these findings: Expert systems
that employ machine learning may generate hypotheses based on patient
data. An expert system is built using rules gleaned from the knowledge of
specialists. It is possible to apply machine learning methods to develop
a systematic description of the clinical traits that are notably described
in a collection of clinical issues that may be utilized as examples. Deci­
sion trees and other decision-tree-like diagrams may be used to represent
this information. KARDIO, which was developed to interpret ECGs, is
an excellent example of this method. The use of statistics to examine
the properties of images in medical image analysis is a sound criterion.
Nuclear medicine imaging equipment is known as a “channeled Hotelling
observer” (CHO). Amenable themes in human visual structure activate
the channels. Image quality assessment may be done using this method,
and the results are positive for medical imaging. It is called a channelized
SVM because of the way it is implemented in the algorithm (CSVM).
Two medical physicists rated fault discernibility in 100 noisy pictures on a
six-point scale and then assessed the score certainty of a lesion's existence
in the current time period. After that, an additional 60 photographs are
Machine Learning for Medical Images 103

included in a training session. The human observers were able to complete

this task for six different flattening filter selections and two different
numbers of repetitions in the OS-EM rebuilding method.23

7.7 CONCLUSION

Since this technique was first conceived 50 years ago, machine learning
technology has advanced tremendously. Initially, the models were simplistic
and “brittle,” which meant that they could not handle any deviations from the
examples supplied in training. Due to the fast advancements in technology,
machine learning systems will soon be taking on tasks formerly reserved
for human beings. During the previous several years, advances in machine
learning have been made. Currently, machine learning algorithms are quite
resilient in actual circumstances, and the frameworks make the most of the
learning process. It was formerly used to describe the practice of medical
imaging, and it is predicted to increase quickly in the near future. The use
of machine learning in medical imaging has substantial implications for
medicine. It is imperative that our research leads to improving patient care.
Machine learning’s benefits must be taken seriously to make the most use of
them feasible.

KEYWORDS

• medical imaging
• machine learning
• image enhancement
• information retrieval
• supervised learning
• unsupervised learning

REFERENCES

1. Erickson, B. J.; Korfiatis, P.; Akkus, Z.; Kline, T. L. Machine Learning for Medical
Imaging. Radiographics 2017, 37 (2), 505–515 [Online]. DOI: 10.1148/rg.2017160130
PMCID: PMC5375621
104 Computational Imaging and Analytics in Biomedical Engineering

2. Latif, J.; Xiao, C.; Imran, A.; Tu, S. In Medical Imaging Using Machine Learning and
Deep Learning Algorithms: A Review, 2019 International Conference on Computing,
Mathematics and Engineering Technologies – iCoMET 2019, March 2019. DOI:
10.1109/ICOMET.2019.8673502
3. Valiant, L. G. A Theory of the Learnable. Commun. ACM 1984, 27 (11), 1134–1142.
4. Robert, C.; In Machine Learning, A Probabilistic Perspective; Taylor & Francis, 2014.
5. Doi, K. Computer-Aided Diagnosis in Medical Imaging: Historical Review, Current
Status and Future Potential. Comput. Med. Imag. Graph. 2007, 31 (4-5), 198–211.
6. Mahesh, M. Fundamentals of Medical Imaging. Med. Phys. 2011.
7. Jannin, P.; Grova, C.; Maurer, C. R. Model for Defining and Reporting Reference-Based
Validation Protocols in Medical Image Processing. Int. J. Comput. Assist. Radiol. Surg.
2006, 1 (2), 63–73.
8. https://www.intel.com/content/www/us/en/artificial-intelligence/posts/difference­
between-ai-machine-learning-deep-learning.html
9. Michalski, R. S.; Carbonell, J. G.; Mitchell, T. M. Eds.; In Machine Learning: An
Artificial Intelligence Approach; Springer Science & Business Media, 2013.
10. Norris, D. J. Machine Learning: Deep Learning. In Beginning Artificial Intelligence
with the Raspberry Pi.; Springer, 2017; pp 211–247.
11. Jankowski, N.; Grochowski, M. In Comparison of Instance Selection Algorithms
I. Algorithms Survey, International Conference on Artificial Intelligence and Soft
Computing; Springer, 2004.
12. Wernick, M. N.; Yang, Y.; Brankov, J. G.; Yourganov, G.; Strother, S. C. Machine
Learning in Medical Imaging. Process. Mag. 2010, 27, 25–38.
13. https://www.educba.com/machine-learning-architecture/
14. Schmidhuber, J. Deep Learning in Neural Networks: An Overview. Neural Netw. 2015,
61, 85–117.
15. Warwick, W.; et al. A Framework to Assess Healthcare Data Quality. Eur. J. Soc. Behav.
Sci. 2015, 13 (2), 1730.
16. Ghassemi, M.; et al. Opportunities in Machine Learning for Healthcare, 2018. arXiv
preprint arXiv:1806.00388.
17. Dua, S.; Acharya, U. R.; Dua, P. In Machine Learning in Healthcare Informatics, 2014.
18. Suzuki, K. Pixel-Based Machine Learning in Medical Imaging. J. Biomed. Imag. 2012,
2012, 1.
19. Agarwal, T. K.; Tiwari, M.; Lamba, S. S. In Modified Histogram Based Contrast
Enhancement Using Homomorphic Filtering for Medical Images, Advance Computing
Conference (IACC); 2014 IEEE International; IEEE, 2014.
20. Nanni, L.; Lumini, A.; Brahnam, S. Local Binary Pattern Variants as Texture Descriptors
for Medical Image Analysis. Artif. Intell. Med. 2010, 49 (2), 117–125
21. Shi, Z.; He, L. In Application of Neural Networks in Medical Image Processing,
Proceedings of the Second International Symposium on Networking and Network
Security; Citeseer, 2010.
22. Bratko, I.; Mozetič, I.; Lavrač, N. KARDIO: A Study in Deep and Qualitative Knowledge
for Expert Systems; MIT Press, 1990.
23. Narasimhamurthy, A. An Overview of Machine Learning in Medical Image Analysis:
Trends in Health Informatics. In Classification and Clustering in Biomedical Signal
Processing; IGI Global, 2016; pp 23–45.
CHAPTER 8

INNOVATIONS IN ARTIFICIAL
INTELLIGENCE AND HUMAN
COMPUTER INTERACTION IN THE
DIGITAL ERA
M. V. ISHWARYA,1 M. SURESH ANAND,2 A. KUMARESAN,3 and
N. GOPINATH4
1
Department of Artificial Intelligence and Data Science, Agni College
of Technology, Chennai, India
2
Department of Computing Technologies, School of Computing, SRM
Institute of Science & Technology, Kattankulathur, India
3
School of Computer Science and Engineering, Vellore Institute of
Technology, Vellore, India
4
Department of Computer Science and Engineering, Sri Sairam
Engineering College, Chennai, India

ABSTRACT

The artificial intelligence (AI) and human–computer interaction (HCI) people

group have frequently been described as having contradicting perspectives
on how people and PCs ought to interface. As the two of them advance,
there is a more profound differentiation that cuts across these networks,
in how scientists imagine the connection among information and plan. By
inspecting the rationalistic and plan directions fundamental groups of work
in the two disciplines, we feature applicable contrasts and opportunities for

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
106 Computational Imaging and Analytics in Biomedical Engineering

powerful communication with PCs. The new ascent of profound learning has
changed AI and has prompted a heap of pragmatic strategies and apparatuses
that fundamentally sway regions outside of center AI.
Specifically, current AI strategies presently power new ways for
machines and people to communicate. Accordingly, it is convenient to
examine how current AI can drive HCI research recently and how HCI
exploration can assist with coordinating AI advancements. This study
offers a gathering for specialists to examine new freedoms that lie in
bringing current AI strategies into HCI research, recognizing significant
issues to explore, exhibiting computational, and logical techniques that can
be applied, and sharing datasets and apparatuses that are now accessible
or proposing those that ought to be additionally evolved. The themes we
are keen on including profound learning techniques for comprehension
and demonstrating human practices and empowering new association
modalities, cross breed knowledge that consolidate human and machine
insight to address troublesome errands, and apparatuses and strategies for
collaboration information duration and enormous scope information driven
plan. At the center of these points, we need to begin the discussion on how
information driven and information driven methodologies of present day
AI can affect HCI.
With regards to these and different difficulties, the jobs of people working
couple with these frameworks will be significant, yet the HCI people group
has been just a tranquil voice in these discussions to date. This article diagrams
a background marked by the fields that recognize a portion of the powers that
avoided the fields as much as possible. Computer-based intelligence was by
and large set apart by an exceptionally aggressive, long haul vision requiring
costly frameworks, albeit the term was infrequently imagined as being the
length of it end up being, while HCI centered more around advancement and
improvement of broadly utilized equipment inside a brief time frame scale.
These distinctions prompted various needs, techniques, and appraisal draws
near.
We propose a human-PC communication stage for the hearing weakened,
that would be utilized in medical clinics and banks. To grow such a system,
we gathered Bosphorus Sign, a Turkish Sign Language corpus in wellbeing
and finance spaces, by counseling communication through signing etymolo­
gists, local clients, and area specialists. Utilizing a subset of the gathered
corpus, we have planned a model system, which we called HospiSign that is
expected to help the Deaf in their clinic visits. The HospiSign stage directs its
clients through a tree-based movement outline by asking specific questions
Innovations in Artificial Intelligence and Human Computer Interaction 107

and requiring the clients to reply from the given alternatives. In order to
perceive signs that are offered as responses to the cooperation stage, we
propose during hand position, hand shape, hand development, and chest area
present highlights to represent signs. To demonstrate the fleeting part of the
signs we utilized dynamic time warping and temporal templates. The order
of the signs is finished utilizing k-Nearest Neighbors and Random Decision
Forest classifiers. We led experiments on a subset of Bosphorus Sign and
assessed the adequacy of the framework in terms of highlights, transient
demonstrating strategies, and characterization techniques. In our experi­
ments, the mix of hand position and hand development highlights yielded
the highest acknowledgment execution while both of the fleeting demon­
strating and classification techniques gave serious outcomes. In addition, we
researched the impacts of utilizing a tree-based action chart and discovered
the way to deal with increment the recognition performance, yet additionally
facilitate the transformation of the clients to the framework. Furthermore,
we researched area variation and facial milestone limitation methods and
examined their appropriateness to the motion and communication through
signing acknowledgment errands.

8.1 INTRODUCTION

Human–computer interaction (HCI) centers around the plan, assessment,

and utilization of data and correspondence advances with an express
objective to further develop client encounters, task execution, and personal
satisfaction. HCI is right now being formed and molding the utilizations of
man-made reasoning (artificial intelligence [AI]) and keen increase (IA).
This is prompting the quick development of new and invigorating examina­
tion themes. These subjects and the inquiries got from them are broadening
and testing our ebb and flow hypothetical establishments and exploration
philosophies. This calls for both reflection and conversation among IS
researchers.
This track plans to give a gathering to talk about, create, and advance a
scope of examination investigating novel speculations, systems, and exact
bits of knowledge identified with marvel relating to HCI, AI, or poten­
tially IA. We invite entries mirroring a breath of examination customs and
approaches, including conduct, financial aspects of IS, econometrics, plan
science, and information science, that address new and arising issues in
these fields.
108 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 8.1 HCI and AI interaction.

Exploration applicable to the track might offer novel hypothetical

points of view on HCI, AI, and IA in an assortment of settings including,
however, not restricted to web-based business, m-trade, work, associations,
human collaborations with shrewd advances, human robot cooperations,
novel interface plans for computer-generated reality/expanded reality (VR/
AR) innovations, and assessment of HCI issues through neurophysiological
devices and gadgets (e.g., EEG, fMRI, GSR, and eye-trackers). Significant
exploration may likewise incorporate examinations analyzing the effect
of AI/IA on business technique improvement, applications in open versus
private associations, the impact of AI available, economy and society, and
the business esteem and unexpected results of AI. The logical and designing
disciplines that make up or are firmly aligned with AI and HCI are of some­
what ongoing beginning. They date back to the 1950s at the soonest, with
the main reference book article on AI showing up in 1967. AI and HCI have
been quickly changing and growing both in scholarly substance and in appli­
cation, particularly somewhat recently. The new sped up speed of progress
has been expected in no little part to the practically amazing advancements
and item improvements in supporting microelectronics, elect optics, and
show advances—all equipment escalated regions.
Development in the measure of new wording and specialized language
consistently goes with such changes and advances in innovation. Perceiving
the uselessness of endeavors to create extensive definitions considering these
quick changes, the board has picked to give brief portrayals of four regu­
larly utilized terms: computerized reasoning, human–PC interface, virtual
universes, and manufactured conditions. Man-made consciousness is the
assortment of calculations that whenever make it conceivable to help clients
Innovations in Artificial Intelligence and Human Computer Interaction 109

to see, reason, and act. Since it is calculations that make up AI, the elements
of seeing, thinking, and acting can be cultivated heavily influenced by the
computational gadget (e.g., PCs or mechanical technology) being referred
to.
Simulated intelligence at least incorporates:
• Representations of “reality,” insight, and data, alongside-related
techniques for portrayal;
• Machine learning;
• Representations of vision and language;
• Robotics; and
• Virtual reality (characterized underneath)
Human–PC interface (HCI) comprises of the accompanying:
• The machine combination and understanding of information and their
show in a structure advantageous to the human administrator or client
(i.e., shows, human insight copied in computational gadgets, and
reproduction and engineered conditions).
Availability and, specifically, versatility, and ease of use are helpful
highlights for a PC – equipment or programming asset, since they univer­
salize the advantages of its legitimate use. Openness implies the joining of
properties to permit utilizing a PC asset by individuals with some sort of
scholarly incapacity or sensorial/actual hindrance, just as the incorpora­
tion and expansion of its utilization to all areas of society. Versatility is
characterized as the limit of a PC asset to work in various conditions by
changing its setup which could be made by end clients, or more than likely
its ability to learn with client collaboration and adjust to accommodate
its clients. At long last, ease of use is identified with the office to learn
and utilize a PC asset. These ideas support the full advancement of avail­
ability, flexibility, and convenience highlights as a client alternative, and
not as an inconvenience of the asset configuration measure. To advance a
superior readability, the term openness is utilized in future as an overall
term for availability, versatility, and convenience. Def physical weakness,
just as the consideration and augmentation of its utilization to all areas of
society. Flexibility is characterized as the limit of a PC asset to work in
various conditions by changing its arrangement—which could be made
by end clients, or probably its ability to learn with client collaboration and
adjust to accommodate its clients. At last, convenience is identified with
the office to learn and utilize a PC asset.
110 Computational Imaging and Analytics in Biomedical Engineering

These ideas build up the full advancement of availability, versatility, and

ease of use highlights as a client choice, and not as a burden of the asset
configuration measure. To advance a superior decipherability, the term avail­
ability is utilized from now on as an overall term for openness, versatility,
and convenience. Characterizing what, how, when and for what reason to
fabricate PC assets with availability highlights is one of the primary worries
of the HCI region. The association between a client and the asset is portrayed
by a two-way course, wherein each incidental affects the conduct or the
outcomes created by the other one. The PC asset configuration measure
needs to consider the pertinent impacts delivered by such connection, and
meeting openness prerequisites includes the advancement of exceptional
systems to further develop this client collaboration what, how, when, and for
what reason to fabricate PC assets with availability highlights is one of the
fundamental worries of the HCI region.
The connection between a client and the asset is described by a two-way
course, where each incidental affects the conduct or the outcomes delivered
by the other one. The PC asset configuration measure needs to consider the
applicable impacts delivered by such cooperation, and meeting openness
necessities includes the advancement of exceptional procedures to further
development of this client association. Simulated intelligence instruments
used to execute it are momentarily examined; at long last, the future
examination bearings and ends are given and the connected book indices
are recorded. The main methodology, which I will call “rationalistic” (see
conversation of this term in tries to show individuals as intellectual machines,
whose inside instruments equal to those we incorporate into computerized
PCs. The clearest articulation of this view is Newell and Simon’s Physical
Symbol System Hypothesis,9 which impacted an age of scientists both in AI
and HCI. It was the hypothetical articulation of what Haugeland5 calls “Ordi­
nary AI” (GOFAI), which was the prevailing worldview in the period of
elevated requirements for the close term making of human-like AI. Newell’s
origination was additionally the vital motivation for the establishing text on
HCI as a discipline of psychological designing, which stays powerful in the
HCI people group today.
A speedy look at the papers in the yearly ACM SigCHI meeting shows
numerous papers that address communication issues from an exact quantita­
tive point of view. The critical suspicions of the rationalistic methodology are
that the fundamental parts of thought can be caught in a formal emblematic
portrayal. Regardless of whether it relates straightforwardly to a conventional
rationale, it works like a rationale in that clear cut algorithmic guidelines can
Innovations in Artificial Intelligence and Human Computer Interaction 111

be applied to models (cycles and information) as image structures. Furnished

with this rationale, we can make shrewd projects and we can plan frameworks
that improve human cooperation. The subsequent methodology is more
enthusiastically to mark. It has fondness to the individuals who call their
methodology “phenomenological,” “constructivist,” and “natural,” and I will
allude to it as a “plan” approach. In the plan approach, the attention isn’t on
demonstrating canny inward operations, yet on the connections between an
individual and the encompassing climate. Alongside this shift of center goes
a change in the sort of understanding that is sought after. A critical piece of
the thing that matters is in the job of formal demonstrating and clarification.
In plan, we regularly work in spaces of human translations and practices
for which we don’t have prescient models. The topic of “Accomplishes it
work?” isn’t drawn nearer as an estimation before development, however, as
an iterative cycle of model testing and refinement. David Kelley, author of
the plan firm IDEO and the head of Stanford’s Institute for Design (dschool.
stanford.edu) is regularly referred to his explanation that “Edified experi­
mentation beats the arranging of impeccable acumen.” This isn’t an assertion
against astuteness, yet rather an affirmation of the impediments of knowing
and demonstrating the intricacies of the genuine human world. Inside the
HCI people group, there has been a continuous movement away from
Newell’s intellectual designing methodology, toward a plan situated position
that draws significantly more from the experience of other plan disciplines.14
The present CHI people group is occupied with progressing banters about
the jobs of investigation and configuration in making new interfaces and
understanding the manner in which individuals associate with them (see, for
instance, Ref. [15]). Obviously, “plan” is deciphered by various individuals
in many related, however, non-indistinguishable manners.
A speedy web look for the terms HCI and design gives a vibe for the variety
and enthusiasm of the conversation. Over a similar period, AI has followed
a comparable to direction. We have seen the overall surrender of GOFAI and
the ascendance of measurable, encapsulated, and constructivist draws near.
It is difficult to do equity to the significant ideas and contrasts among these
in a short remark, yet a string goes through them that resounds with the 1258
T. Winograd/Artificial Intelligence 170 (2006) 1256–1258 plan approach
in HCI. Maybe than putting together AI skill with respect to a coherent
portrayal of the setting and the specialist’s information, there is an exchange
between broad versatile systems and world experience, which leads after
some time to shrewd practices, frequently as the result of broad models and
preparing. Numerous pundits of AI, starting with Dreyfus2 have underlined
112 Computational Imaging and Analytics in Biomedical Engineering

the indistinguishability of human idea and human actual epitome. Taking on

a similar mindset as a human isn’t simply a question of the right cerebrum
engineering or quality grouping, yet of a complex formative communication
between the entire creature and its current circumstance, including others. He
and others have contended that for a gadget to have human-like knowledge
would require its encountering something similar to human turn of events
and childhood. The rationalistic reaction to this is to imitate the outcome of
actual world involvement in an assortment of suggestions, as in CYC,6 a task
which has neglected to satisfy its hopes after numerous long periods of exer­
tion (see Ref. [13]). Obviously, there is a significant job for a rationalistic
methodology in making and understanding the versatile components. Work
in measurable language understanding, neural organizations, or AI depends
on profound investigation and quantitative models of the various compo­
nents and strategies that structure the reason for transformation. In any case,
the analyst isn’t needed (or ready) to unequivocally address the information
or rules of conduct for the smart framework.

FIGURE 8.2 Growths of HCI and AI.

Eventually, “Edified experimentation beats the arranging of perfect mind.

Human and man-made brainpower collaboration are introducing huge and
energizing mechanical freedoms for common advancement even in the
present current innovative environment, however, the genuine potential
for shared improvement is within a reasonable time-frame and past which
can possibly be astounding. With the proceeded with fast advancement of
innovation as well as in quantum research, the potential for AI to develop
Innovations in Artificial Intelligence and Human Computer Interaction 113

at a terrifying rate is inside our grip. The structure squares to accomplish

this incredible jump in innovation comprises of Quantum Computing, Big
Data, and mind PC interface (BCI), each becoming venturing stones and
progressing to such a phase where when innovation just considered in sci-fi
will be an ordinary component of present day human and AI innovation.
Man-made intelligence and people carry their own one of a kind charac­
teristics with each gaining from different, people bring ascribes including
experience, qualities, and judgment which can weave along with the gigantic
traits that AI can contribute. Simulated intelligence and human association
can possibly create and push the limits of innovations on earth as well as
space investigation. For instance, with this coordinated effort comes the
potential for the fruitful turn of events and business utilization of nuclear
fusion to supply a clean new tremendous fuel source. Add this along with
the capacity of AI to nullify the need to rest, to take care of or the need to
breath oxygen, giving AI the capacity for huge space travel while examining
conceivable future human investigation and in any event, driving the limits
further with possible human colonization. Tremendous measures of large
information and the copious stock of organized information are fundamental
for the future headway of AI, the preparation of learning calculations for use
in deep learning and neural networks is vital for make AI more shrewd.
The preparing force of the present PCs is now working related to
enormous information ceaselessly propelling the capacity to handle huge
information, all the more productively developing close by the appearance
of equal preparing and Hadoop with both progressing at a frantic rate which
thus is empowering AI to turn out to be considerably more astute. Envision
the advancement of AI if the force of quantum computing was promptly
accessible with potential preparing power multiple times more impressive
than the present old style PCs with every producer endeavoring to arrive at
quantum supremacy as well as far and past this progressive advancement
Humanity can outfit this astounding innovation with AI and BCI creating
another innovation transformation, this is the future, yet what’s to come is
quick drawing closer. With quantum computing carrying the capacity to
handle enormous information in plenitude, this will thus soar AI improve­
ment prompting the progression of BCIs, with AI being able to pinpoint as
well as to guide and increment for instance DNA planning distinguishing
qualities in strength and knowledge. The opportunities for super strength and
incredibly smart people to coordinate with the enormous progression in AI
insight are achievable. With the peculiarity sure to be outperformed later on
people and AI will take a stab at artificial general intelligence (AGI).
114 Computational Imaging and Analytics in Biomedical Engineering

This potential for humans and AI to develop constantly together is

faltering, however, the topic of security, regulations, and ethics close by
AI and future human BCI headways features the requirement for standard
security and guidelines to be set up. From man-made brainpower and choice
emotionally supportive networks and their utilization in a scope of settings,
the following two articles research non-traditional HCI and moral contem­
plations in regards to explicit socio-specialized frameworks with regards
to crisis and catastrophe the board. For example, Shipman investigates
the moral difficulties related to the plan of foundation that is educated by
human conduct and person on foot elements. While the creator distinguishes
the chances this presents, from supporting individualized encounters and
personalization in brilliant urban areas, to working with improved depar­
tures during threatening or regular crises, the moral dangers are likewise
underlined. The requirement for industry self-guideline is obvious, just like
the significance of analysts and modelers displaying familiarity with the
pertinent moral difficulties.

8.2 HEALTHCARE SYSTEMS

This paper presents a continuous investigation pointed toward EMR quality

evaluation and improvement as a piece of general calculated and mechanical
reason for CDSS building.16 The considered issues show up and are settled
inside an extent of advancement for dynamic in medication and medical
care which utilizes EMR as a center wellspring of data on patients’ curation
in clinic. This makes the EMR quality improvement to be considered as a
significant issue for CDSS advancement. The appropriate plan of EMR GUI
isn’t the lone issue for aversion of slip-ups in EMR. As a matter of fact, just
slip-ups of 2–4 classes portrayed in 2.2 could be tackled thusly. Different
slip-ups classes need more clever arrangements, while the framework ought
to give hints dependent on recently entered data during the current client
meeting. This assignment can be settled utilizing CDSS coordinated with
EMR. Class 1 errors might prompt befuddle in drugs buy that prompt inac­
curate medication consumption.23 Creators in Ref. [24] propose the rundown
of activities for clinical specialists to keep away from such circumstance,
in this way, we accept that the CDSS can help by mistakes checking with
respect to finding and select that error amendment that fits to the analysis.
Another approach to further develop drug solution is to naturally change
drug names to utilize Tall Man lettering, while it can diminish drug name
disarray errors.25
Innovations in Artificial Intelligence and Human Computer Interaction 115

Computerized diagnostics of certain sicknesses dependent on clinical

trial results examination could assist with remedying class 4 errors. Different
account strategies for a similar data are a more muddled issue and there are
various approaches to settle it. One way is perceive current info explanation
and give autocomplete hint for the client. This clue might be founded on
most normal structure for such explanations recovered from EMR. What
direction we ought to choose is the theme for future examination. The CDSS
can perform such undertakings as checking conceivable medication drug
collaborations (DDI) in doctor prescriptions26 or checking the consistency
of field “analysis,” clinical trials, and endorsed drugs. In Ref. [27], creators
led meta-examination of a few investigations on CDSS. Creators notice
that advanced CDSS ought to be not difficult to utilize, and yet, ought to
keep away from the chance of disregarding proposals in the framework.
With everything taken into account, it is an uncommon test to foster HCI
in CDSS. Our flow project is committed to the improvement of a CDSS
for cardiovascular infections coordinated with EMR. The principal point of
this CDSS is to further develop HCI to diminish the quantity of missteps
in EMR. Our framework coordinates with the as of now utilized EMR (see
Fig. 8.1 for theoretical engineering), utilizes the records put away in EMR,
reuses and broadens the accessible EMR (UI). The second permits us to save
existing client experience, stay away from most mix-ups associated with
curiosity and acquire fair assessment of the progressions in the medical care
quality. Specialized issues of the coordination interaction are settled by the
UI expansion. That UI augmentation screens current client meeting in EMR,
cooperates with worker side of DSS for meeting investigation and shows
cautions and ideas when a few occasions are set off.

FIGURE 8.3 HCI and AI for healthcare systems.

116 Computational Imaging and Analytics in Biomedical Engineering

As of now the proposed engineering is effectively being worked out

to give (1) the adaptable incorporation with MIS; (2) gradual update HCI.
To play out certain slip-ups adjustment errands normal language preparing
techniques ought to be utilized and distinctive AI and information recov­
ering strategies, including counterfeit neural organization, choice trees,
and other characterization calculations could be utilized to make the
information base. Work on this task is as yet in progress and we are in
the beginning phase of advancement including the decision of explicit
calculations (Fig. 8.1). Reasonable engineering of the CDSS incorporated
in EMR. To further develop client association of the clinical data frame­
work, we utilize the cycle DMAIC (define, measure, analyze, improve,
and control).28 Currently, we have characterized the issue and distinguished
potential causes as per the primary period of the DMAIC cycle. The exami­
nation was led in genuine settings and gives rich subjective information
to foster the CDSS and the vision of things to come framework. In the
pattern of measure–analyze–improve, we will make another adaptation of
CDSS. Because of the way that clients conjure the MIS consistently, we
will apply a technique for Subjective Mental Effort Question (SMEQ)29 for
estimating fulfilment on each phase of cycle. A seven-digit size of reactions
gives a decent harmony between the quantity of alternatives for settling on
a powerful decision by the respondent and special cases blunder of estima­
tion. Culmination and exactness of information in EMR structures are the
reason for generous advantages, remembering better consideration, and
decline for medical services costs.
The significant piece of the EMR is the plan of HCI. Improvement of
human–PC communication influences definitely a larger number of cycles
than it may appear at the main look. Remembering that HCI influences doctor
appropriation of EMR, it likewise contacts the course of therapy, medical
care quality estimation, clinical information investigation, and numerous
different viewpoints that are not shrouded in this paper. This is the motivation
behind why EMR frameworks ought to be planned, carried out, and utilized
fittingly. Something else, clinical frameworks will bring about accidental
unfavorable results, for example, misdiagnosis, underestimation of infection
seriousness of associative sicknesses or medication drug communications.
The entirety of this might influence patient wellbeing.
Use of coordinated CDSS pointed toward further developing HCI gives
certain advantages and diminishes the quantity of mix-ups in EMR. It likewise
animates the appropriation of EMR by doctors. These progressions ought to
work on the nature of clinical consideration. Great nature of EMR furnishes
Innovations in Artificial Intelligence and Human Computer Interaction 117

information researchers with more information helpful for investigation and

more certainty about the pertinence of the outcomes acquired from EMR
information examination. In the meantime, as referenced above, progress
of human–PC association in EMR requests specialized arrangements, yet
in addition assistance of doctors’ comprehension of the significance of such
frameworks for their normal practice and further utilization of information
put away in such frameworks. CDSS engineers will work intimately with
doctors and information researchers to comprehend their requirements and
hardships.
Along these lines, we will lead ethnographic meetings with doctors,
as we did with information researchers, to comprehend the conduct
and customs of individuals interfacing with the framework. We intend
to utilize DMAIC cycle to work on the ease of use of the framework
being worked on. Shrewd Health with the expanding application and
acknowledgment of digitalization, especially in agricultural nations like
India, the Internet turns into a fundamental factor for medical services.
Savvy wellbeing like e-Health administrations has the critical potential
to upgrade patient analysis and to work on the availability and nature of
treatment using Internet applications and cloud stage.28 With the progres­
sion of ICT, Web 2.0 innovation has carried the degree to expand these
administrations of by permitting patients, patient’s families, and local
area everywhere to partake all the more effectively during the time spent
wellbeing advancement.
What’s more, schooling through long-range interpersonal communica­
tion measure,15 many agricultural nations are receiving e-Health for its
adaptability and speed to defeat hindrances like the comfort of arrangement,
diminished driving expenses and time, decreased patient holding up period,
further developed admittance to clinical subject-matter experts and wellbeing
data alongside directed plans and subsequent meet-ups. Such highlights open
another window for cutting edge and innovation empowered treatment appli­
cations. This likewise makes the consideration cycle a worked on personal
satisfaction for a great many country patients experiencing clinical problems
which can be better overseen.28 Because of the progression of innovation
alongside responsive highlights, a large number of the online applications
are currently available through cell phones and other shrewd gadgets, which
is carrying adaptability and solidarity to the client and saving his time and
cost. To perceive the future extent of cell phone in medical services, Joshi
and Pant have recommended from their exploration that execution of versa­
tile applications in wellbeing administrations is in the underlying stage yet
118 Computational Imaging and Analytics in Biomedical Engineering

its further use in both giving clinical instruction and expert medical care will
change the medical care industry.
HCI professionals give off an impression of being staying away from
a significant region; similarly, individuals working in the space appear
to be ignorant of HCI. For instance, are software engineering divisions
graduating designers ignorant of HCI? Maybe plan/makers/controllers/
obtainment may not be utilizing qualified individuals. Assuming we are
to work on the nature of wellbeing basic regions, HCI must be an inno­
vator in uncovering issues and showing it can keep away from or oversee
abandons. There is a minuscule writing on HCI in safety critical regions,
especially in medical care. In the medical care space, the majority of the
writing focuses on clinical sequelae of antagonistic occasions (e.g., by
directing CPR and a remedy, the patient recuperated from a medication
glut), not on the plan or HCI issues (e.g., the ergonomics of the keypad
urged the attendant to enter some unacceptable medication portion, and
the UI configuration gave no dynamic survey of the portion) that made the
idle conditions for the blunder. In the USA, the obligatory clinical gadget
mistake reports frequently fault the client at whatever point no specialized
breakdown could be recognized.
HCI and dormant conditions prompting the occurrence are by and large
overlooked. One of the conceivable outcomes is that HCI is intricate and
that clinical gadgets and conditions are perplexing. One may likewise add
further reasons like the exceptionally cut-throat market, quickly evolving
advancements, and issues undertaking dependable environmental client
contemplates (e.g., patient classification, or generally unessential clinical
exploration guidelines, for example, randomized control preliminaries). In
actuality, this paper showed that fundamental HCI can add to the medical
care. Obviously, a few issues of security basic frameworks lie outside HCI
itself; maybe plans of action inside an administrative system that disregards
HCI. Consider how we prepare and spur HCI experts to draw in—or stay
away from—significant regions like medical care. The turn of events and
take-up of various advancements might appear to be exceptionally unique in
various pieces of the world, due to accessibility, expenses, and specialized
availability. The Western world has seen an improvement from fixed PCs,
through PCs tablets, and versatile PCs, presently moving toward the Internet
of things, sensors, and universal advancements. It has gone from proficient
business related use to more private and recreation use. Numerous nations
from the creating scene skirt at least one stage in this interaction, furnishing
admittance to versatile advances with no or little utilization of fixed or PCs.
Innovations in Artificial Intelligence and Human Computer Interaction 119

Portable innovations are less expensive, more secure, and simpler to use in
a more customized way.
The methods of utilizing the computerized innovations fluctuate a ton in
various pieces of the world because of culture, yet in addition to accessibility.
In the Western world, a cell phone might be an extremely private device that
one infrequently provides for another person; yet in the country Africa, a cell
phone might be shared by numerous individuals in a more prominent social
setting.
Through the expanding digitalization of advanced education and the
expanding global accessibility through MOOCs and other carefully acces­
sible innovations for instructing and learning, the chances for building
abilities and occupations in the agricultural nations also increment
immensely. A particularly mechanical improvement will, if appropriately
utilized, assume a significant part in democratization and fair freedoms for
advancement and in adding to a few of the maintainability advancement
objectives.
It is important to create and adjust the techniques and cycles for plan and
improvement of computerized change to the particular territorial conditions
in each separate region. By opening up advancement, publicly supporting
improvement to take into account numerous to contribute, by being more
deft and considering changing prerequisites improvement could be made
considerably more applicable for different pieces of the world. Changing
conduct and mentalities to permit a more client focused methodology,
notwithstanding, has so far ended up functioning admirably in various
pieces of the world, regardless of whether the manners by which it has been
embraced change in various districts.
There is likewise a danger for “imperialistic colonialization” of program­
ming improvement in which “advanced superpowers” force their advance­
ment societies without recognizing the neighborhood perspectives in light of
a legitimate concern for digitalization. Supporting global turn of events and
digitalization requires a reasonable cycle of quietude and regard for nearby
customs and societies to deal with a solid presentation of new improvement
measures, instruments, practices, and strategies. In addition, research has
shown that the presentation of new innovation in medical care is once in
a while met with opposition and that dormancy is solid with regards to the
foundation of innovation. At the point when online clinical records were
presented in Sweden doctors emphatically disliked the framework as they
were stressed over their workplace and about patients perusing clinical
records on the web and becoming stressed over what they read.25
120 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 8.4 Healthcare growth with HCI and AI.

All in all, digitalization is in a general sense changing society through

the turn of events and utilization of computerized innovations. It might
profoundly affect the advanced improvement of each country on the planet.
Be that as it may, it should be created dependent on nearby practices, it
needs worldwide help and not to be restricted by any mechanical require­
ments. There is a requirement for a dependable plan measure that includes
a comprehension of neighborhood societies and customs for sound global
frameworks advancement and digitalization. Especially, digitalization to
help worldwide wellbeing requires a significant comprehension of the clients
and their specific circumstance and the whole medical care framework and
accordingly requires a client focused frameworks configuration approach.

8.3 DIGITAL HEARING

It is to some degree secret that sound isn’t utilized all the more broadly in
human–PC cooperation plan as we in reality can hear a ton in any case,
maybe, the hear-able truth is as well “undetectable” and hence hear-able
interface configuration seems, by all accounts, to be troublesome. There
is actually no explanation, aside from availability for outwardly weakened
clients, to add sound to human–PC interfaces that are as of now advanced for
the visual methodology, as it doesn’t actually add to framework convenience.
Innovations in Artificial Intelligence and Human Computer Interaction 121

In original human–PC cooperation ideal models,1–3 like universal, inescap­

able, wearable, and vanishing figuring, intuitive sonification may offer
potential options to the generally predominant visual presentations, opening
up our eyes to see the encompassing scene or do what little visual showcases
don’t progress nicely. In short, utilizing sound for portrayal in collabora­
tion configuration is valuable for standing out to occasions or areas, for
non-visual correspondence as a general rule, including discourse, cautions,
warning, and criticism. Sound is less helpful for consistent portrayal of
items (as we have no earlids), for outright readings (as the vast majority see
hear-able measurements like pitch, din, and tone) as this work was upheld
to a limited extent by Enterprise Ireland under the Basic Research Program
Multimodal Browsing project and by Mikael Fernström who is associated
with the Interaction Design Center, Department of Computer Science and
Information Systems, University of Limerick, Limerick, Ireland and Eoin
Brazil, who is associated with the Interaction Design Center, Department of
Computer Science and Information Systems, University of Limerick, being
relative), for fine-nitty gritty spatial portrayal, and furthermore is tricky in
loud or clamor delicate conditions. Planning intelligent sonifications for
human–PC communication necessitates that various issues are tended to.
We need to consider where and how sonification is suitable. As fashioners,
we should consider the clients’ abilities, while doing errands in genuine
conditions, and furthermore consider that encompassing commotion levels
may veil the framework’s sounds. In case, strong is considered to upgrade
collaboration, methods of making and testing hear-able allegories should
be investigated and how much the utilization of sound adds to the clients’
exhibition and abstract nature of utilization. To have the option to plan with
sound, we need an undeniable level comprehension of what and how we
hear (for example Gaver4,5). While there are a broad measure of studies on
the view of melodic sounds and discourse, moderately little is thought about
different sorts of non-discourse sounds and specifically purported ordinary
sounds. In the new association standards recently referenced, we can interro­
gate how we think regarding connection and human movement. In past work
on hear-able interfaces, going from Gaver’s Sonic Finder6 to Brewster’s
progressive earcons (for example, in Ref. [7]), human activity has generally
been considered in a discrete manner, for example, like kicking a football,
where a client activity begins a cycle that then, at that point finishes with
no further client control. This view may be fitting when pressing catches
or flicking switches. An elective view is activity as a nonstop stream, for
example, a pen stroke, where we consistently move a pencil on a surface,
122 Computational Imaging and Analytics in Biomedical Engineering

depending on our learnt signal through proprioception, just as haptic, visual,

and hear-able criticism. This last view is becoming significant as large
numbers of the new information gadgets like pens, digitizers, cameras, and
so on, are equipped for identifying very unpredictable human activities.
In any case, at the center of our plan space, a major issue is the way to
group and choose sounds to be reasonable for a specific collaboration plan.
Contingent upon our proposed clients, errands and setting there is at first a
wide continuum in this plan space, going from cement to extract portrayals,
for example, from hear-able symbols to earcons.8,9 In case, we are planning
for easygoing ordinary use, we most likely need to think about substantial
structures while in case we are planning for profoundly specific spaces, (for
example, cockpit or interaction control applications) where our clients will
be chosen and prepared, having elite necessities, we may have to zero in on
psychoacoustic issues, for example, while it is accepted that regular sounds
have intrinsic importance, gained from our regular exercises, hearing such
sounds in segregation without setting can be very befuddling. The sound of a
solitary confined stride can for instance be heard as a book being dropped on
a table. Curiously, this issue is fairly like how phonetic homonyms work, for
example, expressions of a similar spelling or sound as others, however, with
various implications.11 To additionally foster our comprehension of what
individuals think they hear, we chose to lead listening tests, a methodology
likewise utilized by numerous different scientists for example.12,13
Made top caliber (44.1 KHz, 16-digit) chronicles of 104 regular sounds
(terms somewhere in the range of 0.4 and 18.2 s) and had 14 postgraduate
understudies pay attention to the recorded sounds in irregular request
utilizing earphones, reacting in free-text configuration to what exactly each
solid was. As a rule, the portrayals given were very rich. In light of Ballas’
strategy for computing causal vulnerability,14–16 the reactions were arranged
and sorted just as assessed if right or not. From the reactions, we removed
activities and instruments/specialists sections of the writings, for example,
what the members thought were the items/materials making the sound and
how the articles collaborated. In our starter examination, we tracked down
that 77% of the activities and 71% of the instruments/specialists were effec­
tively recognized. We likewise noticed that the subsequent informational
index, with all reactions, categorizations, and estimations of causal vulner­
ability and precision could be utilized for proposing conceivable utilization
of sounds in communication plan, to some degree like Barrass’ strategy
for gathering tales about when sound is valuable in regular day to day
existence.17 According to a fashioner’s perspective, it is fascinating to take
Innovations in Artificial Intelligence and Human Computer Interaction 123

note of that the informational collection utilized in this manner contains

data about how individuals depict regular sounds just as estimations of
how precise the distinguishing proof of those sounds is. Taking a gander
at the sequencing between various sounds, we can see that occasionally the
request for sounds influences the distinguishing proof, especially for sounds
with high causal vulnerability. For instance, the sound of falling precipita­
tion is regularly haphazardly recognized either as downpour or fricasseeing
bacon. On the off chance that the sound before the downpour sound is
a birdcall, the reaction is all the more regularly downpour. In the event
that the sound before the downpour sound is conflicting of ceramic plates
and cutlery, audience members will in general react that it is the sound of
singing and the area is a kitchen. Comparative perceptions are examined by
Ballas and Howard14 proposing that there are a few equals between how we
measure language and regular non-discourse sounds with the discoveries
of the listening tests, we can begin to recommend conceivable hear-able
portrayals and allegories for connection plan. In light of Barrass’17 TaDa
approach, we can do an errand and information investigation that permits
us to choose sounds that can impart the measurements and headings that
give clients satisfactory criticism about their activities comparable to the
framework and the status and occasion of the framework. We then, at that
point need to make ways so the framework can deliver the chose sounds
lastly assess the subsequent plan with clients.18 In case we were to simply
play sound records for addressing client activities, they would consistently
solid the equivalent and never (or sometimes) be, for instance, expressive,
for example to be planned to the client’s distinguished exertion or the size of
the articles in question. This was one of the issues tended to by the sounding
object project. The sounding object project investigated new techniques for
truly motivated displaying of sounds for sound blend. Our work was at
first generally educated by natural acoustics and for instance Gaver’s work
on hear-able symbols.4,5,19 We likewise pursued cartoonification of sound
models, for example, improving on the models while holding perceptual
invariants. The models were carried out in pd2 and tried in various ways
going from perceptual analyses to imaginative execution. Contrasted with
normal sound records, sound items can give “live” sound models that can
be parametrically controlled progressively. The distinction can likewise be
depicted as full-fledged articles versus developmental items.20 For a full
grown item, the entirety of its boundaries is realized when actuated while
for a developmental article boundaries might change all through a persistent
use or initiation of the article.
124 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 8.5 AI and HCI on digital hearing.

FIGURE 8.6 Digital hearing.

This suggests that sound items, which can be utilized both as full grown
and transformative articles, are appropriate for hear-able portrayals for new
communication ideal models with non-stop control by client activities. With
cartoonification (or as some like, caricaturization), the sound models can
be made computationally more proficient than completely sensible models;
subsequently, they are bounded to be appropriate for “slender” stages like
wearable or handheld PCs.
Applications we need to truly consider if a bunch of sonification natives
can be incorporated with a working framework. This can thusly bring about
Innovations in Artificial Intelligence and Human Computer Interaction 125

the turn of events and check of solid toolbox for frameworks engineers, like
what is accessible for graphical UIs today. There is likewise a need to instruct
and uphold collaboration fashioners with the goal that they can open up their
innovative intuition toward intelligent sonification, for example, that it is
feasible to give persistent criticism continuously for signal-based gadgets. All
parts in human–PC interfaces likewise have tasteful properties. It is presum­
ably conceivable to plan sonifications that are psychoacoustically right and
very productive yet extremely horrendous to pay attention to. As proposed
by Eric Somers, we need to draw upon the information and thoughts of
Foley specialists (sound plan for film, radio, and TV) just as exercises gained
from different hypotheses of acousmatic music. We have examined some
clever ways to deal with intuitive sonification in human–PC collaboration
plan, specifically, for relaxed use and pervasive and wearable applications.
We recommend that such plans ought to be founded on the outcomes from
listening tests with potential similitudes being extricated from clients’ depic­
tions of ordinary sounds. The listening tests can likewise give direction in
our comprehension of how blends of hear-able symbols can be deciphered
by clients. At long last, we portray an instance of utilizing sound objects
for communication plan and propose spaces of future exploration. Have
executed a programmed interpretation framework, which converts finger-
spelled expressions to discourse and the other way around, in a customer
worker architecture. The objective of the investigation isn’t just to help a
meeting impeded individual yet additionally to assist an outwardly weakened
individual to connect with others. The framework upholds numerous spoken
and communications through signing, including Czech, English, Turkish,
and Russian, and the interpretation between these communicated in dialects
is taken care of utilizing the Google Translate API. The acknowledgment of
multi-lingual finger spelling and discourse was finished utilizing k-Nearest
Neighbors Algorithm (k-NN) and HMMs, separately. In the fingerspelling
synthesis model, a 3D enlivened symbol is utilized to communicate both
manual and non-manual features of an offered hint. He Dicta-Sign25 is a
multi-lingual communication through signing research project that points to
make Web 2.0 applications open for Deaf individuals so they can associate
with each other. In their Sign Wiki model, the creators show how their system
enables gesture-based communication clients to get data from the Web. Like
Wikipedia, in which users are approached to enter text as a contribution from
their console, gesture-based communication clients can search and alter any
page they need, and collaborate with the framework by means of a Microsoft
Kinect sensor in the Dicta-Sign Wiki. The Dicta-Sign is presently accessible
126 Computational Imaging and Analytics in Biomedical Engineering

in four languages: British Sign Language (BSL), German Sign Language

(DGS), Greek Sign Language (GSL), and French Sign Language (LSF).
In comparative way, Karpov et al.26 present their multi-modal synthesizer
system for Russian (RSL) and Czech (CSE) gesture-based communications
that utilize a 3D vivified avatar for blend. VisualComm,12,27 a Chinese Sign
Language (CSL) acknowledgment and translation apparatus, means to assist
the Deaf with speaking with hearing individuals utilizing the Microsoft
Kinect sensor progressively. The framework can decipher a hard of hearing
individual’s sign phrases to text or discourse and a conference individual’s
content or discourse to gesture-based communication utilizing a 3D ener­
gized symbol. In view of 370 day by day states, VisualComm accomplishes a
recognition rate of 94.2% and exhibits that 3D gesture-based communication
acknowledgment should be possible in real-time by utilizing the modalities
given by the Microsoft Kinect sensor.

8.4 DIGITAL HUMANITIES

Advanced humanities (DH) is a space of academic action at the crossing

point of registering or computerized innovations and the disciplines of the
humanities. It remembers the methodical utilization of computerized assets
for the humanities, just as the examination of their application.1,2 DH can
be characterized as better approaches for doing grant that include shared,
transdisciplinary and computationally drew in exploration, instructing, and
publishing.3 It carries advanced apparatuses and techniques to the investiga­
tion of the humanities with the acknowledgment that the printed word is as
of now not the principle vehicle for information creation and circulation.
By creating and utilizing new applications and procedures, DH makes
new sorts of instructing and exploration conceivable, while simultaneously
contemplating and scrutinizing what these mean for social legacy and
computerized culture.2 Thus, a particular component of DH is its develop­
ment of a two-way connection between the humanities and the advanced:
the field both utilizes innovation chasing after humanities examination
and subjects innovation to humanistic addressing and cross examination,
frequently at the same time. The meaning of the advanced humanities is
as a rule constantly figured by researchers and experts. Since the field is
continually developing and changing, explicit definitions can immediately
become obsolete or superfluously limit future potential. The second volume
of debates in the digital humanities recognizes the trouble in characterizing
Innovations in Artificial Intelligence and Human Computer Interaction 127

the field: “Alongside the computerized documents, quantitative investiga­

tions, and apparatus building projects that once described the field, DH
presently envelops a wide scope of techniques and practices: perceptions
of huge picture sets, 3D displaying of recorded ancient rarities, “conceived
advanced” theses, hash tag activism and the examination thereof, substitute
reality games, portable maker spaces, and that’s just the beginning.
In what has been called “enormous tent” DH, it can now and again be
hard to decide with any particularity what, decisively, advanced humani­
ties work involves. Generally, the computerized humanities created out
of humanities processing and has gotten related with different fields, like
humanistic figuring, social registering, and media examines. In substantial
terms, the computerized humanities accept an assortment of subjects, from
curating on the web assortments of essential sources (fundamentally printed)
to the information mining of huge social informational indexes to theme
demonstrating. Advanced humanities fuse both digitized (remediated) and
conceived computerized materials and join the procedures from conventional
humanities disciplines (like way of talking, history, reasoning, phonetics,
writing, workmanship, paleontology, music, and social examinations) and
sociologies, with apparatuses given by figuring (like hypertext, hypermedia,
information representation, data recovery, information mining, insights,
text mining, and computerized planning), and computerized distributing.
Related subfields of computerized humanities have arisen like programming
considers, stage examines, and basic code contemplates. Fields that equal
the computerized humanities incorporate new media studies and data science
just as media hypothesis of organization, game investigations, especially in
regions identified with advanced humanities project plan and creation, and
social examination.
In viable terms, a significant differentiation inside advanced humanities
is the emphasis on the information being handled. For handling literary
information, advanced humanities expands on a long and broad history
of computerized version, computational semantics and regular language
preparing and fostered a free and profoundly particular innovation stack
(generally cumulating in the determinations of the text encoding initia­
tive). This piece of the field is now and again hence put aside from digital
humanities overall as “advanced philology” or “computational philology.”
For the creation and investigation of advanced versions of items or curios,
computerized philologists approach computerized practices, techniques, and
innovations, for example, optical person acknowledgment that are giving
freedoms to adjust the field to the advanced age.
128 Computational Imaging and Analytics in Biomedical Engineering

The possibility of interface and related ideas like plan and convenience,
are the absolute generally vexed in contemporary processing. Meanings of
interface normally summon the picture of a “surface” or a “limit” where at
least two “frameworks,” “gadgets,” or “substances” come into “contact” or
“connect.” Though these terms support spatial translation, most interfaces
likewise exemplify worldly, haptic, and psychological components. The
directing wheel of a vehicle, the control board of a VCR, and the handle of
an entryway are altogether instances of ordinary interfaces that show these
measurements. With regards to PCs and processing, “interface” is regularly
utilized conversely with “graphical UI,” or GUI, most often experienced
as a work area windows climate. The order line brief is maybe the most
popular option in contrast to the GUI, however, there are a plenty of others,
including screen peruses, movement trackers, unmistakable UIs (TUIs, stun­
ningly delivered in the 2002 film Minority Report), and vivid or increased
registering conditions. In the humanities, in the interim, it is progressively
entirely expected to experience the possibility that a book or a page is a sort
of interface, a reaction to the arrangement that the shows of composition and
print culture are no less mechanically resolved than those of the computer­
ized world. Something like one onlooker, Steven Johnson, has characterized
our present recorded second as an “interface culture,” a term he employs
to accept not just the pervasiveness of PCs and electronic gadgets yet in
addition the manner by which interface has come to work as a sort of saying
or social getting sorted out standard—what the modern novel was to the
nineteenth century or TV to the rural American 1950s are his models.

FIGURE 8.7 Role of AI and HCI in humanities.

Innovations in Artificial Intelligence and Human Computer Interaction 129

However, much it is discussed; in any case, interface can now and again
appear to be minimal cherished. Ease of use master Donald A. Norman
states: “The genuine issue with interface is that it is an interface. Interfaces
disrupt everything. I would prefer not to zero in my energies on interface.
I need to zero in at work” (2002: 210). Nicholas Negroponte holds that the
“secret” of interface configuration is to “make it disappear” (1995: 93). To
additionally convolute issue, interface is frequently, practically speaking, an
exceptionally recursive wonder. Take the experience of a client sitting at
her PC and perusing the Web, maybe getting to content at a computerized
humanities website. The website’s inside plan forces one layer of interface
between the client and the substance, and the internet browser—its catches
and menus and edges—quickly force another. The client’s work area climate
and working framework then, at that point force a third layer of interface.
The ergonomics of the circumstance (we’ll accept our client is working with
a console and mouse, taking a gander at a screen situated the suggested 18
inches away) make still another layer of interface, a layer which becomes
clear when one considers choices, for example, getting to a similar substance
with a PDA or a wearable gadget, or in a room-based computer-generated
experience setting like a CAVE. Significantly, each of these “layers,” as I
have been calling them, shows the potential for collaboration with each other
just as with the client. The work area climate administers the conduct of
the program programming, whose highlights and capacities thus straightfor­
wardly influence numerous parts of the client’s cooperation with the sites’
inside plan and content.
While all that I have quite recently been practicing is recognizable enough
in software engineering circles, especially the area known as human–PC
connection (HCI, additionally now and then distinguished as human–PC
interface), parts of this account might appear to be tricky to perusers prepared
in humanities disciplines. It would not be difficult to come by somebody
willing to contend that my whole situation is the result of one more unac­
knowledged interface, a sort of normal social passage whose socially built
belief systems administer our assumptions as to innovation, portrayal, and
admittance to data. In addition, in the situation outlined over, my differentia­
tion between various layers of interface and something I nonchalantly called
“content” is one that contradicts many years of work in scholarly and social
analysis, where structure and content are naturally perceived as inseparable
from each other. Accordingly, the heaviness of set up shrewdness in a field
like interface configuration lays on a basic disengage with the predominant
scholarly presumptions of most humanists—that an “interface,” regardless
130 Computational Imaging and Analytics in Biomedical Engineering

of whether the windows and symbols of a site or the position of a sonnet on

a page, can some way or another be ontologically decoupled from whatever
“content” it ends up epitomizing. This is exactly where Brenda Laurel starts
her difference in computers as theater, her compelling study of standard
interface hypothesis:
Typically, we ponder intuitive figuring as far as two things: application
and interface. In the prevailing perspective, these two things are theoretically
particular: an application gives explicit usefulness to explicit objectives, and
an interface addresses that usefulness to individuals. The interface is what
we speak with—the thing we “talk” to—what intervenes among us and the
inward operations of the machine. The interface is regularly planned last,
after the application is altogether considered and maybe even executed; it
is joined to a previous heap of “usefulness” to fill in as its contact surface.
Shrub is writing to challenge this common perspective. However, in all
actuality, according to a designer’s point of view, the interface is frequently
reasonably unmistakable, yet in addition computationally particular. As John
M. Carroll calls attention to, it wasn’t until the relatively late approach of
dialects like visual basic that it even became pragmatic to program both UI
and application’s basic usefulness with a similar code (Carroll 2002: xxxii).
Like the historical backdrop of hand-press printing, which instructs us that
paper-production, type-setting, scratching or etching, and bookbinding came
to envelop altogether different areas of work and specialized mastery (once
in a while housed under a similar rooftop), asset improvement in the comput­
erized world is regularly exceptionally divided and compartmentalized.
Interfaces are for the most part examined in unremarkable and utilitarian
terms, however, figuring legend contains a lot of occasions where helpless
interface configuration has had deadly and cataclysmic outcomes. One
infamous scene included the Therac-25, a machine utilized for malignancy
radiation treatment during the 1980s, whose unwieldy programming inter­
face added to a few patients passing from overexposure. Moreover, the little
plane mishap that killed vocalist lyricist John Denver has been credited to
an inadequately planned cockpit interface, explicitly the situation of a fuel
switch. While the stakes in computerized humanities research are (joyfully)
other than life and passing, interface is regardless an essential segment of
any task.
To be sure, interface presents various fascinating and remarkable issues
for the advanced humanist. Justifiably determined by logical and utilitarian
necessities, the interface is additionally where portrayal and its chaperon
philosophies are generally prominent to our basic eyes. Apparently married
Innovations in Artificial Intelligence and Human Computer Interaction 131

to the ideal of ease of use, the interface is likewise where we send our most
innovative highlights and creative twists. Time after time set up as the last
period of an undertaking under a tight cut-off time and a much more tight
spending plan, the interface turns into the first and in quite a while the select
insight of the venture for its end clients. Apparently, the most innovative or
instinctive phase of the advancement interaction, the interface is likewise
conceivably the most observational, subject to the thorough quantitative ease
of use testing spearheaded by HCI. This section makes neither endeavor to
offer a thorough overview of the tremendous expert writing on interface and
convenience, nor does it try to fill in as a plan groundwork or manual for
best practices (Perusers intrigued by those themes ought to counsel the part’s
ideas for additional perusing.

8.5 BUSINESS INTELLIGENCE

Business intelligence is a bunch of techniques, cycles, models, and advances

that change crude information into significant and helpful data used to
empower more viable key, strategic, and functional experiences and dynamic.
This theory focuses on the business knowledge frameworks (BIS). BIS are
intelligent frameworks planned to help chiefs use correspondence advances,
information, reports, information, and insightful models to recognize and
tackle issues; accordingly, a BI framework can be known as a choice emotion­
ally supportive network (DSS).9 We focus on online analytical processing
(OLAP) apparatuses that empower clients to intuitively investigate multidi­
mensional information according to various points of view. The center of the
OLAP innovation is the information 3D square as shown in Figure 8.2, which
is a multidimensional dataset model. The model comprises of measurements
and measures. The actions are mathematical information like income, cost,
deals, and spending plan. Those are arranged by measurements, which
are utilized to bunch the information like the “bunch by” administrator in
social datasets. Commonplace measurements are time, area, and item, and
they are regularly coordinated in progressive systems. A progression is a
construction that characterizes levels of granularity of a measurement and
the connection between those levels. A period measurement can for instance
have hours as the best granularity and higher up the progression can contain
days, months, quarters, and a long time. Previously, business knowledge
frameworks were utilized to answer OLAP inquiries in table-based reports
that addressed various questions independently. The new age of BIS has two
132 Computational Imaging and Analytics in Biomedical Engineering

remarkable highlights such as coordination of various inquiry answers and

perception of these answers (for example, utilizing perception dashboards).
This work has been vigorously impacted by the modern climate on the BI act
of SAP Research. SAP is one of the market chiefs in big business application
programming, giving significant venture devices, for example, enterprise
resource planning (ERP), customer relationship management (CRM), and BI
frameworks like SAP crystal reports, SAP business objects dashboards, and
SAP business objects explorer. The fundamental objective of information
representation is its capacity to envision information, imparting data plainly
and successfully. It doesn’t imply that information representation needs to
look exhausting to be practical or incredibly complex to look delightful.
To pass on thoughts adequately, both tasteful structure and usefulness need
to go connected at the hip, giving experiences into a fairly inadequate and
complex data perception assumes a significant part in supporting informa­
tion investigation and sum up the fundamental attributes of an informational
index in a straightforward structure, frequently with visual designs or
outlines. It allows the insightful thinking cycle to turn out to be quicker
and more engaged.12 Representation research looks to enhance human
cognizance by utilizing human visual capacities to figure out theoretical data
nevertheless as clarified by James et al.,14 explicit sorts of visual portrayals
and connection strategies permit clients to see, investigate, and see a lot of
data on the double. In the space of multidimensional representation these
include scatterplot frameworks, multidimensional glyphs (like StarGlyphs
and Chernoff Faces), equal directions or starplots, multidimensional scaling,
even perception procedures and composed numerous perspectives. This
proposal focuses on facilitated and numerous perspectives like dashboards,
portrayed straightaway, as they are the most famous inside the BI space.
All things considered as clarified by James et al.,14 explicit sorts of visual
portrayals and connection methods permit clients to see, investigate, and
see a lot of data immediately. In the space of multidimensional represen­
tation these include scatterplot networks,15 multidimensional glyphs, (for
example, StarGlyphs16–18 and Chernoff Faces), equal directions19–21 or star-
plots, multidimensional scaling,22 plain perception procedures what’s more,
planned and numerous perspectives.25 This theory focuses on facilitated and
numerous perspectives like dashboards, portrayed straightaway, as they are
the most well-known inside the BI space. To be handled and examined, like
observing environment and climate, science, medication, and business. Our
work center is business visual analytics that guides in examining enormous
business informational indexes and allows taking right choices. By the by
Innovations in Artificial Intelligence and Human Computer Interaction 133

our emphasis is fundamentally on the communication and representation

challenges, and less on the programmed investigation strategies.
The primary objective of HCI is to make PCs, all the more user-friendly
by limiting the obstruction between the human’s objective of what they need
to achieve and the PC’s comprehension of the client’s undertaking. Perhaps
the main HCI plan techniques is user-focused plan Figure 8.3, in which
clients become the dominant focal point in the plan of any PC framework,
and is the plan approach we followed to comprehend our clients’ require­
ments and propose arrangements in this work.

FIGURE 8.8 Role of HCI and AI in buisness intelligence.

Human-focused plan measure characterizes an overall interaction for

including human-focused exercises all through an advancement life-cycle.
The four exercises are: (1) Specify the setting of utilization: identify indi­
viduals who will utilize the item, what they will utilize it for, and under what
conditions they will utilize it. (2) Specify necessities: identify any business
prerequisites or client objectives that should be met for the item to be effec­
tive. (3) Create plan arrangements: this piece of the interaction might be
done in stages, working from a harsh idea to a total plan. (4) Evaluate plans:
the main piece of this cycle is that assessment—preferably through conve­
nience testing with real clients—is, however, basic as quality testing may
be to acceptable programming improvement. A focal idea in HCI is that of
ease of use, that is, “The degree to which an item can be utilized by indicated
clients to accomplish determined objectives with adequacy, effectiveness,
134 Computational Imaging and Analytics in Biomedical Engineering

and fulfilment in a predefined setting of utilization.” Numerous techniques

exist for considering convenience, however, the most essential and helpful is
client trying, which has three segments as distinguished by Nielsen34: (1) Get
hold of some delegate clients, for our situation business intelligence special­
ists or tenderfoots. (2) Ask the clients to perform agent undertakings with
the plan. (3) Observe what the clients do, where they succeed, and where
they experience issues with the UI. This is the strategy we have followed
to work on the ease of use of the various models carried out during this
theory. The measure of information in our reality has been detonating, and
effectively breaking down enormous information gives an edge to the two
organizations just as people. BI betters break down enormous informational
collections, and one of the more significant apparatuses accessible to BI
examiners are perception dashboards. By and by, these dashboards stay hard
to make and modify by people with little preparing. Besides, they actually
miss the mark in supporting nitty gritty investigation record keeping, and
correspondence of examination results to other people. This postulation
centers around improving client cooperation with BI dashboards, through
(1) simple creation and customization of dashboard for a more extensive
client range, (2) adding novel explanation supports for examination, and (3)
utilization of visual narrating for imparting dashboard investigation results.
The propensity for growing dashboard end-client populace with assorted
necessities, requires clients, who are regularly tenderfoots in data percep­
tion, to get fast admittance to adaptable and simple to fabricate business
intelligence data dashboards. This is reflected in progresses in effectively
adjustable dashboard perception frameworks (for example Ref. [35,36]), it
corresponds with progresses in effectively available data perception condi­
tions for beginners, (for example, ManyEyes37), and is additionally reflected
in the expanding focal point of the InfoVis research local area on the most
proficient method to help amateur clients.38,39 By and by, there are as yet
numerous inquiries regarding how fledgling clients associate with genuine
data representation frameworks and that’s only the tip of the iceberg so with
dashboards. In the BI investigation space, programming merchants center for
the most part around adding more highlights into the following dashboard
discharge, with no an ideal opportunity to do the exploration expected to find
what really works, or even to step back and see how their items are truly being
utilized. To discover the responses to such inquiries, we assembled experi­
ences with respect to fledgling client rehearses on exploration views (EV), a
framework that permits beginner representation clients to handily construct
and tweak business intelligence data dashboards. EV gives a natural climate
Innovations in Artificial Intelligence and Human Computer Interaction 135

to progressively making, revamping, looking, and investigating numerous

visual information portrayals from different information sources. These
angles help clients to more readily recover, test, and acclimate themselves
with their information. We give new rules that expand past work on plan­
ning for perception beginners, with regards to dashboards. Only assortments
of visual portrayals don’t represent themselves; to become significant they
require understanding and clarification.
Because of the restrictions of human memory40 recollecting every single
thinking subtlety and achievements in the investigation interaction becomes
testing. We report dashboard constraints in investigation record-keeping,
information revelation, and social occasion experiences. We led top to
bottom meetings with dashboard specialists and revealed the need of sorting
out their representations by commenting on them with their discoveries. This
prompted new explanation needs for multi-outline perception frameworks,
on which we based the plan of a dashboard model that upholds information
and setting mindful comments that can be shared across representations
and datasets. We zeroed in especially on clever explanation perspectives,
for example, multi-target comments, comment straight forwardness across
graphs, just as joining comments to an information layer divided between
various information sources and BI applications, permitting cross applica­
tion comments. Albeit the work practices of BI experts’ might vary, in the
end they all develop stories that include: assortments of visual parts (dash­
boards, graphs, or tables) to help their accounts, verbal clarifications during
presentation(s), or itemized literary explanations and portrayals as composed
reports and outside assets, for example, wikis. This cycle stays awkward
and requires incessant exchanging between various programming (dash­
board investigation instruments, screen catches, content managers, show or
drawing devices, and so on) BI investigators need a simpler instrument to
share their examination results and recount stories with their information.
Our work means to help BI investigators with their narrating needs, to foster
high examination esteem. Our meetings show that their recorded experiences
and information should be coordinated and moved to other people, just as
stay accessible for future clients. To accomplish this, we need a planning and
sharing instrument to work with the exchange of information and encounters.
We accordingly gathered new necessities identified with BI visual narrating.
In light of these necessities, we planned and carried out a narrating model
apparatus that is coordinated in a dashboard investigation device and permits
simple progress from examination to story creation and sharing. Section 5
gives subtleties on this work.
136 Computational Imaging and Analytics in Biomedical Engineering

The measure of information in our reality has been detonating, and effec­
tively investigating huge information gives an edge to the two organizations
just as people. BI betters dissect huge informational collections, and one
of the more significant apparatuses accessible to BI experts is representa­
tion dashboards. By and by, these dashboards stay hard to make and alter
by people with little preparing. Besides, they actually miss the mark in
supporting point by point examination record keeping, and correspondence
of investigation results to other people. This postulation centers around
the upgrading client cooperation with BI dashboards, through (1) simple
creation and customization of dashboard for a more extensive client range,
(2) adding novel explanation support for investigation, and (3) utilization of
visual narrating for conveying dashboard examination results.

8.6 MARKET ANALYSIS

A market investigation is a quantitative and subjective appraisal of a market.

It investigates the size of the market both in volume and in esteem, the
different client portions and purchasing behaviors, the opposition, and the
financial climate as far as boundaries to passage and guideline.
The “Hyper-Converged Infrastructure (HCI) Market” report offers a
nitty-gritty outline of colossal development potential as far as income age
and this development factors that will affecting the worldwide business
scope. The report shows most recent market bits of knowledge, expanding
development openings, business techniques, and plans received by signifi­
cant players. Hyper-converged infrastructure (HCI) market report gives key
measurements on market size, share, and changing serious scene by covering
major topographical locales. It likewise offers an exhaustive examination of
market elements, current market patterns, item portfolio with SWOT and
store network investigation.
Hyper-merged foundation is a product characterized IT framework that
virtualizes the entirety of the components of ordinary “equipment character­
ized” frameworks. HCI incorporates, at the very least, virtualized figuring, a
virtualized SAN and virtualized organizing. HCI ordinarily runs on business
off-the-rack workers.
The fundamental factors that are driving the market development
incorporate expanding interest for information assurance and catastrophe
recuperation, profoundly versatile arrangements, and expanded utilization
for virtualization. In any case, disadvantage of packaged arrangements is the
Innovations in Artificial Intelligence and Human Computer Interaction 137

main consideration impeding the market. What’s more, developing acknowl­

edgment of hyper combination arrangements in coming years would be a
huge development opportunity for the market. This exploration study looks
at the current market patterns identified with the interest, supply, and deals,
notwithstanding the new turns of events. Significant drivers, restrictions, and
openings have been covered to give a comprehensive image of the market.
The investigation presents inside and out data with respect to the turn of
events, patterns, and industry arrangements and guidelines carried out in
every one of the geological locale. Further, the by and large administrative
structure of the market has been thoroughly covered to bring to the table
partners a superior comprehension of the key variables influencing the
general market climate.
The hyper-merged foundation market size was esteemed at $3.84 billion
of every 2018, and is projected to reach $33.16 billion by 2026, developing
at a CAGR of 30.7% from 2019 to 2026. Hyper-joined foundation (HCI)
is a solitary framework system that incorporates capacity, registering, and
network assets that assist organizations with effectively sending and make
do with a solitary UI.
Hyper-intermingling framework has acquired high foothold from the new
past among ventures to attract its capability to smooth out the sending of new
jobs, ease foundation the executives, and enhance framework costs. This has
additionally upheld the development of the hyper-combined foundation market
during the estimate time frame. HCI engineering is intended for a pay-more
only as costs arise cost model and can scale steadily which has emphatically
affected the development of the market. Business-basic applications which
are right now conveyed on three-level IT foundation will progress to hyper-
converged framework as it offers incorporated stack frameworks, coordinated
framework frameworks, and coordinated reference designs.
Among the business upward, the BFSI business overwhelmed the
worldwide hyper-joined framework market as far as income in 2018 and
is required to proceed with this pattern during the estimate time frame. As
the monetary area is exceptionally compelling toward meeting their client
needs. With rising digitization has provoked this area to convey incred­
ible computerized encounters. Accordingly, expansion of strong advanced
change systems currently incorporates hyper-combined foundation to build
its framework readiness. Besides, monetary firms are slanting toward HCI
to drive readiness and lower costs for the essential drives like dispersed it
and far off workplaces/branch workplaces, virtual work area arrangement,
business-basic applications, and calamity recuperation. This is a central
138 Computational Imaging and Analytics in Biomedical Engineering

point expected to drive the development of the hyper-met framework market

among the BFSI business. The hyper-united foundation market in North
America overwhelmed as far as income share in 2018. Ascend in shift toward
inventive innovation that gives cloud-like financial aspects to existing server
farms without the need to think twice about its presentation, accessibility, or
dependability is a main consideration that drives the selection of HCI market
in the district. This permit HCI adopters to observe worked on functional
productivity, versatility, and sped up arrangement time.
The report centers around the hyper-joined foundation market develop­
ment possibilities, limitations, and investigation. The investigation gives
Porter’s five powers examination of the hyper-combined framework industry
to comprehend the effect of different factors like dealing force of providers,
serious power of contenders, danger of new contestants, danger of substi­
tutes, and bartering force of purchasers available.
The hyper-joined foundation market division incorporates part, applica­
tion, industry vertical, and area. By segment, it is sorted into equipment
and programming. Based on application, it is partitioned into far off office
or branch office, virtualization work area framework (VDI), server farm
solidification, reinforcement recuperation or fiasco recuperation, basic
applications virtualization, and others. Contingent upon industry vertical, it
is sorted into BFSI, IT, and broadcast communications, government, medical
care, fabricating, energy and utilities, schooling, and others. In view of area,
the market is broke down across North America, Europe, Asia-Pacific, and
LAMEA.
The central members working in the worldwide hyper-combined foun­
dation market gauge incorporate Cisco Systems, Inc., Dell Inc., Hewlett
Packard Enterprise Company, Huawei Technologies Co., Ltd., Microsoft
Corporation, NetApp, Inc., Nutanix, Inc., Pivot3, Scale Computing, and
Vmware, Inc. Significant players working in this market have seen high
development sought after for hyper-combined foundation, particularly due
to on-going server farm modernization tasks or drives among server farm
industry across the globe. This investigation incorporates hyper-united foun­
dation market examination, patterns, and future assessments to decide the
impending venture pockets. Developing requirement for information insur­
ance and fiasco recuperation and decrease in capital expenditure (CAPEX)
and operational expenditure (OPEX) are central point driving the develop­
ment of the hyper-combined framework market patterns. These components
are credited to the development of the market as significant foundation
merchants are embracing half and half and multi-cloud organizations as either
Innovations in Artificial Intelligence and Human Computer Interaction 139

reinforcement systems or information calamity recuperation choices or as an

option for on-premises framework. Nonetheless, seller lock-in is a central
point expected to frustrate the market development somewhat. Moreover,
ascend in interests in server farm foundations is an entrepreneurial factor of
the market.
Hyper-merged foundation (HCI) is a solitary framework structure that
incorporates capacity, processing, and organization assets that assist orga­
nizations with effectively sending and deal with a solitary UI. Significant
drivers for the development of hyper-converged infrastructure market size
incorporate benefits like diminished capital venture and working consump­
tion and adaptability for recuperation from fiascos.
This investigation incorporates the market opportunity, examination,
patterns, and future assessments to decide the fast approaching venture
pockets. The hyper-converged infrastructure market report presents key
drivers, restrictions, and openings for the hyper-joined foundation industry.
The HCI market in North America overwhelmed as far as income share
in 2018. Ascend in shift toward inventive innovation that gives cloud-like
financial aspects to existing server farms without the need to think twice
about its presentation, accessibility, or dependability is a central point that
drives the reception of HCI market in the district. This permits the HCI
adopters to observe worked on functional productivity, adaptability, and
sped up sending time.

FIGURE 8.9 Fundamental requirements for market analysis.

140 Computational Imaging and Analytics in Biomedical Engineering

8.7 CONCLUSION

Portraying the use of a PC as a semiotic interaction among people and PCs is

both expressive and takes into account considerable generative force.
Considering the semiotic viewpoint, it could bring about an assortment of
undertakings on the crossing point of AI and HCI that have both more acces­
sible techniques and a more profound hypothetical establishment. These
could be utilized to acknowledge Douglas Engelbart’s vision of “working on
the compelling usage of the scholarly force of society’s concern solvers.”1
This could appear as client focused, assessment zeroed in probes frameworks
that use the method for man-made reasoning, design acknowledgment, and
information science. Exploratory information investigation, as portrayed
here Ref. [12], could turn into an appropriate strategy both as a technique and
as a subject for collaboration plan. Particularly, since the quantity of acces­
sible sensors and the delivered information are quickly expanding. There is
now research that intends to empower and work with a mutual perspective of
the computerized material.18 This is a decent beginning stage, despite the fact
that it is as yet restricted to materials. In a more all-encompassing manner,
the explorative methodology could be reached out to entire themes and
examination questions. An exploratory information investigation approach
could likewise help solid ideas. Solid ideas are contestable, faultless, mean­
ingful, and conceptual examples.5 The information science toolset could help
recognize and validate solid examples like social route, for example, by
imagining multidimensional information, uncovering relationships, and
empowering exploratory information investigation. All the more by and
large, an association of AI and HCI could empower clients to all the more
effectively and altogether investigate an assortment of data sources and
information streams. A vital interdisciplinary examination project is “The
Human Speechome Project Symbol Grounding and Beyond.” The Spee­
chome utilized varying media chronicles for a longitudinal report on language
advancement of a solitary kid. The information was then examined to see
more about language securing.15 This can be utilized for research; however,
it could likewise be applied to educate the lives regarding individuals, for
example, concerning their own wellbeing and their accounts. Lieberman
proposes that penmanship and voice acknowledgment will turn into a signifi­
cant application area for both HCI and AI innovation.7 Despite the fact that
they acquired an awful standing as the original of executions didn’t work,
Moore’s law will ultimately make factual language models sufficiently
incredible to dependably settle penmanship and voice acknowledgment.7
Innovations in Artificial Intelligence and Human Computer Interaction 141

Profound learning research previously showed that it is feasible to prepare a

face indicator without marking pictures as containing a face or not and that
solo learning is able to do consequently learning significant level ideas, for
example, feline countenances and human bodies.6 There have been different
distributions with respect to intelligent AI.2,17,19 Intuitive AI was proposed as
“a characteristic method of incorporating foundation information into the
demonstrating stage.”19 One of the objectives was to empower clients to
“work connected at the hip with AI frameworks.”17 This was investigated
utilizing a verbally process study,17 a technique generally utilized in HCI.
The collaborations of man-made consciousness and human–computer
connection, particularly intelligent AI, could turn out to be extremely impor­
tant in clinical and business applications. It could likewise turn into a pivotal
instrument in acknowledging Seymour Papert’s fantasy of “computational
familiarity for everybody.”14 Resnick recommended that this will require
another age of innovations, exercises, and instructive techniques.14 Papert
evoked the idea of a Mathland, where numerical reasoning is regular—like
communicating in French is normal in France.13 Papert composed that while
kids might have issues learning French in French class in a US school, all
youngsters in France will ultimately learn French. Such a Mathland could be
acknowledged utilizing intelligent AI. The framework could both evaluate
how much a client definitely thinks about programming (which equivalents
to how enormous their jargon is) and what the person is possible attempting
to accomplish (which equivalents to their goals and which is like how a
human language educator poses the right inquiries to direct an understudy).
Like in human interaction and correspondence, the best associations depend
on a profound comprehension of one another and effectively gathering the
inward state just as the aims of the other. Like an instructor, who can say for
sure what an understudy comprehended and what the person didn’t and who
likewise realizes how to help that person advance (a Socratic thought of the
educator as a birthing specialist). Measurable AI strategies could give clients
the perfect measure of help and help and empower designers to make better
client encounters. Proactive and strong innovation could become accommo­
dating in close to home wellbeing following, brilliant home administration,
individual accounting the executives just as interdisciplinary science. Be that
as it may, it could likewise make programming simpler and significantly
further develop training. Computer-based intelligence and HCI could
commonly profit with a nearer joint effort. The AI people group could build
their utilization of HCI techniques like the Thinking so anyone might hear
convention and Wizard of Oz prototyping and consider social constructivist
142 Computational Imaging and Analytics in Biomedical Engineering

positions. HCI could additionally expand human capacities utilizing new

advancements. The two disciplines can be hypothetically established by
semiotic hypothesis. An interdisciplinary group of AI and HCI scientists
could foster uncommon applications, for example, by further developing
schooling and making homes more brilliant. Since the time PCs were
conceived, there have been different connections among individuals and PCs
to make PCs more receptive to the people’s necessities. The consistent
improvement in human interest and interest drives the advancement of HCI
innovation and astute robot advances. A lot of exploration works has been
completed to make HCI more regular and amicable and, simultaneously,
make robots more wise and versatile. With the quick improvement of AI
innovation as of late, it gives uncommon advancement freedoms to the
examination of these two advances. This paper sums up the advancement
status of HCI from the part of communication capacities and presented the
connected advances of a keen robot. From there on, the difficulties for these
two fields later on improvement and conceivable examination approaches
are elucidated. The conversation introduced in this segment is wide and
illustrative, however, not totally comprehensive. Some encouraging employ­
ments of AI methods to fabricate great arrangements were not unequivocally
examined, and may be viewed as when the goal is to carry out some specific
viewpoints in game openness. Some illustration of such AI strategies and
their uses are: (1) Both artificial neural network and case-based reasoning
should be thought of if the answer for an issue can be situated in past encoun­
ters. The previous is proper in non-emblematic or sub-emblematic settings,
while the last ought to be utilized when using emblematic portrayal is
possible; (2) Evolutionary computing is appropriate to take care of issues in
which there are a bunch of boundaries to be optimized; (3) Mappings and
forecast issues are very much addressed for certain particular artificial neural
networks architectures; (4) Fuzzy approximated reasoning is appropriate to
manage vulnerability—a typical component found in the human conduct
presentation and investigation—and to perform fine settings in streamlining
and planning process. Figure 8.1 sums up, through a connection lattice, the
chances of utilizing AI to execute solutions to make client communication in
games more open. Lines continue the issues—in regards to the HCI scope—
proposed through the conversation in this segment; sections list the AI
procedures early chose.
The space of assistive innovations advancement identified with games,
particularly in regards to the utilization of AI to enhance the legitimate assets
of client interaction with the different kinds of uncommon necessities, actually
Innovations in Artificial Intelligence and Human Computer Interaction 143

requires thorough investigations which can direct endeavors to innovative

work of accessible games. One of the possible spaces of work is the guide
ping of the genuine cooperation between gamers with extraordinary necessi­
ties and existing games, both for the exemplary games and for those which as
of now have openness highlights. Instances of ongoing works are introduced
by Zaharias and Papargyris (2009) and Kort and Ijsselsteijn (2008), which
albeit not zeroed in on gamers with extraordinary requirements, fill in as
motivation for research that intends to more readily comprehend the view­
points identified with the accessibility world. The previous examinations the
connections between gamers’ way of life and games convenience; and the last
talks about how games can impact the social associations of gamers in the
genuine world. Specifically in the setting examined in this section, there are
two regions which actually need of strongholds to arrive at worthy degrees
of value concerning availability: ressions, body developments, or discourse
construction (oral, composed, or gestured); (5) Translation/planning between
various sorts of dialects and portrayals: oral language to gestural language;
composed language to pictographic language; chart is portrayal to sound
portrayal; complex discourse to basic discourse.
The primary goal of this part was to examine a few viewpoints identi­
fied with game availability, high-lighting how the HCI and AI regions can
manage this inquiry. In this, contents were examined a few perspectives
which conceivably address a hindrance for the adequate utilization of a game
by a client with extraordinary requirements, and some potential answers for
defeat such difficulties. In reality, the AI region as of now gives different
instruments which could be applied to implement the uncommon highlights
in games, in any case, the endeavors to utilize them viably for the accessible.

KEYWORDS

• HCI
• AI
• framework
• frequency

REFERENCES

1. Wang, T. M.; Tao, Y.; Liu, H. Current Researches and Future Development Trend of
Intelligent Robot: A Review. Int. J. Automat. Comput. 2018, 15 (5), 525–546.
144 Computational Imaging and Analytics in Biomedical Engineering

2. Abdelaziz, A. H. Comparing Fusion Models for DNN-Based Audiovisual Continuous

Speech Recognition. IEEE/ACM Trans. Audio Speech Lang. Process. 2018, 26 (3),
475–484.
3. Ren, S.; He, K.; Girshick, R.; Zhang, X.; Sun, J. Object Detection Networks on
Convolutional Feature Maps. IEEE Trans. Pattern Anal. Mach. Intell. 2017, 39 (7),
1476–1481.
4. Young, T.; Hazarika, D.; Poria, S.; Cambria, E. Recent Trends in Deep Learning Based
Natural Language Processing. IEEE Computat. Intell. Magaz. 2018, 13 (3),55–75.
5. Tirri, K.; Nokelainen, P. Measuring Multiple Intelligences and Moral Sensitivities in
Education; SensePublishers, 2011.
6. Kim, J.; El-Khamy, M.; Lee, J. Residual LSTM: Design of a Deep Recurrent Architecture
for Distant Speech Recognition, 2017. arXiv preprint arXiv:1701.03360
7. Chen, Y. N.; Wang, W. Y.; Gershman, A.; Rudnicky, A. Matrix Factorization with
Knowledge Graph Propagation for Unsupervised Spoken Language Understanding.
ACL-IJCNLP 2015, 1, 483–494.
8. Yu, A. W.; Dohan, D.; Luong, M. T.; et al. QANet: Combining Local Convolution
with Global Self-Attention for Reading Comprehension, 2018. arXiv preprint
arXiv:1804.09541.
9. Liu, F.; Flanigan, J.; Thomson, S.; Sadeh, N.; Smith, N. A. Toward Abstractive
Summarization Using Semantic Representations. arXiv preprint arXiv:1805.10399.
10. Carlisle, J. H. Evaluating the Impact of o±ce Automation on Top Management
Communication, In Proceedings of National Computer Conference and Exposition,
1976; pp 611–616.
11. Card, S. K.; Moran, T. P.; Newell, A. The Keystroke-Level Model for User Performance
Time with Interactive Systems. Commun. ACM 1980, 23 (7), 396–410.
12. Hewett, T. T.; Baecker, R.; et al. ACM SIGCHI Curricula for Human-Computer
Interaction. ACM 1992. DOI: 10.1145/2594128.
13. D'Amico, G.; Bimbo, A. D.; Dini, F.; et al. Natural Human-Computer Interaction, In
Multimedia Interaction and Intelligent User Interfaces; Springer: London, 2010; pp
85–106.
14. Wang, H. A.; Tian, F. Natural and e±cient Human Computer Interaction, In 10000
Selected Problems in Sciences; Science Press: Beijing, 2011; pp 625–627.
15. Cao, X. What is the Natural Characteristics of Natural User Interface. Commun. China
Comput. Federat. 2011, 11, 14–18.
16. Legg, S.; Hutter, M. A Collection of de¯nitions of Intelligence. Front. Artif. Intell. Appl.
2007, 157, 17–24.
17. Gottfredson, L. S. Mainstream Science on Intelligence: An Editorial with 52 Signatories,
History, and Bibliography. Intelligence 1997, 24 (1), 13–23.
18. Neisser, U.; Boodoo, G.; Bouchard, T. J.; et al. Intelligence: Knowns and Unknowns.
Am. Psychol. 1996, 51 (2), 77–101.
19. Ren, F.; Bao, Y. Int. J. Info. Tech. Dec. Mak. 2020, 19, 5–47. www.worldscientific.com
20. Matthews, G.; Zeidner, M.; Roberts, R. D. In The Science of Emotional Intelligence:
Knowns and Unknowns; Oxford University Press: New York, NY, 2007.
21. Gardner, H. Frames of Mind: The Theory of Multiple Intelligences. Quart. Rev. Biol.
1985, 4 (3), 19–35.
22. Legg, S.; Hutter, M. Universal Intelligence: A de¯nition of Machine Intelligence, In
Minds and Machines; Kluwer Academic Publishers, 2007; vol. 17 (4), pp 391–444.
Innovations in Artificial Intelligence and Human Computer Interaction 145

23. Zhou, H.; Huang, M.; Zhang, T.; Zhu, X.; Liu, B. Emotional Chatting Machine:
Emotional Conversation Generation with Internal and External Memory, 2017. arXiv
preprint arXiv:1704.01074.
24. Ren, F.; Huang, Z. Automatic Facial Expression Learning Method Based on Humanoid
Robot XIN-REN. IEEE Trans. Hum. Mach. Syst. 2016, 46 (6),
25. Minsky, M. L. In The Society of Mind; Simon & Schuster Press, 1988.
26. Picard, R. W. A®ective Computing; MIT Press, 1997.
27. Nagamachi, M. Kansei/A®ective Engineering; CRC Press, 2011.
28. Tu, X. Arti¯cial Emotion, The Paper Assembly of the 10th Annual CAAI; Guangzhou,
China, 2000.
29. Wang, Z.; Xie, L. Arti¯cial Psychology-an Attainable Scienti¯c Research on the Human
Brain; IPMM: Honolulu, 1999; vol 2, pp 1067–1072.
30. Wang, Z. Arti¯cial Psychology and Arti¯cial Emotion. CAAI Trans. Intell. Syst. 2006,
1 (1), 38–43.
31. Ledoux, J. E. Emotion Circuits in the Brain. Ann. Rev. Neurosci. 1999, 23 (23), 155–184.
32. Cardinal, R. N.; Parkinson, J. A.; Hall, J.; et al. Emotion and Motivation: The Role of the
33. Amygdala, Ventral Striatum, and Prefrontal Cortex. Neurosci. Biobehav. Rev. 2002, 26
(3) 321–352.
34. Hossain, M. S.; Muhammad, G. Audio-visual Emotion Recognition Using
Multidirectional Regression and Ridgelet Transform. J. Multimodal User Interfaces
2016, 10 (4), 325–333.
35. Shaver, P.; Schwartz, J.; Kirson, D.; et al. Emotion Knowledge: Further Exploration of
a Prototype Approach. J. Pers. Soc. Psychol. 1987, 52 (6), 1061.
36. Nicolaou, M. A.; Zafeiriou, S.; Pantic, M. In Correlated-Spaces Regression for Learning
Continuous Emotion Dimensions, ACM International Conference on Multimedia, 2013;
pp 773–776.
37. Barrett, L. F. Discrete Emotions or Dimensions? The Role of Valence Focus and Arousal
Focus. Cogn. Emot. 1998, 12 (4), 579–599.
38. Kamarol, S. K. A.; Jaward, M. H.; Kälviäinen, H.; et al. Joint Facial Expression
Recognition and Intensity Estimation Based on Weighted Votes of Image Sequences.
Pattern Recognit. Lett. 2017, 92 (C), 25–32.
39. Mehrabian, A. Pleasure-Arousal-Dominance: A General Framework for Describing
and Measuring Individual Di®erences in Temperament. Curr. Psychol. 1996, 14 (4),
261–292.
40. Breazeal, C. Function Meets Style: Insights from Emotion Theory Applied to HRI.
IEEE Trans. Syst. Man Cybern. C 2004, 34 (2), 187–194.
41. Ren, F.; Quan, C.; Matsumoto, K. Enriching Mental Engineering. Int. J. Innov. Comput.
Inf. Control 2013, 9 (8), 3271–3284.
42. Xiang, H.; Jiang, P.; Xiao, S.; et al. A Model of Mental State Transition Network. IEEJ
Trans. Electron. Inf. Syst. 2007, 127 (3), 434–442.
43. Mathavan, S. A.; et al. An Emotion Recognition Based Fitness Application for Fitness
Blenders. Euro. J. Mol. Clin. Med. 2020, 7 (2), 5280–5288.
44. Ren, F. A®ective Information Processing and Recognizing Human Emotion. Electron.
Notes Theor. Comput. Sci. 2009, 225, 39–50.
CHAPTER 9

COMPUTER-AIDED AUTOMATIC
DETECTION AND DIAGNOSIS
OF CERVICAL CANCER BY USING
FEATURE MARKERS
P. SUKUMAR1, R MURUGASAMI1, A. RAJAN2, and S. SHARMILA3
1
Department of Computer Science and Engineering, Nandha
Engineering College (Autonomous), Erode, Tamil Nadu, India
Department of ECE, Sree Rama Engineering College, Thirupathi,
2

Andhra Pradesh, India

3
Department of Civil Engineering, Nandha Engineering College, Erode,
Tamil Nadu, India

ABSTRACT

A computer-aided automatic detection and diagnosis method for cervix

cancer using Pap smear image is described in this chapter. Cervix disease is
the foremost reason behind feminine genital cancers and also severe reason
for feminine disease death around the world. A papanicolaou smear is a
process performed by the therapist during which a trail of cells are taken
from the cervix-uteri using a minor swab and examined for any abnormal
microscopic appearances due to HPV infections. The proposed methodology
constitutes the following stages: preprocessing, feature extraction, nuclei
region segmentation, and classification. Morphological operations are
used to segment the nuclei cell region. The features are extracted from the

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
148 Computational Imaging and Analytics in Biomedical Engineering

re-processed papnicolaou spot cell image. False error rate can be decreased
by using the automated process.

9.1 INTRODUCTION

Cancer sickness is most common in women affecting the genital region

characterized by cervical tumor, with an increasing incidence in the last
periods. Cervix cancer deaths occur most normally among women aged
between 16 and 43 years. The region of cervix, also called uterine cervix,
is located in the lower part of the uterus (in women). It connects the
body of the uterus, that is, endo-cervix with the birth canal by means of
exo-cervix.
The cervix is covered with cells namely, squamous cells and the glandular
cells. Cervical cancers originate in the core layer of the cervix, intersection
of the vagina and at the uterus due to Human Papilloma Virus (HPV) mostly
transmitted through sexual intercourse from male to female. Two common
types of cervical cancers exist, namely, adenocarcinoma and squamous cell
carcinoma, such that 80–90% of the cervical cancers occur due to the squa­
mous cell carcinoma, which begins at the junction of the endo-cervix and the
exo-cervix.
Women who have had more than three full-term pregnancies, and 17 years
aged or younger women while they are at their first full-time pregnancies are
subjected to a higher threat of having cervix cancers as per the Statistics of
Indian Cancer Society. Even, passive smoking, which damages the DNA
around the cervical cells, is considered a cofactor for getting the hazard of
cervix cancer in women.
The symptoms of cervical cancer are abnormal vaginal bleeding, post­
menopausal bleeding, pelvic pain, etc. In advanced stages of cervical cancer,
woman may experience bone fractures around the pelvic area, vaginal
discharge with blood, leakage of urine, etc.
Surgery is the primary treatment in the majority of women with endome­
trial cancer. Radiotherapy is also used when adjuvant therapy is indicated.
However, indication for adjuvant therapy varies from country to country.
Chemotherapy and hormonal treatment are only given in clinical trial
contexts or as palliation in case of an advanced stage.
Cervical cancer is generally diagnosed using Pap smear test and women
being screened for at least once in their lifetime at an age of 30–40 may
reduce their risk of cancer by 25–36%. Papanicolaou smear is a screening
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 149

test for cervix cancer in women used to detect nonvisible Human Papilloma
Virus (HPV) infection.
A papanicolaou smear is a process performed by the therapist during
which a trail of cells are taken from the cervix-uteri using a minor swab and
examined for any abnormal microscopic appearances due to HPV infections.
Papanicolaou smear test is done either at the age of eighteen or at earlier.
Papanicolaou smear test should be done 1–5 years depending on the women’s
risk factor. Smoking may make cervical cells more susceptible to infection
and the development of cervical cancer because the carcinogens in cigarettes
pass from the lungs into the blood stream and into the cervical mucus.
A woman’s socioeconomic status is important because it is an indicator
of her access to health care. Women from lower socioeconomic backgrounds
tend to have less access to basic healthcare, which decreases the chance that
abnormal cervical cell changes will be caught before they lead to cancer.
Women with first-degree relatives who have had breast cancer are more
likely to perform breast self-examinations than are women without a family
history of breast cancer. Daughters whose mothers have had breast cancer are
also more likely to be involved in the medical setting and are more likely to
seek out relevant information. Such women are more likely to have physical
examinations and to have such examinations more frequently.
No doubt the daughters of these women are also less informed about
Pap smears since mother–daughter communication surrounding anything
perceived to be sexual is likely to be low. Generational acculturation may
also influence mother–daughter communication; as a generation becomes
more acculturated they are more likely to take on the morals/values of
the host country. Open and candid discussions about sex or the body may
become more frequent when a family has lived in this culture for two or
more generations.
Studies show that women, in general, are severely undereducated about
the Pap smear and about HPV. The purpose of the Pap smear is poorly under­
stood. Some women think a Pap smear screens for Sexually Transmitted
Infections (STIs) other than HPV, for HIV, or for pregnancy. Others think
that the test screens for reproductive tract cancers other than cervical cancer
such as ovarian, endometrial, and uterine cancer.
Both age and education level influence women’s perceptions of the
importance of Pap smears. As a woman ages, she is more likely to believe
that Pap smears are important to her reproductive health.
This is even more obvious as women’s education levels increase. Adoles­
cents and older women with minimal education are more likely than more
150 Computational Imaging and Analytics in Biomedical Engineering

educated women to fear Pap smear testing and to have more misconceptions
about the purpose of Pap smears.
The detection of cervical cancers from the papanicolaou smear images is
a challenging task in medicinal image processing. This can be improved in
two ways. One way is by selecting suitable well-defined exact features and
the other is by selecting the best classifier.
Several spontaneous and semi-spontaneous methods have been suggested
in various times to identify various periods of cervical cancer. These
methods were not supported in attaining the purposes of providing dignified
variables which could eradicate the clarification errors and interobserver
discrepancy.

9.2 OBJECTIVES OF THE STUDY

• To develop an automated process detection process using medical images.

• To apply multi-resolution Gabor transform for cervical cancer detection.
• False error rate can be decreased by using automated process.
• To improve the performance evaluation parameters.
• Validate the results obtained.

9.3 LITERATURE SURVEY

9.3.1 GENERAL

There are lots of researches dealing with general diagnosis of cervical

cancers. In this chapter, we have considered some papers which presented
various diagnosing techniques for cervical cancers, including pap smear
segmentation, texture-based detection, etc. In our literature survey, we go
over some related works that account for high efficient cervical cancer
diagnosis schemes as well as its performance in classification which are in
proximity to our proposed concept.

9.3.2 TECHNIQUES FOR CERVICAL CANCER DETECTION

Carcinoma formation in the internal lining of cervix. Various stages of

cervical cancer can be identified from the location of infection and the extent
to which the region is affected.
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 151

Varghese et al. (2022) fixated on learning the association between DNA

methylation and miRNA expression, which had provided new visions into
the pathogenesis of cervical cancer. DNA-based methylation markers, along
with usual screening technologies, may help in improving the exactness
of the current selection techniques used for analysis and prognosis of the
disease. Taken together, recognized DNA promoter methylation, foremost to
its cancer causing agent, can be a conceivable mark for analysis and remedy.
Revisions interrogating the translational impact of methylation in early
detection of cervical cancer may help in a better prediction for the afflicted
individuals, leading to longer survival for cervical cancer patients.
The suggested system consists of the resulting stages as preprocessing,
Gabor transform, feature extraction, and classification. Gabor transform
is used to convert the time domain cervical image into multi-resolution
domain and then the watershed subdivision technique is used to categorize
the abnormal regions from the cervical image. Structures are detached from
the segmented region and classified using ANFIS. Morphological processes
are identified to spot the cancer area in the image. The planned automation
technique for cervical cancer detection is applied on the publicly available
datasets. The planned system achieves exactness of 99.5% with respect to
ground truth images.
Mariarputham and Stephen (2015) presented texture-based representa­
tion. They used texture characteristics and selected grading. Infectious cell
is characterized based on parameter analysis. The set of tools was used to
extract the information like local binary pattern and histogram analysis.
The output of support vector machine provided accuracy rate of carcinoma
84.10% for moderate dysplasia
Edwin et al. (2015) used neural networks and SVM classifier to screen
the disease using Pap smear cell images. Average precision 92.33% precision
is obtained in this paper. Oscanoa et al. (2015) have proposed an efficient
recognition method of ecto-cervical cell employed.
Jusman et al. (2014) have studied the techniques, their benefits and
drawbacks. The alphanumeric data of the selection procedures were used
as data for the processor rerun system as changed in the skilled study. Four
phases of the processer system used in their study were enhancement, feature
extractions, feature assortment, and categorization. The computer classifi­
cation based on spectra data achieved better exactness data. The exactness
in classification using neural networks was 78.7%. The overall shows that
cytology structures and the features provided advanced exactness of about
90%.
152 Computational Imaging and Analytics in Biomedical Engineering

Huagng et al. (2014) have established a procedure for segmenting nucleus

and cytoplasm counters. Efficient system categorizes the papanicolaou smear
cells into anyone of four dissimilar categorizations of classes using SVM.
Two trials were conducted to authenticate the sorting concert which showed
the best routine outputs. In the principal experiment, the outcomes presented
that average truths of 97.16% were obtained individually, for separating
dissimilar varieties of cells. In another trial, 70% of the cell pictures were
used for training and 30% for investigation, attaining an exactness of 96.12%
An algorithm for Pap smear cell nuclei segmentation was proposed which
uses an elastic segmentation algorithm for determining the exact shape of the
nuclei. Then, noise is removed filters, and edges are extracted with a Canny
edge detection algorithm.
Finally, a randomized Hough transform processed by a level set algo­
rithm is applied to find the candidate nuclei. The segmentation results were
higher than conventional methods. In Bamford et al. (1996), morphological
watersheds were applied over gray scale images of low firmness and results
in the identification of the locality of remote cells in each image, whereas
cell nuclei were not detected.

9.4 METHODOLOGY

9.4.1 GENERAL

The proposed system constitutes reprocessing, feature succession, and nuclei

partition. Internal structure processes are used to partition cell center area.
The Gray equal, wavelet, and GLCM structures are mined from standard
plus diseased cell center.
The mined structures are proficient and categorized using ANFIS catego­
rization. The planned method attained 98.74% exactness for dysplasia cell
subdivision.

9.4.2 MATERIALS AND METHODS

The DTU/Herlev Pap Smear Database (2005) contains papnicolaou-spot

facts attained from images of fit and tumorous spot coming from the Herlev
University. Images are remained by several investigators for their certain
revisions.
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 153

9.4.3 METHODS

The planned papnicolaou spot cell subdivision process contains a repro­

cessing process primarily. The reprocessing is completed to eliminate the
noises and then improve the better particulars of the image. Formerly,
features are mined from the reprocessed image in the training method from
regular normal and abnormal cell images.
These features are used to train the ANFIS categorization in the training
mode as shown in Figure 9.1. Figure 9.2 labels the testing mode of catego­
rization in which the papnicolaou spot image is categorized into normal or
dysplasia based on its mined features.
The dysplasia image attained is exposed to morphological processes for
nuclei subdivision and finally the routine of subdivision is assessed.

FIGURE 9.1 Dysplasis nucleus segmentation in training mode.

FIGURE 9.2 Dysplasis nucleus subdivision in categorization method.

9.5 FEATURE EXTRACTION

The features are extracted from the re-processed papnicolaou spot cell
image. Features are used to distinguish the usual and irregular papnicolaou
spot images for cervical cancer identification.
154 Computational Imaging and Analytics in Biomedical Engineering

Local binary pattern, GLCM, wavelet, and Law’s consistency structures

are extracted, purpose of analysis of cervical cancer in papnicolaou spot
images.

9.5.1 LOCAL BINARY PATTERN OPERATIVE

LBP is extremely skilled operative employed created on texture features. LBP

label examines image element in an image by thresholding the adjacent pels
element of each image element and substitutes the image element value with
a binary numeral. It is added forcefully to mono gray-scale variations initi­
ated by radiance differences. LBP operator has a computational simplicity;
henceforth potential for examining the images in the present access.

9.5.1.1 ORIGIN OF LBP OPERATIVE

In difference to the LBP using 8 image elements in a 3 × 3 pels block,

this general construction of the operative sets no boundaries to the size of
the adjacent region or to the amount of sample points. Reflect a monoto­
nous image f (x, y) and gp represents the gray level of a subjective picture
element. Furthermore, let gp represent the rate of sample point which is
equally set apart round adjacent sample facts in addition to that range R
surface of (x, y)

=sp I ( xy,=
yt ), s 0,......s − 2 (9.1)

x + R sin(2πp / Y )
xy = (9.2)

y − s cos(2πp / y)
ys = (9.3)

=p I ( xy,=
yt ), p 0,......P − 2 (9.4)

x + R cos(2πp / Y )
xp = (9.5)
The boundary operators are operated in the monotonous image to obtain
the expected value.
Standard operative is resultant from dual circulation. It is obtained from
threshold changes based on rules. The standard operative is
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 155

∑ (9.6)
p=1
, R ( g c , yc )
LBPP= p=0
s ( s p − sc )2 p

9.5.1.2 LBP PROCESS OVER DIRECTIVE PIXEL AREA

LBP operative describes a specific area roughness by 2 balancing procedures,

namely, dull scale contrast. Unique LBP operative thresholds the 3 × 3 near
image element around the center image element, thereby producing separate
tags for each adjoining picture element in the image and changing the picture
element with an expected numeric. Thus, the statistical principles based on
28 = 256 sole markers are used as texture signifier.

FIGURE 9.3 LBP reckoning operation.

9.5.1.3 PROCEDURE FOR LBP RECKONING OPERATION

Step 1: Original image is used for analysis in the order of three rows and
three columns
Step 2: Location of sub-image based on medical image applications
Step 3: Conversion of sub-image into mathematical transformation
Step 4: Values attained based on the transformation results
Step 5: Transformation results attained based on mathematical formulation
principle.
156 Computational Imaging and Analytics in Biomedical Engineering

By altering the sum of picture element in the adjacency and its radius
using a round adjacency and bilinear inserting values, the LBP might be
applied for dissimilar sized adjacency area. The gray-scale differences in the
local nearest might be used as the consistent difference portion. The picture
element adjacency is based on adjacent points.
Figure 9.3 shows the operational procedure of Reckoning operation.
Original image is subdivided into a number of segments as per the matrix
computation. Values are considered based on intensity measures as per the
operational procedure with 3 × 3 matrix.
The center pixel value is 54; it is compared with the other values in the
matrix. The values greater than center pixel value should be considered as
binary number high component as 1. The values lesser than center pixel
value should be considered as binary number low component as 0.
Now the sub-image value is converted into binary value as 1 and 0.
The intensity range of pixels varies based on the present value. The matrix
block contains 1 as high resolution and the matrix block contains 0 as low
resolution.
Sub-image is converted into binary values and it is converted into decimal
value. Before the conversion of decimal operations, in clockwise order the
values are considered 1,101,011 by converting into decimal value as 203.

H
= s ∑ x, y
I { fl (k ,=
z ) b=
} s 0,1,...., s −1 (9.7)
where, “n” is the numeral of LBP labels.

9.5.2 GLCM FEATURES

The co-occurrence structures can be mined from each co-occurrence medium.

Co-occurrence features are energy, contrast, correlation, and homogeneity.
GLCM Medium at 45 degree.

1 5 3 4

2 2 4 1

3 4 5 5

4 2 1 2
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 157

The data are modeled based on repeated occurrence from the original
medium. It is calculated based on specific diagonal parameters. Contrast is
computed using this technique. The comparisons between first order to first
order, second order to second order, third order to third order, fourth order
to fourth order, and finally fifth order to fifth order are framed in the matrix
of computation and the expected values are attained based on the above
principles.

GLCM 1 2 3 4 5

1 0 1 0 0 0

2 0 0 0 1 1

3 1 1 0 0 0

4 1 0 0 0 1

5 0 1 0 1 0

Contrast
• It measures the strength distinction among a picture element and its
nearest.
Energy
• Yields the amount of shaped origins in the specific matrix.
Homogeneousness
• Yields a charge that actions the nearness of the circulation of basics in
the array of element.
Recorrelation
• Yields a portion of how connected a picture element to nearest in
comparison with the original values.
Law’s Energy Texture Features
• Law’s texture energy measures are derived from three simple vectors
of length 3,
• L3 = (1,2,1), E3 = (−1,0,1), S3 = (1,2, −1).
• Convolving these features.
158 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 9.4 Trained pap smear cell images (normal case).

FIGURE 9.5 Trained pap smear cell images (abnormal-dysplastic cell).

FIGURE 9.6 (a) Dysplasia Papanicolaou smear cell image. (b) Dysplasia cell subdivided
image using planned technique. (c) Dysplasia cell physically subdivided image.
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 159

9.5.3 PERFORMANCE EVALUATION PARAMETER

• Sensitivities [Ses = Tpos / (Tpost + Fnegt)]

• Specificities [Sps = Tneg / (Tnegt + Fpost)]
• Positive predictive values [Ppvs = Tpost / (Tpots+Fpost)]
• Negative predictive values [Npvs=Tnegt / (Tnegt+Fnegt)]
• Accuracies [Accs=(Tpost+Tnegt) / (Tpost+Fnegt+Tnegt+Fpost)]

TABLE 9.1 Performance Evaluation for Dysplastic Cell Segmentation.

Performance evaluation parameters Quantitative results (%)
Sensitivity 92.68
Specificity 99.65
Positive predictive values 51.0
Negative predictive values 93.01
Accuracy 98.74

TABLE 9.2 Mined GLCM Features for Regular and Dysplasia Cells.
GLCM features Standard cells Dysplasia cells
Energies 1.56×10 -5
1.623×10-5
Entropies 1.1078 1.106
Autocorrelations 1.616×10 4
1.596×104
Contrasts 1.13×104 1.306×104

9.6 CONCLUSION

• Cervix disease is foremost reason for feminine genital cancers and

also severe reason for feminine disease death around the world.
• Computer-aided automated detection technique for cervical cancer is
proposed in this chapter.
• The mined structures are skilled and categorized with support of
classifier.
• Planned system achieves 100% classification rate and further
processes, detects the edges of the diseased region.
• Routine of the planned system is tested using specific images.
160 Computational Imaging and Analytics in Biomedical Engineering

KEYWORDS

• pap smear
• human papilloma virus
• cervical cancer
• medical imaging
• image analysis
• morphological operations

REFERENCES

1. WHO Report. Cervical Cancer Screening in Developing Countries; Report of the World
Health Organization, 2002. http://whqlibdoc.who.int/publications/002/9241545720.pdf
2. National Cancer Institute. SEER Incidence and US Mortality Rates and Trends for the
Top 15 Cancer Sites by Race/Ethnicity; National Health Institute: Bethesda, 2005.
3. World Health Organization. Global Burden of Disease Report: Causes of Death in 2004;
Global Burden of Disease Report, World Health Organization: Geneva, 2004.
4. Roland, K. B.; Benard, V. B.; Greek, A.; Hawkins, N. A.; Manninen, D.; Saraiya, M.
Primary Care provider Practices and Beliefs Related to Cervical Cancer Screening with
the HPV Test in Federally Qualified Health Centers. Prev. Med. 2013, 57 (5), 419–425.
5. American Cancer Society (ACS). What is Cervical Cancer [Online] 2010. http://www.
cancer.org/cancer/cervicalcancer/detailedguide/cervical-cancer-what-is-cervical-cancer.
6. American Cancer Society (ACS). Key Statistics About Cervical Cancer
[Online] 2015. http://www.cancer.org/cancer/cervicalcancer/detailedguide/
cervical-cancer-key-statistics.
7. Song, D.; Kim, E.; Huang, X.; Patruno, J.; Muñoz-Avila, H.; Heflin, J.; Long, R. L.;
Antani, S. Multimodal Entity Coreference for Cervical Dysplasia Diagnosis. IEEE
Trans. Med. Imaging 2015, 34 (1), 229–235.
8. Gordon, S.; Zimmerman, G.; Greenspan, H. In Image Segmentation of Uterine Cervix
Images for Indexing in PACS, Proceedings of 17th IEEE Symposium on Computer-
Based Medical System, 2004; pp 298–303.
9. Ji, Q.; Engel, J.; Craine, E. Classifying Cervix Tissue Patterns with Texture Analysis.
Pattern Recognit. 2000, 33 (9), 1561–1574.
10. Park, S. Y.; Sargent, D.; Lieberman, R.; Gustafsson, U. Domain-Specific Image Analysis
for Cervical Neoplasia Detection Based on Conditional Random Fields. IEEE Trans.
Med. Imaging 2011, 30 (3), 867–878.
11. Horng, J. T.; Hu, K. C.; Wu, L. C.; Huang, H. D.; Lin, F. M.; Huang, S. L.; Lai, H. C.;
Chu, T. Y. Identifying the Combination of Genetic Factors that Determine Susceptibility
to Cervical Cancer. IEEE Trans. Inf. Technol. Biomed. 2004, 8 (1), 59–66.
12. Greenspan, H.; Gordon, S.; Zimmerman, G.; Lotenberg, S.; Jeronimo, J.; Antani, S.;
Long, R. Automatic Detection of Anatomical Landmarks in Uterine Cervix Images.
IEEE Trans. Med. Imaging 2009, 28 (3), 454–468.
Computer-Aided Automatic Detection and Diagnosis of Cervical Cancer 161

13. Alush, A.; Greenspan, H.; Goldberger, J. Automated and Interactive Lesion Detection
and Segmentation in Uterine Cervix Images. IEEE Trans. Med. Imaging 2010, 29 (2),
488–501.
14. Herrero, R.; Schiffman, M.; Bratti, C.; Hildesheim, A.; Balmaceda, I.; Sherman, M.
Design and Methods of a Population-Based Natural History Study of Cervical Neoplasia
in a Rural Province of Costa Rica: The Guanacaste Project. Rev. Panam. Salud. Publica.
1997, 1, 362–375.
15. Holland, J. H. In Adaptation in Natural and Artificial Systems, University of Michigan
Press: Ann Arbor, MI, 1975.
16. Kim, E.; Huang, X. A Data Driven Approach to Cervigram Image Analysis and
Classification; Color Medical Image Analysis, Series, In Lecture Notes in Computational
Vision and Biomechanics; Celebi, M. E., Schaefer, G., Eds.; Springer: Amsterdam, The
Netherlands, 2013; vol 6, pp 1–13.
17. Chang, S.; Mirabal, Y.; Atkinson, E.; Cox, D.; Malpica, A.; Follen, M.; Richards-
Kortum, R. Combined Reflectance and Fluorescence Spectroscopy for In Vivo Detection
of Cervical Pre-cancer. J. Lower Genital Tract Dis. 2005, 10 (2), 024031.
18. Davey, E.; Assuncao, J.; Irwig, L.; Macaskill, P.; Chan, S. F.; Richards, A.; Farnsworth,
A. Accuracy of Reading Liquid Based Cytology Slides Using the ThinPrep Image
Compared with Conventional Cytology: Prospective Study. Br. Med. J. 2007, 335
(7609), 31.
CHAPTER 10

A STUDY ON SENTIMENT ANALYSIS

M. MOORTHY1, YOHANNES BEKUMA BAKARE2, and
BALACHANDRA PATTANAIK2
1
Muthayammal Engineering College, Rasipuram, India
2
Department of Electrical and Computer Engineering, College of
Engineering and Technology, Wollega University, Ethiopia, Africa

ABSTRACT

The collection of people opinion with the use of social media in terms
of product selection is referred as sentiment analysis. Social media is the
method of sharing data with a large number of people. It can be addressed
as a medium of propagating information through an interface. In this paper,
fundamental concept and different algorithms used in sentiment analysis were
explained; also, it gives information about datasets for helping researchers
to do research in sentiment analysis. Based on the analysis of responses in
terms of views and feedbacks, three types of sentiments can be found—posi­
tive, negative, and neutral.

10.1 INTRODUCTION

The channels used as social media have stormy contact environments; it is

imperative to relay sensitive knowledge regarding person’s thinking on any
kind of device, concept, or policy through these social network channels.1
To both customers and suppliers, this data is valuable. In general, persons
view other people’s opinions about the product especially during any kind

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
164 Computational Imaging and Analytics in Biomedical Engineering

of purchasing. Based on the customer’s sentiment, the manufacturer can

learn about its product benefits and drawbacks. For researchers, it is a very
interesting area to examine and sum up the opinions conveyed in this broad
opinion text content in the development of data science instead of human
devices ready to analyze and solve the problems by finding the patterns in
every data set on their own.
A machine-learning algorithm uses a specific type of data to reply to
set of questions using the features extracted from the data. Besides, the
advancement of data science has helped the creation of models that analyzes
large volumes of complex data quickly and precisely. To obtain the highest
value from big data, businesses need to know precisely how to fit the correct
algorithm with a particular learning process or resources (Machine Learning
and its Applications Outsource to India).3
Social media is the method of sharing data with a huge and vast audi­
ence. It can be addressed as a medium of propagating information through
an interface. Social media in tandem with social networks helps individuals
cater their content to a wider society and reach out to more people for sharing
or promotion.4
Most utilize social destinations to express their feelings, convictions, or
suppositions about things, spots, or characters.5
Sentiment analysis, as a technique, can be used in extracting meaningful
information from any given set of documents.6
When it comes to human interactions and communication between
individuals, emotions are of utmost importance. Identifying the underlying
emotions from text is beneficial in deciding the axioms of human–computer
interaction that governs communication and many additional key factors.7–10
Sentiment analysis is the procedure of categorizing the views expressed
over a particular object. With the advent of varied technological tools, it
has become an important measure to be aware of the mass view in busi­
ness, products, or in matters of common like and dislike. Tracking down the
emotion behind the posts on social media can help relate the context in which
the user shall react and progress.
In this paper, we propose effective ways to perform sentiment analysis
or opinion-mining using various resources. The next part of the paper is
arranged as follows: Section 10.2 of the paper presents sentiment classifica­
tion; Section 10.3 briefs about the approaches for sentiment analysis; Section
10.4 talks about the machine learning-based sentiment analysis methodology;
Section 10.5 talks about the outline of machine learning techniques; Section
10.6 deals with the architecture of sentimental analysis for social media
A Study on Sentiment Analysis 165

analytics; Section 10.7 discusses about open source sentiment analysis tools;
and Section 10.8 presents the limitations of sentiment analysis. The paper is
concluded in Section 10.9.

10.2 SENTIMENT CLASSIFICATION

Sentiment analysis is an automated method of determining whether a usage-

produced text conveys a positive, negative, or common view of an object
(i.e., the item, the individual, the subject, the case, etc.). Sentiment classifi­
cation can be achieved at the four levels such as Document level, Sentence
level, and Aspect or Feature level2 as shown by the Figure 10.1.
A. Document level: In this level, the available record is classified into
a positive or negative class as a simple information category. In this
type of classification, the whole document can be represented as
“positive,” “negative,” or “neutral.11
B. Sentence level: In this type of classification, all the sentences can be
differentiated by subjective or objective, also each sentence can be
represented as “positive,” “negative,” or “neutral.” At this level, each
sentence is named “positive,” “negative,” or “fair-minded.11
C. Aspect or Feature level: The Aspect or Feature-level classification
involves the determination and finding the features from the avail­
able data. Also, in which the whole document or the sentences can be
classified by considering some parts of sentences or record which is
called as “perspective-level assessment grouping.”11

FIGURE 10.1 Sentiment classification.

166 Computational Imaging and Analytics in Biomedical Engineering

10.3 APPROACHES FOR SENTIMENT ANALYSIS

There are several techniques available for sentiment analysis:

A. Lexicon-based approach:
The optimistic and pessimistic words available in a dictionary used by
Lexicon are applied to assess the polarity of opinion. The count of optimistic
and pessimistic words is discussed in the text. If the text is more positive,
a positive score will be assigned to the text. The text is awarded a negative
score if it has high amount of negative or pessimistic words. The score will
be neutral if the text contains the same number of negative or pessimistic
words. A lexicon of opinion (positive and negative opinions) is developed to
finalize the word as positive or negative. There are numerous ways to build
and compile a dictionary.12
• Dictionary-based approach: A small number of words of opinion
with established guidelines are gathered manually.12 In corpora like
WordNet or thesaurus, synonyms, and opposite from these words are
then searched and appended to the group until a new term reaches
the process of assembling, but it is not fast. This method has the
inconvenience of depending on the dictionary scale, the intensity
of the sentiment classification. As the dictionary size increases, this
approach is wrong.
• Corpus-based approach: This approach mainly depends on thinking
patterns of enormous companies. The created words are context-
specific and it needs a large dataset labeled.
B. Machine learning-based approach:
Machine learning techniques in the classification of sentiment depends on
the use of well-known machine learning technology on text data as shown by
Figure 10.2. The classification of the sentiment based on machine learning
can be categorized primarily into supervised and unsupervised methods of
learning.13
• Supervised learning: Supervised methods of learning rely on labeled
training manuals. Supervised learning is an effective classification
method and has been used with very promising results for classifying
opinions. The regularly used supervised classification techniques
in sentiment analysis are Support Vector Machine (SVM), Naïve
Bayes (NB) Maximum Entropy (ME), and Artificial Neural Network
(NN), and Decision Tree (DT) classifiers. Some less commonly
A Study on Sentiment Analysis 167

used algorithms are Logistic Regression (LR), K-Nearest Neighbor

(KNN), Random Forest (RF), and Bayesian Network (BN).13
• Unsupervised learning: This technique does not use pre-listed data
to train the classifier, unlike supervised learning. The more common
instance of unsupervised machine learning algorithms are Kmeans
and Apriori Algorithms. Unsupervised machine learning may also be
divided into clusters and associations.14

FIGURE 10.2 Machine learning-based approach.

C. Hybrid-based approach:
The hybrid-based approach uses both ML and lexicon-based classification
approach. Few research techniques propose a mixture of lexicon-based and
automated learning techniques to enhance the classification of sentiment.
This hybrid approach is primarily advantageous as it can achieve the best of
both. The combination of Lexicon and Learning has demonstrated increased
accuracy.
168 Computational Imaging and Analytics in Biomedical Engineering

10.4 MACHINE LEARNING-BASED SENTIMENT ANALYSIS

METHODOLOGY

The polarity of an analysis data is calculated by various techniques. Machine

learning basic sentiment analysis technique is the most popular and efficient.
As discussed below, the polarity in analysis data and the most successful
algorithm are calculated.15 Data collection for any kind of text classification
task-specific in size as to the number of words, data sets can be used. Such
data sets were used after slight preprocessing for sentiment analysis such as
case folding, word deletion, etc.
Data Preprocessing: This pre-processing phase seeks to prepare text data
for further processing.
Feature Selection and Feature Vector Construction: A computer is not
able to process text data straight away, which is an inherent problem. Text
data must also be numerically interpreted. Terms are usually used as the
characteristics of the text. This gives the text representation a high dimen­
sion. Features need to be filtered to reduce dimensions and remove noise to
improve classification performance and processing efficiency.
Classification Algorithms for Sentiment Analysis: Several popular and
commonly used classification algorithms such as the Multinomial Naïve
Bayes Algorithm or the K-Nearest Neighbor Algorithm are commonly used
to identify sentiment polarity of users’ opinions based on given opinion data
Support Vector Machines Algorithm. Evaluation Metrics Measuring every
algorithm’s output using parameters such as confusion matrix, efficiency,
recall, and F-measurement.

10.5 OUTLINE OF MACHINE LEARNING TECHNIQUES

Mechanical learning is an Artificial Intelligence (AI) branch, which inves­

tigates machines for the establishment of new knowledge and skills and the
identification of existing knowledge. In the area of data mining, computer
vision, processing of natural languages, search engines, biometrics, medical
diagnostics, credit card fraud detection, a market analysis of stocks, DNA
sequence, speech and handwriting recognition, strategy games and robotics,
machine learner has been widely used.16 The popular machine learning
algorithms are:
1. Linear Regression: A linear regression is defined as the value of
the dependent or reliant variable is estimated using independent
A Study on Sentiment Analysis 169

variables for statistical techniques? A relationship consists of

mapping a dependent and independent variable on a line and that
line is known as regression line shown by Y = a*X + where Y is the
Dependent variable, X is the Independent Variable is the Intercept
and a is the slope.
2. Logistic Regression: This approach is used to define the discrete
dependent variable from the set of separate variables. Logistic
regression provides the coefficients to estimate a probability logistic
transformation.
3. Decision Tree: For classification and regression, the decision tree
may be used tree structure. In which the available dataset is divided
into set of subsets and the decision tree is used to develop a training
model to estimate the type of the final variable.
4. Support vector machine: A Binary Classifier (BC) is a Support
Vector Machine (SVM). On the n-dimensional point, row data is
drawn. In this, a hyper plane separating the data sets is drawn. This
enhanced separation maximizes the training data margin.
5. Naive-Bayes: This method is based on the theorem of Bayes used by
increasingly sophisticated classification methods. It is a classifica­
tion technique. It learns how an entity with certain characteristics
belonging to a certain category or class is possible.
6. KNN: This technique is used for classification and regression. This
is one of the simple machine learning algorithms. It saves the cases
and searches most k-neighbors it resembles for new data. It saves the
cases. With a testing dataset, KNN makes clear predictions.
7. K-means Clustering: It is an unsupervised algorithm for learning to
reach the cap. The initial partition is achieved by Euclidean distance
for grouping the datasets into clusters.
8. Random Forest: It is the category of the supervised algorithm. The
random forest algorithm involves the generation of several decision
trees. The regression as well classification process used these deci­
sion trees together. Based on the targets and features of the training
dataset this algorithm contain rules.
9. Dimensionality Reduction Algorithms: This implies that the
number of random variables is decreased by acquiring those key
variables. Function extraction and selection of features are methods
for reducing dimensionality. The main component analysis can be
done by Principal Component Analysis (PCA), which is a method of
removing primary variables from a wide range of variables.
170 Computational Imaging and Analytics in Biomedical Engineering

10. Gradient boosting and Ada Boost Algorithms: The algorithm of

gradient boosts is a classification and regression algorithm. AdaBoost
selects only those features that enhance model prediction. It operates
by selecting a base algorithm such as decision trees and refining it
iteratively by taking into account the wrong examples in the training
data set.

FIGURE 10.3 Flow chart of machine learning-based sentiment analysis technique.

10.6 ARCHITECTURE OF SENTIMENTAL ANALYSIS FOR SOCIAL

MEDIA ANALYTICS

Feeling assessment, opinion of customers, feedback, writing emotions,

and attitudes involves in the process of investigation of human feelings.
The review of the product as many is negative and positive is a decision-
making process. The key method of the rejection norm and the classification
of negative and positive feelings earned by the users or customers in the
A Study on Sentiment Analysis 171

social community.17 Data Pre-processing: The preprocessing technique is

more useful to identify and remove meaningless, noisy, and incompatible
data. Eliminating URLs: URLs will not help to inspect the emotion in the
non-formal text. Questions: The terms question will not help to reduce the
ambiguity of polarity such as when, when, who, how, etc. Removing Special
Characters: Special characteristics like, and, (), [] {}/' are separated to
eliminate inconsistencies by the function of polarity. Removal of Retweets:
The re-tweeting process doubles the tweet of another user and redistributes
it to peers. This also happens whenever a user tries to tweet another user.
Retweets are normally reduced.

10.7 OPEN SOURCE SENTIMENT ANALYSIS TOOLS

The following open source tools are all free and available for building and
maintaining your own sentiment analysis infrastructures and other NLP
systems. However, do keep in mind that in order to make use of the tools
below, you or someone on your team will need the necessary programming
and development skills to handle the coding and ML integration.
• NLTK: This includes lexical analysis, named entity recognition,
tokenization, PoS tagging, and sentiment analysis. It also offers some
great starter resources.
• Spark NLP: Considered by many as one of the most widely used
NLP libraries, Spark NLP is 100% open source, scalable, and
includes full support for Python, Scala, and Java. You’ll find a whole
host of NLP features, pre-trained models and pipelines in multiple
languages. There’s also an active Slack community for discussion and
troubleshooting.
• TextBlob: Built on the shoulders of NLTK, TextBlob is like an
extension that simplifies many of NLTK’s functions. It offers an easy
to understand interface for tasks including sentiment analysis, PoS
tagging, and noun phrase extraction. TextBlob is a recommended
natural language processing tool for beginners.
• Doccano: This open source text annotation tool has been designed
specifically for text annotation. It allows for the creation of labeled
data for sentiment analysis, named entity recognition, and text
summarization. This is a good option to look at for smaller datasets
and building initial proof of concept projects.
172 Computational Imaging and Analytics in Biomedical Engineering

10.8 LIMITATION OF SENTIMENT ANALYSIS

Customer’s sentiment about the product or brand may be influenced by many

factors. He might have a bad day and it may directly influence his remark
negatively. Also, sentiment can change over a period based on his mood.
So, it is advisable to go with large sample of data. It will be difficult for an
algorithm to understand the sarcasm and ironic language while interpret the
sentiment in isolation. So, there is a need to train the model rigorously.

10.9 CONCLUSIONS

Analyzing the amount of data that is being generated through internet

media and micro blogging by users of facilities available like Twitter,
Reddit,Facebook, etc., and understanding the behavior of people using
internet is useful; it can generate both informative and revenue for the busi­
ness. Using sentiment features rather than conventional text classification
gives high accuracy. This system is used to rank satisfactory classifiers for
sentiments and helps business organizations for making their future business
plans related to the product. Science and technology becomes fruitful only
when it is put to solve real problems. In this context, sentiment analysis can
serve as the solution to several business problems. It is the nexus of several
technologies and accumulated knowledge. It is safe to say that sentiment
analysis is a multi-disciplinary domain. To accurately infer sentiments and
emotions, we draw knowledge from computer science, psychology, linguis­
tics, machine learning, data mining, deep learning, and statistics.20 Sentiment
analysis is in an evolving phase and shows great promise. In terms of tech­
nological evolution, sentiment analysis has lost its passive state of rule-based
implementations and entered into an assertive state augmented by advanced
data mining and machine learning techniques.

KEYWORDS

• sentiment analysis
• approaches
• open source sentiment tools
• sentiment analysis classification
• social network channels
A Study on Sentiment Analysis 173

REFERENCES

1. Yi, S.; Liu, X. Machine Learning Based Customer Sentiment Analysis for Recommending
Shoppers, Shops Based on Customers’ Review. Complex Intell. Syst. 2020, 1 (1). DOI:
https://doi.org/10.1007/s40747-020-00155-2
2. Vohra, S.; Teraiya, J. A Comparative Study of Sentiment Analysis Techniques. Int. J. Inf.
Knowl. Res. Comput. Eng. 2013, 2 (2), 313–317.
3. Machine Learning & its Applications Outsource to India. [Online] May 18, 2020.
https://www.outsource2india.com/software/articles/machine-learning-applications­
how-it-works-whouses-it.asp.
4. Leskovec, J. In Social Media Analytics: Tracking, Modeling and Predicting the Flow
of Information Through Networks, Proceedings of 20th International Conference
Companion World Wide Web, 2011; pp 277–278.
5. Hasan, A.; Moin, S.; Karim, A.; Shamshirband, S. Machine Learning-Based Sentiment
Analysis for Twitter Accounts. Math. Computat. Applicat. 2016, 21 (1), ISSN:
2297–8747.
6. Padmaja, S.; et al. Opinion Mining and Sentiment Analysis – An Assessment of Peoples’
Belief: A Survey. Int. J. Ad Hoc Sens. Ubiq. Comput. 2013, 4 (1).
7. Sahu, T. P.; Ahuja, S. In Sentiment Analysis of Movie Reviews: A Study on
Feature Selection & Classification Algorithms, 2016 International Conference on
Microelectronics, Computing and Communications (MicroCom), 2016.
8. Akhtar, Md. S.; Kumar, A.; Ekbal, A.; Bhattacharyya, P. In A Hybrid Deep Learning
Architecture for Sentiment Analysis, International Conference on Computational
Linguistics: Technical Papers, 2016; pp 482–493.
9. Anil Kumar, K. M.; Rajasimha, N.; Reddy, M.; Rajanarayana, A.; Nadgir, K. In Analysis
of Users’ Sentiments from Kannada Web Documents, International Conference on
Communication Networks, 2015; vol 54, pp 247–256.
10. Mittal, N.; Aggarwal, B.; Chouhan, G.; Bania, N.; Pareek, P. In Sentiment Analysis of
Hindi Review Based on Negation and Discourse Relation, International Joint Conference
on Natural Language Processing, 2013; pp 45–50.
11. Wagh, R.; Punde, P. In Survey on Sentiment Analysis using Twitter Dataset, 2nd
International Conference on Electronics, Communication and Aerospace Technology
(ICECA 2018) IEEE Conference, 2018; ISBN: 978-1-5386-0965-1.
12. Medhat, W.; Hassan, A.; Korashy, H. Sentiment Analysis Algorithms and Applications:
A Survey. Ain Shams Eng. J. 2014, 5 (4), 1093–1113. DOI: https://doi.org/10.1016/j.
asej.2014.04.011
13. Aydogan, E.; Akcayol, M. A. In A Comprehensive Survey for Sentiment Analysis
Tasks Using Machine Learning Techniques, International Symposium on Innovations
in Intelligent Systems and Applications, 2016; vol 1 (1), pp 1–7. DOI: https://doi.
org/10.1109/INISTA.2016.7571856
14. Ahmad, M.; Aftab, S.; Muhammad, S. S.; Ahmad, S. Machine Learning Techniques for
Sentiment Analysis: A Review. Int. J Multidiscip. Sci. Eng. 2017, 8 (3), 27–35.
15. Yogi, T. N.; Paudel, N. Comparative Analysis of Machine Learning Based Classification
Algorithms for Sentiment Analysis. Int. J. Innov. Sci. Eng. Technol. 2020, 7 (6), 1–9.
16. Patel, A. Machine Learning Algorithm Overview. Medium [Online] May 18, 2020. https://
medium.com/ml-research-lab/machine-learning-algorithm-overview-5816a2e6303.
174 Computational Imaging and Analytics in Biomedical Engineering

17. Mahendran, N.; Mekala, T. A Survey: Sentiment Analysis Using Machine Learning
Techniques for Social Media Analytics. Int. J. Pure Appl. Math. 2018, 118 (8), 419–422.
18. Abdul-Mageed, M.; Diab, M. T.; Korayem, M. In Subjectivity and Sentiment Analysis
of Modern Standard Arabic, Proceedings of the 49th Annual Meeting of the Association
for Computational Linguistics: Human Language Technologies: Short papers, 2011; vol
2.
19. Nakov, P.; Ritter, A.; Rosenthal, S.; Sebastiani|, F.; Stoyanov, V. In SemEval-2016
Task 4: Sentiment Analysis in Twitter, Proceedings of SemEval2016; Association for
Computational Linguistics, 2016.
20. Xie, H.; Wong, T.; Wang, F. L.; et al. Editorial: Affective and Sentimental Computing.
Int. J. Mach. Learn. Cybern. 2019, 10, 2043–2044.
CHAPTER 11

APPLICATIONS OF MAGNETIC
RESONANCE IMAGING TECHNIQUES
AND ITS ADVANCEMENTS
V. RAMESH BABU,1 S. MARY CYNTHIA,2 K. SAVIMA,3 and
G. LAKSHMI VARA PRASAD4
1
Department of CSE, Sri Venkateswara College of Engineering,
Sriperumbudur, India
2
Department of ECE, Jeppiaar Institute of Technology, Chennai, India
Department of Computer Science, S.T.E.T. Women’s College,
3

Mannargudi, Tamil Nadu, India

4
Department of IT, QIS College of Engineering and Technology, Andhra
Pradesh, India

ABSTRACT

Magnetic resonance imaging (MRI) is widely used in biomedical research

and clinical applications. The emergence of new techniques increases the
applications of MR imaging remarkably. MRI is established with the use of
magnetization properties and RF signal for generating images of organs to
diagnose diseases. In this paper, the types of MRI and their several applica­
tions were explained and it gives the significance of different MRI images in
medical field. In general, the value of MRI is affected by some factors like
the availability of experts, number of scanners, and cost. The MRI scanning
process is comparatively more secure than other methods, like X-ray CT and
PET, in terms of ionization.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
176 Computational Imaging and Analytics in Biomedical Engineering

11.1 INTRODUCTION

Magnetic resonance imaging (MRI) is widely used in biomedical research

and clinical applications. The emergence of new techniques increases the
applications of MR imaging remarkably. MRI is established with the use of
magnetization properties. To capture MRI, a strong magnetic field is applied
to randomly oriented protons of human body available in water molecules
to make changes in the alignment. Then this magnetization is disordered
by applying external radio frequency wave. This absorbed RF energy is
then emitted with the help of several relaxation processes and the protons
realigned. Subsequently, the emitted signals are determined.
The frequency information present in the emitted signal for each loca­
tion is then converted into corresponding intensity levels. By changing
the cycle of RF pulses enforced and acquired different MRI images can
be constructed. The value of time required between two applied pulses is
called repetition time (TR) and the value of time between the distribution
of RF pulse and the collection of the echo signal is called time to echo
(TE).

11.2 TYPES OF MRI

11.2.1 T1-WEIGHTED IMAGES

T1 images indicate the time required for the protons to realign with the
direction of applied magnetic field which leads to short TE and repetition
time. Generally, fat takes short time for realignment compared with water
so fat appears bright while water emerges dark. The T1-weighted image
should contain short TR otherwise all protons appear with the same intensity
value. If the selected TR time is shorter than tissue’s recovery time then only
contrast images can be obtained.

11.2.2 T2-WEIGHTED IMAGES

The longer value of TE and TR times are used in the development of

T2-weighted images. The transverse relaxation time (T2) is defined as the
value of time required to give up phase adherence of protons.
Applications of Magnetic Resonance Imaging Techniques 177

11.2.2.1 APPLICATIONS
11.2.2.1.1 Identify Hematologic Marrow Diseases

Initially, from the T1- and T2-weighted MRI images, the bone marrow was
segmented and lot of features were extracted; among them the important
characteristics were selected based on principle component analysis (PCA)
and least absolute shrinkage and selection operator (LASSO).
Then, at last, classification models random forest (RF) and logistic regres­
sion (LR) were used to classify bone chondrosarcoma, metastatic diseases,
and osteoporosis.9

11.2.2.1.2 Prediction and Classification of Gliomas

Gliomas is a type of brain tumor. The conventional MRI images were used
to classify Gliomas in any one of the four grades given by World Health
Organization.10 This classification plays vital role in decision making and
planning of medical treatment.

11.2.2.1.3 Achilles Tendon Ruptures (ATRs) Treatment

T2-weighted images were used for the diagnosis and treatment of Achilles
tendon ruptures (ATR). The experimental results emphasized that this
method provides better positive correlation and higher specificity compared
with manual contour tracing (MCT).11

11.2.2.1.4 Diagnosis of Colorectal Cancer

The T2-weighted imaging method provides better results compared with

other methods to discriminate colorectal cancer. This scanned image can
give required information about medical treatment. The features of the
collected data, after extraction, were applied to deep learning algorithms for
classification process.12

11.2.2.1.5 Prediction of Out-of-Field Recurrence (OFR) of Cervical

Cancer

The selection of the method of medical treatment for cancer is based on the
stage of the disease. Also, the level of recurrence for the same variety of
178 Computational Imaging and Analytics in Biomedical Engineering

cancer may change from person to person. The conventional T1-, T2-weighted
images used to estimate Out-of-field recurrence (OFR) subsequences based
on this the treatment method is chosen from surgery, radiotherapy, and
chemotherapy methods.13

11.2.2.1.6 Early Detection of Lung Cancer

T2-weighted MR images along with convolutional neural network is used to

identify the lung nodule region. A false positive (FP) degradation method is
used to protect the true nodule. 14

11.2.2.1.7 Automated Stroke Lesion Segmentation

The fully automated quantitative segmentation of stroke lesion with the help
of T2-weighted MRI images as input is a very efficient method compared
with manual segmentation.15

11.2.3 FLUID ATTENUATED INVERSION RECOVERY (FLAIR)

The length a T2-weighted image depends on the value of TE and TR times

selected. So the CSF fluid is completely repressed and abnormalities appears
bright which yields easy determination abnormalities. In FLAIR images,
grey matter occurs brighter than white matter as like T2-weighted images
but cerebra spinal fluid (CSF) appears dark.

FIGURE 11.1 T1-, T2-weighted, and FLAIR magnetic resonance images.

Applications of Magnetic Resonance Imaging Techniques 179

11.2.4 SPIN ECHO MAGNETIC RESONANCE IMAGING

The SEM image captured with the help of spin-echo pulse sequence consists
of 90° excitation pulse and a 180° inversion or refocusing pulse. These pulses
are applied to the tissues present in the region of interest.

11.2.5 GRE (GRADIENT ECHO IMAGING)

The GRE image is obtained with the use of gradient-echo sequences which
are having flip-angle changing over a range of 10–80°. If the flip angle value
is large then it gives more T1-weighting to the image and if the value is small
then it provides T2-weighting to the image.

11.2.6 DIFFUSION TENSOR IMAGING (DTI)

Diffusion tensor imaging (DTI) is a type of MRI which is based on the flow
of water molecules present in the white matter of central nervous system
(CNS). Since DTI provides the information about the structural connectivity
of the brain white matter, its demand has increased over the last two decades.
Because of limited resolution and contrast, the conventional MRI techniques
were not able give information about axonal organization. But it is possible
using DTI because it primarily depends on the diffusion of water molecules;
its value is high in the axonal bundles compared to the normal direction so
the axonal direction can be easily determined.

11.2.6.1 MAPPING CEREBRAL CONNECTIVITY CHANGES AFTER

MTBI

Since TBI is a very diverse neurological form, WM structures are affected

differently by injury. The identification of WM degradation for mTBI
patients is possible by combining (1) automatic; atlas informed labeling
of WM streamline clusters, (2) streamline prototyping, and (3) Rieman­
nian matching of elastic curves to quantitate within-subject WM changes,
focusing on the arcuate fasciculus. The mTBI commonly results in traumatic
axonal injury (TAI) but the information about manifestations of disease is
not clearly available in T1- and T2-weighted MR images; it creates curiosity
180 Computational Imaging and Analytics in Biomedical Engineering

in mapping the issues of mTBI onto the white matter functionality of the
brain over time. It is possible with the use of DTI.1
The fractional anisotropy (FA) of water in the brain is quantified if its
value is less—it indicates that the occurrence of TAI. In this work, small
white matter bundles shorter than 4 cm length were neglected because
clustering algorithms used in DTI is not suitable for processing short
streamlines.

11.2.6.2 AUTISM SPECTRUM DISORDER DIAGNOSIS

Different types of MRI are used to diagnose the autism spectrum disorder
(ASD); whereas structural MRI (sMRI) can be used to study physiological
characteristics and brain functions can be studied by using functional MRI
(fMRI); also DTI is involved in the diagnosis of ASD disorder by studying
brain connectivity.
In general, ASD, called as autism, which contain variety of symptoms
like struggling in social interaction, interpersonal skills, and restricted and
repetitive behaviors.
In which water molecules direction is measured in minimum of six direc­
tions, by using this diffusion of water molecules in any other direction can
be determined. The mathematical representation of these directions can be
represented by diffusion tensor which is a 3 × 3 matrix and it is graphically
represented by an ellipsoid. A lot of characteristics may be taken out from
this diffusion tensor matrix, especially FA, axial and radial diffusivity, and
mean diffusivity (MD) which gives information about connectivity and
microstructure of white matter. Also the parameters calculated from these
important features such as trace, skewness, rotational invariance, and others
can be used to diagnosis ASD effectively.2 Image fusing will improve the
accuracy of diagnosis. This process contain three important steps. The first
step is preprocessing step which eliminates image artifacts result of improper
operation of imager and non-brain tissues. Feature extraction is the second
step for that any efficient atlas-based segmentation technique can be used
to calculate, extract, and select features. The final step is the classification
step the linear SVM classifier is used to classify ASD and TD (typically
developed) subjects. In this work, six different output features FA, mean
diffusivity (MD), axial diffusivity (AD), radial diffusivities in the directions
of two minor axis of diffusion ellipsoid, and skewness were used for the
determination of anisotropy.
Applications of Magnetic Resonance Imaging Techniques 181

11.2.6.3 DETECTION OF FOCUSED ULTRASOUND-INDUCED

BLOOD-BRAIN BARRIER OPENING

Also, DTI was used to detect blood–brain barrier (BBB) opening at the
without the use of MRI contrast agent. In this method, diffusion-weighted
images are captured over several directions minimum six directions in
conjunction with an image captured with the absence of weights in order to
populate the diffusion tensor which is a three-by-three, symmetric, positive
definite matrix.3 In this, the word diffusion means that movement of particles
of a body move with the same velocity along parallel paths propelled by
the thermal energy of particles. While movement, these molecules explore
the neighboring tissues at a small scale. The consequences of this displace­
ment corresponds to tissue structure is acquired by diffusion-weighted MRI
images. So it is used in many applications like diagnosis of stroke, edema
formation, subarachnoid hemorrhage, and multiple sclerosis.
FA mapping can identify structural changes in axons ensuing traumatic
brain injury.

11.2.6.4 CARDIOVASCULAR DISEASE DIAGNOSIS

To improve the study of in vivo cardiac DTI features a dense-encoder and

decoder trained in an unsupervised training method. Then the important
characteristics obtained from diffusion-weighted images were fused based
on maximum intensity principle.4 This method improved the quality of
image and calculated diffusion metrics.

11.2.7 FUNCTIONAL MAGNETIC RESONANCE IMAGING (FMRI)

The fMRI technique is used to determine the minute changes in blood flow
that happened during brain functions. It can predict anomaly brain functions
which cannot be determined with other imaging modalities.

11.2.7.1 DETERMINATION OF BRAIN FUNCTIONAL ACTIVITIES

In this work, resting state fMRI (rs-fMRI) is used to determine brain func­
tional activities based on blood oxygen level-dependent (BOLD) signals
then the ASD was diagnosed with the help of the combined framework of
182 Computational Imaging and Analytics in Biomedical Engineering

convolutional neural network with a prototype learning (CNNPL).5 The ASD

affected persons face problems in social interactions and communications.

11.2.7.2 DETECTION OF POST-TRAUMATIC STRESS DISORDER

(PTSD)

In this work, post-traumatic stress disorder (PTSD) is detected with the use
of rs-fMRI data and the highly affected brain region is detected with the help
of artificial neural network (ANN). The resting-state fMRI is very helpful to
provide functional relationship between the areas of the brain. The ANN is
used to provide dominance level of classification among the affected brain
regions left and right regions of the hippocampus, medial prefrontal cortex,
and amygdala. The experimental results show that the left hippocampus is
the highly influenced brain area in PTSD individuals.6

11.2.7.3 EARLY DETECTION OF ASD USING TASK-BASED FMRI

ASD is a neuro-developmental disorder which is mostly defined by impaired

social interaction and communication. The severity of autism can be reduced
with the help of early detection and initial treatment. This is possible for
12–14 months old persons and the use of structural MRI (sMRI), fMRI,
and DTI. In this work, a computer-aided grading framework in infants and
toddlers (between 12 and 40 months) dependent on the analysis of brain
activation in response to a speech experiment.7
Task-based fMRI is used to determine blood oxygen level-dependent
(BOLD) signals in all brain areas and their homogeneity in response to
experiments in different domains. The conclusion of this work is task-based
fMRI with machine learning technique is an effective tool for the early detec­
tion of autism disorder. In which task-based fMRI were captured during the
stimulation of three categories of audio record they were played as simple
forward speech, complex forward speech, and backward speech. This speech
experiment is repeated every 6 min and 20 s and separated by rest blocks
with the duration of 20s.
Task fMRI images were captured with the help of echo-planar imaging
techniques. fMRI session scan composed of 154 volumes. Each volume
is combines slices acquired with the alternating in the plus direction slice
acquisition pattern. Initially pre-processing is done to make the fMRI images
ready for the first level analysis using general linear model (GLM). The
Applications of Magnetic Resonance Imaging Techniques 183

classification process is based on the features determined from the statistical

maps created by GLM as well as group analysis. Each group is classified
against remaining groups in terms of the parameters accuracy, sensitivity,
and specificity. The best accuracy obtained for classifying between mild,
moderate and severe against the remaining two groups are 73%, 84%, and
83%, respectively, by random forest classifier compared with SVM and
multi-layer perceptron.

11.2.7.4 DETECTION OF EMOTIONAL CHANGES

In this work, emotional changes induced by social support can be detected

with the use of fMRI images. This experiment consists of three stages high
support stage, medium support stage, and low support stage.8

11.2.8 DIFFUSION KURTOSIS IMAGING (DKI)

In DTI images, diffusion are based on Gaussian distribution only. If diffusion

behavior is non-Gaussian, then DKI are used so that they are captured at
multiple b values with multiple gradient orientations. The generally used
performance metrics for DKI are mean kurtosis and axial and radial kurtosis.

11.2.9 HIGH ANGULAR RESOLUTION DIFFUSION IMAGING

(HARDI)

In white matter the diffusion is anisotropic and WM injury detection is

challenging one because of crossing and kissing fibers. HARDI data having
ability to resolve cross fibers and it is typically very useful in the case of
complex tract arrangement of brain. But the main disadvantage of DKI and
HARDI method is scanning time is large.

11.2.10 SUSCEPTIBILITY-WEIGHTED IMAGING (SWI)

Susceptibility-weighted imaging is a newly developed neuro imaging tech­

nique which uses magnitude and phase images. It is used to detect compo­
nents having paramagnetic, diamagnetic, and ferromagnetic properties like
blood products, iron, and calcium. These components couldn’t find using
184 Computational Imaging and Analytics in Biomedical Engineering

conventional MRI. Also, SWI have the ability to discriminate calcium from
hemorrhage. It is very essential for traumatic brain injury patients to classify
their severity.

FIGURE 11.2 SWI magnitude and phase image.

KEYWORDS

• T1 and T2-weighted MRI image

• FLAIR
• fMRI
• DTI
• DKI
• SWI

REFERENCES

1. Irimia, A.; Fan, D.; Chaudhari, N. N.; Ngo, V.; Zhang, F.; Joshi, S. H.; O'Donnell, L.
J. In Mapping Cerebral Connectivity Changes After Mild Traumatic Brain Injury in
Older Adults Using Diffusion Tensor Imaging and Riemannian Matching of Elastic
Curves, IEEE-17th International Symposium on Biomedical Imaging (ISBI), 2020; pp
1690–1693.
Applications of Magnetic Resonance Imaging Techniques 185

2. Elnakieb, Y. A.; Ali, Md. T.; Soliman, A.; Mahmoud, A. H.; Shalaby, A. M. Computer
Aided Autism Diagnosis Using Diffusion Tensor Imaging. IEEE 2020, 191298–191308.
3. Karakatsani, M. E.; Pouliopoulos, A. N.; Liu, M.; Jambawalikar, S. R.; Konofagou, E. E.
Contrast-Free Detection of Focused Ultrasound-Induced Blood-Brain Barrier Opening
Using Diffusion Tensor Imaging. IEEE Trans. Biomed. Eng. 2021, 68 (8), 2499–2508.
4. Deng, Z.; Wang, L.; Wu, Q.; Chen, Q.; Cao, Y.; Wang, L.; Cheng, X.; Zhang, J.; Zhu, Y.
Investigation of In Vivo Human Cardiac Diffusion Tensor Imaging Using Unsupervised
Dense Encoder-Fusion-Decoder Network. IEEE Access, 2020, 8, 220140–220151.
5. Liang, Y.; Liu, B.; Zhang, H. A Convolutional Neural Network Combined With Prototype
Learning Framework for Brain Functional Network Classification of Autism Spectrum
Disorder. IEEE 2020, 8, 2193–2202.
6. Shahzad, M. N.; Ali, H.; Saba, T.; Rehman, A.; Kolivand, H.; Bahaj, S. A. Identifying
Patients With PTSD Utilizing Resting-State fMRI Data and Neural Network Approach.
IEEE Access 2021, 9, 107941–107954.
7. Haweel, R.; Shalaby, A.; Mahmoud, A. H.; Ghazal, Md.; Seada, N.; Ghoniemy, S.;
Casanova, M. A Novel Grading System for Autism Severity Level Using Task-Based
Functional MRI: A Response to Speech Study. IEEE Access 2021, 9, 100570–100582.
8. Candemir, C.; Gonul, A. S.; Selver, A. M. Automatic Detection of Emotional Changes
Induced by Social Support Loss using fMRI. IEEE Trans. Affect. Comput. 2021, 1–12.
9. Hwang, E. J.; Kim, S.; Jung, J. Y. Bone Marrow Radiomics of T1-Weighted Lumber
Spinal MRI to Identify Diffuse Hematologic Marrow Diseases: Comparison With
Human Readings. IEEE Access 2020, 8, 133321–133329.
10. Ge, C.; Gu, I. Y. H.; Jakola, A. S.; Yang, J. Enlarged Training Dataset by Pairwise GANs
for Molecular-Based Brain Tumor Classification. IEEE Access 2020, 8, 22560–22570.
11. Regulsk, P. A.; Zielinski, J. Multi-Step Segmentation Algorithm for Quantitative
Magnetic Resonance Imaging T2 Mapping of Ruptured Achilles Tendons. IEEE 2020,
8, 199995–200004.
12. Yang, T.; Liang, N.; Li, J.; Yang, Y. Intelligent Imaging Technology in Diagnosis of
Colorectal Cancer Using Deep Learning. IEEE Access 2019, 7, 178839–178847.
13. Ikushima, H.; Haga, A.; Ando, K.; Kato, S.; Yuko, K.; Uno, T. Prediction of Out-of-
Field Recurrence After Chemo Radiotherapy for Cervical Cancer Using a Combination
Model of Clinical Parameters and Magnetic Resonance Imaging Radiomics: A Multi-
institutional Study of the Japanese Radiation Oncology Study Group. J. Radiat. Res.
2022, 63 (1), 98–106.
14. Li, Y.; Zhang, L.; Chen, H.; Yang, N. Lung Nodule Detection With Deep Learning in 3D
Thoracic MR Images. IEEE Access 2019, 7, 37822–37832.
15. Liu, Z.; Cao, C.; Ding, S. Towards Clinical Diagnosis: Automated Stroke Lesion
Segmentation on Multi-Spectral MR Image Using Convolutional Neural Network.
IEEE 2018, 6, 57006–57016.
CHAPTER 12

A HYBRID CLUSTERING APPROACH

FOR MEDICAL IMAGE SEGMENTATION
M. MALATHI1, K. SEKAR2, MAHENDRAKAN K3, and P. SINTHIA4
1
Department of Electronics and Communication Engineering,
Rajalakshmi Institute of Technology, Chennai, Tamil Nadu, India
Department of Electrical and Electronics Engineering, Hindusthan
2

College of Engineering and Technology, Coimbatore, Tamil Nadu, India

Department of Electronics and Communication Engineering,
3

Hindusthan Institute of Technology, Coimbatore, Tamil Nadu, India

Department of Biomedical Engineering, Saveetha Engineering College,
4

Chennai, Tamil Nadu, India

ABSTRACT

Different medical imaging techniques were used to various different kinds

of medical images. It provides the thorough interior composition of the
body organs. Radiologist detects the abnormality of the body parts from
this image. X-ray, CT, MRI, other tomographic modalities (SPECT, PET,
or ultrasound) are the various medical imaging modalities. After image
capturing segmentation is the next important process. The segmentation
process helps to determine the region of interest partially or automatically.
Manual segmentation refers to the partitioning and naming of an image by
hand by a human operator or physician. On a three-dimensional volumetric
image, segmentation is done slice by slice. Partitioning of medical images is
an easy or difficult process based on the presence of artifacts.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
188 Computational Imaging and Analytics in Biomedical Engineering

But manual segmentation is a time-consuming process. The experienced

physicians have to completely check the images, slice by slice in order to
remove the region of interest from the target image. The accuracy of the
manual segmentation is based on the physician’s experience. When the
segmentation is performed manually it is possible to perform a large number
of segmentation operations for images without any mistakes. Current trends
in the diagnosis of diseases employ CT and MRI imaging techniques. Radi­
ologists require computer-aided design to obtain useful clinical information
from the images. The computer-aided design of the imaging modalities also
helps in segmenting ample amount of data with the same accuracy. It means
that the result is not affected by fatigue, missing of any manual steps. This
automatic segmentation is performed by various segmentation algorithms.
But the type of algorithm used depends on the type of brain imaging tech­
niques, what kind of body part to be studied. Currently, a number of novel
techniques have been proposed for the hybrid clustering algorithm. Hybrid
clustering implements the combination of two different algorithms to perform
the segmentation process. As a result, hybrid clustering techniques evolve
the advantages of the two algorithms employed to perform the clustering
process. The predominant algorithms used in hybrid clustering techniques
are K-means, adaptive K-means, and spatial fuzzy c-means algorithm.
By combining the K means algorithm along with a spatial fuzzy c-means
algorithm, the hybrid clustering segmentation yields higher accuracy level
for tumor detection and also minimizes the processing time required for the
segmentation process.

12.1 USE OF SEGMENTATION IN MRI BRAIN IMAGE

To partition tissue and body parts, the segmentation technique is utilized.

From Ref. [1], based on applications, the segmentation is used to detect the
border in angiograms of coronary, surgical planning, tumor detection and
segmentation, brain growth study, mass detection in mammograms, image
registration, and heart segmentation in cardiac images. The region bound­
aries1 were signified established on the segmentation of each homogeneous
or a similar region. After image segmentation, it classifies the brain image2
into the tissue class like white matter, gray matter, and cerebro-spinal fluid.
Basically, segmentation is classified into two types
• Manual segmentation.
• Automatic segmentation.
A Hybrid Clustering Approach for Medical Image Segmentation 189

12.1.1 MANUAL SEGMENTATION

Manual segmentation is a tedious process where the human operator or

physician executes segmentation and labeling of an image by hands-on
method. The 3-D volumetric image obtained as slice-by-slice pixels is the
outcome of the segmentation process. The complexity of the technique
depends upon the nature of the artifacts present in the images. But the entire
task completion process for manual segmentation is time consuming. The
experienced physicians have to completely check the images, slice by slice
in order to remove the region of interest from the target image. The accuracy
of the manual segmentation is based on the physician’s experience. Hence,
the above difficulty in the manual segmentation has to be overcome by auto­
matic segmentation. The ground truth value for the quantitative analysis of
automatic segmentation is also obtained by manual segmentation.

12.1.2 NEED FOR AUTOMATIC SEGMENTATION

When the segmentation is performed manually it is possible to perform a

large number of segmentation operations for images without any mistakes.
Current trends in the diagnosis of diseases employ CT and MRI imaging
techniques. Radiologists require a computer-aided design to obtain useful
clinical information from the images. The computer-aided design of the
imaging modalities also helps in segmenting ample amount of data with the
same accuracy. It means that the result is not affected by fatigue, missing
of any manual steps. This automatic segmentation is performed by various
segmentation algorithms. But the type of algorithm used depends on the type
of brain imaging techniques, what kind of body part to be studied. During
imaging techniques, the brain image is affected by many factors like an arti­
fact, noise, etc. The authors in Ref. [7] state that the brain imaging is mostly
affected by the partial volume effect, but in the thorax the motion artifact
occurs. Certain problems are common to both CT, MRI medical images.3
These can be listed as follows:
• Partial volume effect.
• Motion artifacts and ring artifacts.
• The occurrence of noise caused by the sensors and the associated
electronic components.
There is no common algorithm for all the medical data, because every
imaging system has its personal specifications and boundaries. To speed
190 Computational Imaging and Analytics in Biomedical Engineering

up the automatic segmentation process very high-speed computers are now

available at moderate cost.

12.2 MRI IMAGING TECHNIQUE

An MRI technique uses a magnet, radio waves, and a computer to produce

the internal structure of body parts. The patient is located on the bed that
is connected to the magnet. The hydrogen atoms of the human body are
aligned with the help of the strong magnetic field produced by the magnet.
Further, this atom can be exposed to a beam of radio waves, and it rotates the
photons of a human body and produces a weak signal. It has to be detected
by the receiver of the MRI. Next, the signal from the receiver is sent to the
computer in order to produce the image. Finally, the MRI method relatively
produces complete organization of body parts.

FIGURE 12.1 Flow Diagram of Image Segmentation.

Neurosurgeons or doctors mostly prefer MRI imaging technique,4 because

of its high accuracy in disease detection for the entire body. Neurosurgeons
use the MRI technique to diagnose the abnormalities in the head like stroke,
A Hybrid Clustering Approach for Medical Image Segmentation 191

brain tumors, and inflammation of the spine. Neurosurgeon uses the MRI not
only for the study of brain anatomy, but for the reliability of the spinal cord
after trauma. MRI scanners generate 1500 images/second. With the help
of MRI imaging, it generates high-contrast images for studying soft tissue
anatomy. After doing many kinds of literature on brain tumor segmentation,
MRI images are mostly used by researchers.

12.2.1 CLASSIFICATION OF SEGMENTATION

This topic briefly gives the various segmentation methods. And it also
provides how the segmentation is performed on the MRI brain image. Andac
Hamamci,Nadir Kucuk, Kutlay Karaman, Kayihan Engin& Gozde Unal
2012 state that MRI brain image segmentation is a challenging job since
the captured image is affected by magnetic noise and other image artifacts.
Hence, many segmentation methods are implemented for processing MRI
images. But there is no one method which is appropriate for every image.5
Every method will be suitable for certain specific images. For instance,
the spatial information will be obtained from the texture features associ­
ated with an image. But the intensity-based approach basically depends on
the gray level histogram, which does not provide spatial information. But
the segmentation based on the theory of graph cuts is applied to any type
of images like gray or binary images. An unsupervised fuzzy clustering
finds many applications like sensing of remote areas, geology, biomedical,
molecular or biological imaging.

12.2.2 IMAGE SEGMENTATION STRATEGIES

Generally, the entire segmentation algorithm is based on the two fundamental

properties of intensity level values.
• Discontinuity.
• Similarity.
Discontinuity-based segmentation method partition of an image depends
on the sudden changes in intensity. Similarity-based segmentation partition
of an image is based on some predefined condition, that is, the similarity
between the two regions. Figure 12.2 shows the classification of image
processing techniques.
192 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 12.2 Various image segmentation techniques.

12.3 CLUSTERING METHODS

The following topics give the detailed discussion on clustering and its types.
Shijuan He et al. (2001) told clustering is one of the simple unsupervised
learning algorithms. It is defined as a grouping of pixels with similar intensi­
ties without using any training images. This classification operation on pixels
is performed without the knowledge of prior information, The clustering
algorithm trains by its own, using the available data.

12.3.1 CLASSIFICATION OF CLUSTERING

Basically, the clustering is classified into two types

• Hard computing clustering.
• Soft computing clustering.

FIGURE 12.3 Clustering algorithms classification.

A Hybrid Clustering Approach for Medical Image Segmentation 193

12.3.2 HARD COMPUTING CLUSTERING

Clustering is a technique which partitions the input image into different clus­
ters by repeatedly calculating the centroid, and the pixel is forced to move to
the nearest cluster center. This is called hard clustering,9 because it pushes
each pixel into particular cluster center through the continuous iteration. The
authors in Ref. [9] state that there are three common types of hard clustering.
• K-means clustering.
• Fuzzy c-means clustering.
• Expectation and maximization (EM algorithm).

12.3.3 SOFT COMPUTING CLUSTERING

Conventional clustering creates a pattern by grouping the pixel belongs to

one and only cluster. In hard clustering, the clusters are disconnected. But
in soft computing, the pattern is related to every clustering by arranging the
membership function, namely every cluster is a fuzzy set of all patterns.
Fuzzy clustering is one of the best examples for the soft computing clustering.

12.3.3.1 K-MEANS CLUSTERING

Macqueen proposed the algorithm in the year 1997. This comes under the
category of unsupervised algorithms. The algorithm is initiated by assigning
random values of the number of cluster K. Next, centroid is computed from
the cluster center. Each pixel value is estimated in contrast with the centroid.
Then the pixel is located to the particular cluster having the shortest path
among all. The same process7 is repeated by reestimating the centroid for
the next pixel. This process is repeated till convergence of the center. The
algorithm steps are explained as follows:
Step 1: Choose random values for the C cluster center.
Step 2: Euclidean distance has been evaluated among each pixel to cluster
center.
Step 3: Every pixel is assigned to the specific cluster, which has shortest
distance.
Step 4: The chief objective of the algorithm is to reduce the squared error
Xi –Vi is the Euclidean distance between Xi, Vi
194 Computational Imaging and Analytics in Biomedical Engineering

C is the number of clusters

Ci is the number of data points in the ith cluster. Next, calculate the cluster
center by using the following formula:
Ci
1
C= C ∑X
j =1
i (12.1)
i

The advantages are

• An easy implementation is possible.
• The algorithm is easy to understand.
The disadvantages are
• The choice for choosing a K-value is essential to accomplish careful
segmentation.
• It is more sensitive to disturbances and artifacts.
Clustering finds many concerns in a variety of areas like image processing,
data mining, Image retrieval, pattern recognition, image segmentation, etc.

12.3.3.2 FUZZY C-MEANS ALGORITHM

It is one of the best-unsupervised algorithms used for medical image

segmentation. It is developed by DQMM12 and modified by Bezdek. Fuzzy
clustering finds the application in pattern recognition. This is an iterative
type clustering method. It is one of a more suitable method for segmentation
when the K-value is predefined. A Fuzzy algorithm is capable of processing
the overlapped dataset. The technique is able to give good segmentation
results for noise-free images. The algorithms are described as follows:
Step 1: After assigning the K-value, assign the membership value of each
data point based on the cluster center and the data point. The main motive of
the algorithm is to minimize the following function:
2
∑ ∑
N C
=J =i 1 =j 1
xj − cj (12.2)

where N is the number of data points

C refers to the required number of clusters.
dij defines the value of membership for ith data point xi in the cluster
J = x j − c j represents the measure of the closeness of the datapoint xi to the
center vector cj of the cluster j
A Hybrid Clustering Approach for Medical Image Segmentation 195

The above formula estimates the distance between the data point and
cluster center.
Step 2: Next, the data points near to the particular cluster center have the
largest membership value of that specific center. The membership value is
calculated by using the following formula:
Let xj be a data point, and let its degree of membership to a particular
cluster j be calculated as follows:
1
δij = 2
x −C m−1
∑ k =1 xi − C i
C
(12.3)
i k

where m is the fuzziness coefficient and cj is calculated as follows:

∑ (δ x )
N m
ij j
Cj = i =1
(12.4)
∑ δ
N m
i =1 ij

δijm is degree of membership value.

The membership value lies between 0 and 1. It means that ∑δij = 1
Step 3: The fuzziness coefficient values lie between 1 < m < ∞. The
values define the amount of cluster that can overlap with each other. Number
of iterations of the segmentation process depends on the accuracy of degrees
of membership values. This accuracy of degrees of membership is measured
using the amount of membership value varied from one iteration to the next.
∈= ∆ iN ∆ cj δijk +1 − δijk (12.5)

δijk +1 , δijk are membership values for K, K+1 iterations, respectively.

Step 4: The algorithm ends when the cluster center is stabilized.
The disadvantages of FCM are
• All the membership values for a data point in the entire cluster are
one, but the outlier points have the value more than one. It is difficult
to process this kind of process.
• The algorithm is more suitable for noise-free images.

12.3.3.3 SPATIAL FUZZY C-MEANS ALGORITHM

The conventional algorithm uses the single membership value to characterize

the desired pattern. It is not sufficient to perform the segmentation exactly.
For the detection of brain tumors in MRI images, using pixel intensity as
196 Computational Imaging and Analytics in Biomedical Engineering

the sole parameter is not sufficient to classify the brain tissue. When any
dissimilar structure appears, the conventional FCM14 is not sufficient for
segmentation. This can be avoided by adding the spatial information of
neighboring pixels which is considered to define the probability function of
each pixel. This spatial information helps to find new membership values
for each pixel. It leads to reducing the problem due to noise and intensity
inhomogeneity and increases the accuracy of the result.
Consider that X= {x1, x2, x3….. xn} is the set of data points
C=c1,c2,c3,…..cn} is the set of centers.
The following two equations are used to calculate membership and the
cluster center is updated for each iteration.
µ 1
ij =
(12.6)
2
−1
c  dij m
∑ 
k =1  d
 ik



Cj = ∑
( ij ) i
n  µ m x 

 µij m  (12.7)
i =1
 
dij—refers the distance between the ith data and the jth cluster
C—represents the number of clusters
m—fuzziness index
mij—membership of the ith data to the jth cluster data
n—number of data points
Cj—represents the jth cluster center

12.4 HYBRID SEGMENTATION

Determination of the correct choice of the segmentation algorithm for the

given applications is a challenging task in medical image segmentation
(Zhang Xiang et al. 2002). A novel method is introduced which combines
any two segmentation methods to acquire the accurate lesion segments from
the MRI brain image. The cognitive process of uniting any two segmentation
algorithms is named as hybrid segmentation. The main purpose of combining
the various algorithms is to take away the disadvantages of two different
methods and to better the quality of segmentation. The EM segmentation17 is
combined with the active contour models, binary mathematical morphology
to implement the segmentation of adult 2D images. The cerebral blood is
segmented by combining model-based region growing with morphological
segmentation. The authors in Ref. [9] combine fuzzy c-means with K-means
A Hybrid Clustering Approach for Medical Image Segmentation 197

which is called a KIFCM technique to perform accurate tumor detection

from an MRI image.
The advantage of hybrid segmentation is
• The combination of two algorithms improves overall segmentation
accuracy.
The disadvantages are listed as follows:
• It is difficult to implement the new model for segmentation.
• It reduces the computation time, but a large number of parameters
should be tuned to perform segmentation for certain applications.
• Care should be taken while designing the modeling to provide the best
quality of segmentation.
In this context, the tissue characteristics and all physical properties are
closest to the in vivo properties. This kind of phantom images is produced
by the MRI scanner. It provides the image which is more genuine than the
images produced by software simulators. But the method is not flexible, like
software simulators; it is more costly and needs more labor.

12.4.1 IMPLEMENTATION OF CLUSTERING TECHNIQUES

The following topic describes the brief discussion18 about results for various
segmentation algorithms. The algorithms are like K-means clustering, adap­
tive k-means clustering, spatial fuzzy, c-means clustering used for segmenta­
tion of brain image. The performance analysis between the various stages of
results of K-means, adaptive k-means algorithms is compared in terms of
accuracy, time, PSNR, and area. For the segmentation process the sample
brain images are acquired from hospital. The next topic discusses the various
results of the K means algorithm.

12.4.2 ABOUT MATLAB

The brain tumor segmentation algorithm is implemented using Mat Lab

software. This is one of the commonly used computational tools in various
fields like science and engineering, which includes the fields of physics,
chemistry, math, and all engineering streams. It is used in a range of applica­
tions including
198 Computational Imaging and Analytics in Biomedical Engineering

• Signal processing and communications.

• Image and video processing.
• Image segmentation.
• Control systems.
• Computational finance.
• Computational biology.

12.5 CONCLUSION

Three different types of algorithm are implemented in this research work.

The hybrid segmentation which is proposed in this work is entirely based
on providing the best results and the accurate calculations for estimating the
area and time when compared to the other two algorithms. Because of the
integration of the two algorithms, that is, K-means segmentation and spatial
fuzzy C-means algorithm, the distinct features of the two methods are also
combined.
It can be concluded that the compounding of two algorithms yields
better results. Based on this motivation, the research work combines the
two different algorithms, namely SFCM and K-means algorithm and it is
called KISFCM. K-means algorithms are used to detect the tumor as faster
than FCM, but it gives a good result for only smaller values of K. The next
algorithm FCM is utilized to discover the tumor cells that are not adjacent
by K-means. FCM is also not considering the spatial characteristics of
brain images because of the fact that this kind of characteristics is very
essential to sort out the complex structures. But it is a time-consuming
segmentation process. Hence, the research work integrates the K-means
with SFCM titled as KISFCM to acquire faithful area detection of various
algorithms which are represented in terms of area and time.

KEYWORDS

• segmentation
• clustering
• computer tomography (CT)
• magnetic resonance imaging (MRI)
• K-means
• fuzzy c-means
A Hybrid Clustering Approach for Medical Image Segmentation 199

REFERENCES

1. Banerjee, A.; Maji, P. Rough Sets and Stomped Normal Distribution for Simultaneous
Segmentation and Bias Field Correction in Brain MR Images. IEEE Trans. Image
Process. 2015, 24 (12), 5764–5776.
2. Gooya, A.; Biros, G.; Davatzikos, C. Deformable Registration of Glioma Images Using
EM Algorithm and Diffusion Reaction Modeling. IEEE Trans. Med. Imaging 2011, 30
(2), 375–389.
3. Ism, A.; Direkoglu, C.; Sah, M. In Review of MRI Based Brain Tumor Image Segmentation
Using Deep Learning Methods, Proceedings of 12th International Conference on
Application of Fuzzy Systems and Soft Computing; Vienna, Austria, Aug 29–30, 2016.
4. Roniotis, A.; Manikis, G. C.; Sakkalis, V.; Zervakis, M. E.; Karatzanis, I.; Marias,
K. High Grade Glioma Diffusive Modeling Using Statistical Tissue Information and
Diffusion Tensors Extracted from Atlases. IEEE Trans. Inf. Technol. Biomed. 2012, 16
(2), 255–263
5. Asanambigai, V.; Sasikala, J. Adaptive Chemical Reaction Based Spatial Fuzzy
Clustering for Level Set Segmentation of Medical Images. Ain Shams Eng. J. 2016, 9
(3), 459–467.
6. Islam, A.; Syed, M. S.; Khan, M. I. Multifractal Texture Estimation for Detection and
Segmentation of Brain Tumors. IEEE Trans. Biomed. Eng. 2013, 60 (11), 3204–3215.
7. Arizmendi, C.; Daniel, A. S.; Alfredo, V.; Enrique, R. Automated Classification of Brain
Tumours from Short Echo Time In Vivo MRS Data Using Gaussian Decomposition and
Bayesian Neural Networks. Expert Syst. Appl. 2014, 41 (11), 5296–5307.
8. Chen, L.; Weng, Z.; Yoong, L.; Garland, M. An Efficient Algorithm for Automatic
Phase Correction of NMR Spectra Based on Entropy Minimization. J. Magn. Reson.
2002, 158 (1), 164–168.
9. Eman, A. M.; Mohammed, E.; Rashid, A. L. Brain Tumor Segmentation Based on a
Hybrid Clustering Technique. Egypt. Inform. J. 2015, 16 (1), 71–81.
10. Xing, F.; Xie, Y.; Yang, L. Automatic Learning-Based Framework for Robust Nucleus
Segmentation. IEEE Trans. Med. Imaging 2016, 35 (2), 550–566.
11. Hai, S.; Xing, F.; Yang, L. Robust Cell Detection of Histopathological Brain Tumor
Images Using Sparse Reconstruction and Adaptive Dictionary Selection. IEEE Trans.
Med. Imaging 2016, 35 (6), 1575–1586.
12. Kalbkhani, H.; Mahrokh, G. S.; Behrooz, Z. V. Robust Algorithm for Brain Magnetic
Resonance Image (MRI) Classification Based on GARCH Variances Series. Biomed.
Signal Process. Control 2013, 8 (6), 909–919.
13. Yao, J.; Chen, J.; Chow, C. Breast Tumor Analysis in Dynamic Contrast Enhanced MRI
Using Texture Features and Wavelet Transform. IEEE J. Select. Top. Signal Process.
2009, 3 (1), 94–100
14. Jainy, S.; Kumarb, V.; Gupta, I.; Khandelwalc, N.; Kamal, C. A Package SFERCB
Segmentation, Feature Extraction, Reduction and Classification Analysis by Both SVM
and ANN for Brain Tumors. Appl. Soft Comput. 2016, 47, 151–167.
15. Jothi, G.; Inbarani, H. H. Hybrid Tolerance Rough Set Firefly Based Supervised
Featureselection for MRI Brain Tumor Image Classification. Appl. Soft Comput. 2016,
46, 639–651.
200 Computational Imaging and Analytics in Biomedical Engineering

16. Sallemi, L.; Njeh, I.; Lehericy, S. Towards a Computer Aided Prognosis for Brain
Glioblastomas Tumor Growth Estimation. IEEE Trans. Nanobiosci. 2015, 14 (7),
727–733.
17. Valarmathy, G.; Sekar, K.; Balaji, V. An Automated Framework to Segment and Classify
Gliomas Using Efficient Segmentation and Classification. Int. J. Adv. Sci. Technol.
2020, 29 (10S), 7539–754.
18. Valarmathy, G.; Sekar, K.; Balaji, V. An Automated Framework to Segment and Classify
Gliomas Using Efficient Shuffled Complex Evolution Convolutional Neural Network.
J. Med. Imag. Health Inf. 2021, 11, 2765–2770.
19. Malathi, M.; Sujatha, K.; Sinthia, P. Brain Tumour Segmentation Using Clustering And
EM Segmentation. Int. J. Appl. Eng. Res. 2015, 10 (11), 29105–29119.
20. Malathi, M.; Sujatha, K.; Sinthia, P. Detection and Classification of Brain Tumour using
Back Propagation Algorithm. Int. J. Control Theory Appl. 2016, 9 (24), 299–306.
CHAPTER 13

APPROACHES FOR ANALYZING

DENTAL IMAGES WITH MEDICAL
IMAGE PROCESSING WITH ITS
STATISTICS
G. BALANAGIREDDY1, P. SUBRAMANIAN2,
J. MARTIN SAHAYARAJ3, and S. MUTHUKUMAR4
1
Department of Electronics and Communication Engineering,
Rajiv Gandhi University of Knowledge Technologies – Ongole Campus,
Ongole, Andhra Pradesh, India
Department of Computer Science Engineering, Mohamed Sathak
2

A. J. College of Engineering, Chennai, India

3
Department of Electronics and Communication Engineering,
Sri Indu College of Engineering and Technology, Hyderabad,
Telangana, India
4
Department of Computer Science Engineering, B.S. Abdur Rahman
Crescent Institute of Science and Technology, Chennai, Tamil Nadu,
India

ABSTRACT

In dental morphological images, there exhibits a lot of inconsistency due to

noise, teeth visualized per image. So, providing insightful experience would
nurture the needs for stomatological purposes. This chapter deals with such

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
202 Computational Imaging and Analytics in Biomedical Engineering

aspects considering radiographical images. Measuring the distance between

the mandible inferior border and the superior border of the “alveolar” for
the image is done manually. Thresholding via various methods such as
“Haung,” “Otsu” and their corresponding values are provided for analysis.
Both thresholding algorithm used in the subset of the image dataset shows
unbiased results of mandible region.

13.1 INTRODUCTION

The X-ray image acquired for dental purposes can be classified as intraoral
and extraoral. The former states images are acquired within the mouth and
the latter states that images are acquired outside the mouth. The majority of
the survey has been carried out using “intraoral segmentation” via threshold-
based segmentation. The extraoral dataset had been discussed in Ref. [1],
with a review of varying segmentation algorithms. The oral cavity has been
analyzed with malignant and benign lesions using deep learning techniques
specifically for dentigerous cysts.2

FIGURE 13.1 Two-dimensional classification of X-ray images.

Based on their geometrical projection, two-dimensional radiographic

images are shown in Figure 13.1.4 “Cephalometric” analysis using Artificial
Intelligence has been done in Ref. [5], for localization of landmarks for
classification and segmentation purposes. The impact of mandibular region
segmentation in the image and the dose given for radiological perspective
has been analyzed with scanners in phantoms.12. In Ref. [11], segmenting the
“mandibular bone” from the teeth is difficult with grayscale images. Hence,
preprocessing with a “Gaussian filter” and postprocessing algorithms have
been carried out with active contours. Images were taken from the imaging
repository as in Ref. [13], and the ImageJ tool has been used for processing.
Approaches for Analyzing Dental Images 203

Otsu level of thresholding divides the object into segments of two groups
based on variance levels.18 Haung level of thresholding incorporates object
attributes for segmenting an image.19 Both thresholding algorithms used in
the image dataset show unbiased results of the mandible region in sample 1.
Section 13.2 deals with a literature survey of dental imaging, and
segmentation. Section 13.3 deals with the steps involved in the proposed
system development. Section 13.4 deals with statistical and medical imaging
algorithms. Section 13.5 concludes the overall work with future scope.

13.2 LITERATURE SURVEY

Interpretation of X-rays for three-dimensional image constructions has

evolved in various means like subtraction radiography, computer-aided
learning for the acquired image.10 Early, discussion of multislice computed
tomography (CT) states that it provides better accuracy in terms of volume
in vivo and in vitro studies.6 In terms of detection accuracy, the images of
bitewing and periapical were better than the panoramic images considering
the posterior part of the teeth.3 The discussions in Ref. [7] state the need for
morphological operation differences between dentistry annotated data and
automated segmented data. The states finding states the source of errors, and
similarity indices obtained via Jaccard’s coefficient. Local ternary pattern
filter had been discussed in Ref. [8]; for image segmentation, it calculates
three values specified for the threshold pixel. The first “1” value signifies the
neighbor pixel has a value greater than the threshold constant plus the center
of the pixel. The second “−1” value specifies the neighbor pixel has a value
lesser than the center of the pixel minus the threshold constant. The third
value “0” denotes that lies in a range greater than the threshold constant plus
the center of the pixel to the threshold constant minus the center of the pixel.8
Thus a connected component analysis and easy as well as automatic segmen­
tation had been provided. Deep learning model had been used in Ref. [9], for
instance, tooth segmentation and its region of interest. Bounding boxes had
been used for analysis as a comparative index between the missing region
and its ground truth. In Ref. [13], the image registration technique followed
by extracting the mandible contour from the existing mandible atlas has
been done. An inferior border has been extracted with a canny edge detector
where vertical variance obtained using a Gaussian filter is 2 times lesser
than horizontal variance. A semiautomatic model for incorporating manual
segmentation has been associated with the body axis plane.14 The acquired
204 Computational Imaging and Analytics in Biomedical Engineering

images and their age had been more than 20 years in Ref. [13] to avoid
deciduous teeth. The demographic detail of age less than 20 years has been
considered in Ref. [14]. Asymmetric mandibles exist wherein it is difficult
to analyze the three-dimensional shape via conventional techniques. Further,
the study states the influence of landmarks and angular measurements in
accessing the morphology.15
An extensive review of dental X-rays is given in Ref. [16] as it states
the evolution of image processing, deep learning, and machine learning
practices. Variation of intensity in an irregular manner in some cases might
degrade the image quality. In Ref. [17], the discussed works state strange­
ness in acquired medical images has to be visualized via image processing
for extracting its features. A canny edge detector has been used in DICOM
images for pinpointing the exactness of boundaries without loss of its features.
Examining patients based on verbal exchange has been enhanced using deep
learning techniques.21 The support vector technique state it provides more
accuracy for diagnostic and prognostic features.

13.3 METHODOLOGY

13.3.1 PROCESSING OF MANDIBLE REGION IN SAMPLE 1

Panoramic Images were obtained from the imaging archive13 with a segmented
mandible region. Length measurement across the mandible region has been
performed by the mandible inferior border and the superior border of the
“alveolar” for the image.
Then, Otsu thresholding has been done using the ImageJ plugin. It is a
single intensity value that separates the pixels into its classes. It signifies
there is minimal interclass variance that exhibits in an image.
Subsequently, Haung thresholding is done based on the function of
Shannon entropy.

13.3.2 PROCESSING OF MANDIBLE REGION IN SAMPLE 2

Panoramic Images were obtained from the imaging archive13 with a

segmented mandible region.
Edge detection is done to process the edges. It is followed by similar
multi-thresholding via the Haung model and its maximum threshold values.
Approaches for Analyzing Dental Images 205

13.4 RESULTS AND DISCUSSION

A complete image followed by two dentists segmented with the mandible

region segmentation has been taken for analysis with ImageJ software. The
ImageJ software has been used in Ref. [20] since it gives accuracy in results
used for volumetric analysis in oral bone.

FIGURE 13.2 Original grayscale image-1 taken without grayscale segmentation.

Figure 13.2 shows full image with the mandible region included. Figure
13.3 is a segmented image available in the dataset by dentist 1. Figure 13.4
is a segmented image available in the dataset by dentist 2.

FIGURE 13.3 Segmented mandible region of interest by dentist 1.

206 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 13.4 Segmented mandible region of interest by dentist 2.

FIGURE 13.5 Segmented mandible length measured manually indicated via yellow lines.

In Figure 13.5, a length of 582.5 has been measured between mandible

inferior border and the superior border of “alveolar” for image obtained from
dentist 1.
Approaches for Analyzing Dental Images 207

FIGURE 13.6 Segmented mandible length measured manually indicated via yellow lines.

In Figure 13.6, a length of 579 has been measured between the mandible
inferior border and the superior border of the “alveolar” for the image
obtained from dentist 2.

FIGURE 13.7 The threshold value of the image 2.

208 Computational Imaging and Analytics in Biomedical Engineering

In Figure 13.7, a threshold value of 20 is chosen automatically with the

Huang model. The minimum and maximum value used in the Haung model
is shown in Figure 13.8.

FIGURE 13.8 The threshold values used in Haung.

FIGURE 13.9 The threshold value of the image 2 via Otsu thresholding.
Approaches for Analyzing Dental Images 209

In Figure 13.9, a threshold value of 21 is chosen automatically with Otsu

thresholding. The minimum and maximum value used in is shown in Figure
13.10.

FIGURE 13.10 The threshold values used in “Otsu”.

FIGURE 13.11 Original grayscale image 2 taken without grayscale segmentation.

210 Computational Imaging and Analytics in Biomedical Engineering

Figure 13.11 shows whole image with the mandible region included.
Figure 13.12 shows a segmented image available in the dataset by dentist 1
for an original image shown in Fig. 13.11. Figure 13.13 shows segmented
image available in the dataset by dentist 2 for an original image shown in
Fig. 13.12.

FIGURE 13.12 Segmented mandible region of interest by dentist 1.

FIGURE 13.13 Segmented mandible region of interest by dentist 2.

Approaches for Analyzing Dental Images 211

FIGURE 13.14 Edge detection technique for mandible region.

Figure 13.14 shows the edge detection region of interest for dentist 1
image shown in Figure 13.12. Figure 13.15 used the Haung threshold
maximum value lies at 6 for interpretation.

FIGURE 13.15 Haung threshold method for edge detection image for mandible region.
212 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 13.16 Edge detection technique for dentist image 2.

FIGURE 13.17 Haung thresholding for dentist image 2.

Figure 13.17 shows the Haung threshold for dentist image 2 where the
maximum value lies at 6.

13.5 CONCLUSIONS

Two samples of oral panoramic X-ray images were taken for analysis
followed by processing them via segmented mandible regions from experts.
Approaches for Analyzing Dental Images 213

In the first case, the image and its measurement between the inferior border
and the superior border of the mandible region were performed. The work
was further analyzed by thresholding the region via Haung and Otsu-based
analysis. In sample 2, the image obtained was processed via edge detection
and analyzed with subsequent Haung-based thresholding. The future work
will incorporate the impact of age and related dosage associated with volume
metrics analysis in the mandible region.

KEYWORDS

• thresholding
• edge detection
• intraoral
• extraoral
• segmentation algorithms

REFERENCES

1. Silva, G.; Oliveira, L.; Pithon, M. Automatic Segmenting Teeth in X-ray Images:
Trends, a Novel Data Set, Benchmarking and Future Perspectives. Expert Syst. App.
2018, 107, 15–31.
2. Yang, H.; Jo, E.; Kim, H. J.; Cha, I. H.; Jung, Y. S.; Nam, W.; Kim, D. Deep Learning
for Automated Detection of Cyst and Tumors of the Jaw in Panoramic Radiographs. J.
Clin. Med. 2020, 9 (6), 1839.
3. Akarslan, Z. Z.; Akdevelioglu, M.; Gungor, K.; Erten, H. A Comparison of the
Diagnostic Accuracy of Bitewing, Periapical, Unfiltered and Filtered Digital Panoramic
Images for Approximal Caries Detection in Posterior Teeth. Dentomaxillofacial Radiol.
2008, 37 (8), 458–463.
4. Akarslan, Z. Z.; Peker, I. Advances in Radiographic Techniques Used in Dentistry;
IntechOpen, 2015; Chapter 34.
5. Hung, K.; Montalvao, C.; Tanaka, R.; Kawai, T.; Bornstein, M. M. The Use and
Performance of Artificial Intelligence Applications in Dental and Maxillofacial
Radiology: A Systematic Review. Dentomaxillofacial Radiol. 2020, 49 (1), 20190107.
6. Cavalcanti, M. D. G. P.; Ruprecht, A.; Vannier, M. W. 3D Volume Rendering Using
Multislice CT for Dental Implants. Dentomaxillofacial Radiol. 2002, 31 (4), 218–223.
7. Kanuri, N.; Abdelkarim, A. Z.; Rathore, S. A. Trainable WEKA (Waikato Environment
for Knowledge Analysis) Segmentation Tool: Machine-Learning-Enabled Segmentation
on Features of Panoramic Radiographs. Cureus 2022, 14 (1).
214 Computational Imaging and Analytics in Biomedical Engineering

8. Salih, O.; Duffy, K. J. The Local Ternary Pattern Encoder–Decoder Neural Network for
Dental Image Segmentation. IET Image Process. 2022, 1–11. https://doi.org/10.1049/
ipr2.12416.
9. Park, J.; Lee, J.; Moon, S.; Lee, K. Deep Learning Based Detection of Missing Tooth
Regions for Dental Implant Planning in Panoramic Radiographic Images. Appl. Sci.
2022, 12 (3), 1595.
10. van der Stelt, P. F. From Pixel to Image Analysis. Dentomaxillofacial Radiol. 2021, 50
(2), 20200305.
11. Nafi'iyah, N.; Fatichah, C.; Astuti, E. R.; Herumurti, D. The Use of Pre and Post
Processing to Enhance Mandible Segmentation using Active Contours on Dental
Panoramic Radiography Images. In 2020 3rd International Seminar on Research of
Information Technology and Intelligent Systems (ISRITI); IEEE, Dec 2020; pp 661–666.
12. Loubele, M.; Jacobs, R.; Maes, F.; Denis, K.; White, S.; Coudyzer, W.; ... & Suetens, P.
Image Quality vs Radiation Dose of Four Cone Beam Computed Tomography Scanners.
Dentomaxillofacial Radiol. 2008, 37 (6), 309–319.
13. Abdi, A. H.; Kasaei, S.; Mehdizadeh, M. Automatic Segmentation of Mandible in
Panoramic X-ray. J. Med. Imag. 2015, 2 (4), 044003.
14. Chuang, Y. J.; Doherty, B. M.; Adluru, N.; Chung, M. K.; Vorperian, H. K. A Novel
Registration-Based Semi-Automatic Mandible Segmentation Pipeline Using Computed
Tomography Images to Study Mandibular Development. J. Comput. Assist. Tomogr.
2018, 42 (2), 306.
15. Inoue, K.; Nakano, H.; Sumida, T.; Yamada, T.; Otawa, N.; Fukuda, N.; Mori, Y. A
Novel Measurement Method for the Morphology of the Mandibular Ramus Using
Homologous Modelling. Dentomaxillofacial Radiol. 2015, 44 (8), 20150062.
16. Kumar, A.; Bhadauria, H. S.; Singh, A. Descriptive Analysis of Dental X-ray Images
Using Various Practical Methods: A Review. PeerJ Comput. Sci. 2021, 7, e620.
17. Chikmurge, D.; Harnale, S. Feature Extraction of DICOM Images Using Canny Edge
Detection Algorithm. In International Conference on Intelligent Computing and
Applications; Springer: Singapore, 2018; pp 185–196.
18. Otsu, N. A Threshold Selection Method from Gray-Level Histograms. IEEE Trans. Syst.
Man Cybern. 1979, 9 (1), 62–66.
19. Huang, L. K.; Wang, M. J. J. Image Thresholding by Minimizing the Measures of
Fuzziness. Pattern Recogn. 1995, 28 (1), 41–51.
20. Fadili, A.; Alehyane, N.; Halimi, A.; Zaoui, F. An Alternative Approach to Assessing
Volume-of-Interest Accuracy Using CBCT and ImageJ Software: In Vitro Study. Adv.
Radiol. 2015.
21. Menon, N. G.; Shrivastava, A.; Bhavana, N. D.; Simon, J. Deep Learning Based
Transcribing and Summarizing Clinical Conversations. In 2021 Fifth International
Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC); IEEE,
2021; pp 358–365.
CHAPTER 14

AN INVESTIGATION ON DIABETES
USING MULTILAYER PERCEPTRON
J. SURENDHARAN1, A. KARUNAMURTHY2, R. PRAVEENA3, and
K. SHEBAGADEVI3
1
HKBK College of Engineering, Bengaluru, Karnataka, India
2
BWDA Arts and Science College, Villupuram, India
3
Department of ECE, Muthayammal Engineering College, Namakkal,
India

ABSTRACT

The breakthroughs in public healthcare infrastructure have resulted in a large

influx of highly sensitive and critical healthcare information. The application
of sophisticated data analysis techniques can aid in the early detection and
prevention of a variety of fatal diseases. Diabetes can cause heart disease,
renal disease, and nerve damage, all of which are life-threatening complica­
tions of the disease. The goal of this work is to identify, detect, and forecast the
emergence of diabetes in its earliest stages by employing machine learning
(ML) techniques and algorithms. When it comes to diabetes classification,
a multilayer perceptron (MLP) is used. The experimental evaluation was
carried out using the Pima Indian diabetes dataset. According to the study
findings, MLP outperforms the competition in terms of accuracy, with an
accuracy rate of 86.08%. Following this, a comparison of the suggested tech­
nique with the existing state of the art is carried out, proving the flexibility
of the proposed approach to a wide range of public healthcare applications.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
216 Computational Imaging and Analytics in Biomedical Engineering

14.1 INTRODUCTION

Diabetes is referred to as diabetes mellitus (DM) and hypoglycemia, often

known as low blood sugar, in which the body is unable to synthesize glucose
in the bloodstream. Type 1 diabetes is a prevalent type, while type 2 diabetes
is the second most common.1 Type 1 is more likely to develop in children
than in adults. The immune system in a person attacks the pancreas with type
1, causing it to generate antibodies that cause the organ to malfunction and
cease generating insulin. It is possible to develop substantial issues and the
nervous system, despite the fact that it is less severe than type 1 diabetes.2
Diabetes mellitus is a key topic of medical study because of the disease’s
enormous social impact and the massive amount of data that is generated as
a result of this impact.3 Machine learning (ML) approaches in DM diagnosis,
management, and other clinical administration elements are certainly of
major concern. An ensemble strategy for diabetes classification utilizing ML
has been proposed in this article, which builds on a range of methodologies
that have been explored previously.4–8
Obstacles such as gestational DM and obesity occur during pregnancy
and have a long-term influence on the health of both the mother and the child.
In nonpregnant women, because of the relationship between the microvas­
cular block risk and the level of glucose in the blood, several criteria for
recognizing diabetes have been developed. As a result of the improvement in
living standards, diabetes has become increasingly prevalent. The only way
to avoid its problems is to discover and diagnose them early.9
There has been a significant amount of research on disease prediction,
including diagnosis, prediction, classification, and treatment. Various ML10–15
methods have been developed and applied to the identification and prediction
of diseases throughout the last few years. The implementation of ML with
conventional approaches has resulted in significant improvements. Machine
learning has proved its ability to deal with enormous numbers of variables
in an efficient and effective manner while simultaneously developing robust
prediction models. Supervised ML methods are used to investigate the inde­
pendent terms and variables of the dependent term.
Biomedical datasets are transformed into usable information through
this process, which allows for top-tier clinical research while also enhancing
patient care. As previously noted, there have been numerous improvements
in ML approaches as a result of the requirement to classify diabetic patients.
The Pima Indian diabetes dataset is available to the public. There is a total
An Investigation on Diabetes Using Multilayer Perceptron 217

of 769 data points, 500 of which are free of diabetes and 268 of which are
positive for the presence of diabetes.
According to the research history, a variety of ML algorithms is used on
this dataset for the aim of disease classification, none of which has reached an
accuracy of more than 76%, according to the research history. As a result, we
came up with the idea of improvising as a group rather than individually. The
subject of this research is ML models, and it investigates their performance,
theory, and attributes in greater depth.
The classification approach has been adopted by scientists in place of the
regression strategy for making disease predictions. Its performance has been
assessed using the accuracy, precision, recall, and F1-score of the aforemen­
tioned algorithm as measures of its effectiveness.

14.2 RELATED WORKS

Healthcare professionals rely on correct diagnoses to perform their duties

effectively. Diagnosing a patient’s type of diabetes necessitates a variety of
different tests, making it one of the most difficult duties faced by medical
practitioners. Considering too many variables at the same time when
diagnosing has the potential to produce erroneous conclusions. As a result,
determining the kind of diabetes that someone has is incredibly difficult to
accomplish. Recent years have seen a significant impact on the healthcare
industry, particularly as a result of ML techniques. Numerous investigations
have been conducted to determine the classification of diabetes.
Qawqzeh et al.15 came up with a logistic regression for the classifica­
tion of type 2 diabetes. It was built using data from 459 training, and 128
testing. When their proposed system was used to categorize 552 people as
nondiabetic, they were able to attain an accuracy rate of 92%. The proposed
method, on the other hand, has not been evaluated in comparison to existing
methods.
Pethunachiyar16 developed a technique for categorizing patients with DM
by employing a support vector machine (SVM). According to him, linear
SVM is more efficient than any of the other models that have been discussed
thus far. Despite this, there is no comparison of current state-of-the-art tech­
nology, and there are no specifics about parameter selection.
Diabetes was classified using Naive Bayes (NB) classifiers and SVMs
classification. These conclusions were drawn from the Pima Indian diabetes
218 Computational Imaging and Analytics in Biomedical Engineering

study, which used data from the Pima Indian diabetes study. The current as
well as the level of accuracy obtained, are notably absent from the document.
Choubey et al.18 conducted a study in which they compared several
diabetes classification techniques. The datasets utilized were a local diabetes
dataset and Pima Indian datasets. Feature engineering was carried out using
principal component analysis (PCA) and linear discriminant analysis (LDA),
both of which were shown to be beneficial in boosting the accuracy of the
classification method and removing undesired features from the dataset.
An ML paradigm that they developed was used to identify and predict
diabetes by Maniruzzaman et al.19 They employed four ML techniques for
the classification of diabetes: Naive Bayes, decision trees, AdaBoost, and
random forests, among others. Additionally, they used three alternative parti­
tion techniques in addition to the 20 trials they conducted. The researchers
used data from the National Health and Nutrition Survey (NHNS) for both
diabetic and nondiabetic patients to put their innovative technique through
its paces.
Ahuja et al.20 conducted an examination of ML algorithms, including
neural networks, deep learning, and multilayer perceptrons (MLPs), on
the Pima dataset for diabetic classification. When compared to the data,
MLP was determined to be superior to other classifiers. According to the
authors, fine-tuning and efficient feature engineering can help to increase the
performance of MLP algorithms. Recent research by Mohapatra et al.21 has
demonstrated the use of MLP to classify diabetes.
Singh and Singh22 proposed an ensemble method for the prediction of
type 2 diabetes to improve accuracy. The Pima dataset from the University
of California, Irvine Machine Learning Repository was used in this work.
The bootstrap approach with cross-validation was used to train the four base
learners of the stacking ensemble, which were used to train the stacking
ensemble four base learners. However, neither variable selection nor a
comparison of the current state of the art is mentioned, though.
Kumari et al.23 created a soft computing-based diabetes prediction system
based on an ensemble of three commonly used supervised ML algorithms.
It was discovered that they had used Pima datasets in their investigation.
When they compared their system performance to that of state-of-the-art
individual and ensemble approaches, they discovered that it outperformed
them by 79%.
To forecast diabetes in its early or onset stage, Islam et al.24 used a
combination of techniques. Training methods included cross-validation and
percentage splits, which were both employed in this study. They collected
An Investigation on Diabetes Using Multilayer Perceptron 219

data from 529 Bangladeshi patients, both diabetic and nondiabetic, using
questionnaires administered at a hospital in the nation. The experimental
results reveal that the random forest algorithm outperforms them all by a
significant margin. Although there is no comparison to the present state of
the art, there is no clear reporting of the accuracy that was attained in this
study.
The use of ML approaches to predict early and ongoing DM in females has
been demonstrated in several studies.25 They employed typical ML methods
to construct a framework for predicting diabetes to better understand the
disease.
Using ML models for diabetes prediction that were published between
2010 and 2019, Hussain and Naaz26 did a comprehensive evaluation of the
literature on this topic. They evaluated the algorithms based on the Matthews
correlation coefficient and discovered that Naive Bayes and random forests
outperformed the other algorithms in terms of overall performance.

14.3 PROPOSED METHOD

FIGURE 14.1 Proposed architecture for diabetes classification.

A variety of ML techniques have been applied to make decisions in the

proposed diabetes classification system. For classification, we used and fine-
tuned MLP for the first time, owing to its good performance in healthcare,
notably in the prediction of diabetic complications, and the illustration is
given in Figure 14.1.
220 Computational Imaging and Analytics in Biomedical Engineering

14.3.1 CLASSIFICATION

The study makes improvements to three commonly used cutting-edge

strategies for identifying diabetes that were previously published. For
the most part, the proposed ways of putting a person into one of the two
categories of diabetes are compared and contrasted against one another.
The following are the various diabetic treatment options that have been
offered.
MLP: Our experimental setup incorporates a well-tuned MLP for diabetes
classification that was developed by our team of researchers. As indicated in
Figure 14.2, a classification method is represented as a network of numerous
layers that are connected to one another. Perceptrons are the fundamental
building blocks of our paradigm, and they are linear combinations of input
and weights. When developing Algorithm 14.1, we employed a sigmoid unit
as the activation function.
The suggested method is comprised of three major phases, as follows:
The weights are initially established, then the output layer (δk) output is
calculated by applying the sigmoid activation function to the weights and
layer outputs. Second, the inaccuracy of all concealed units is determined at
the hidden layer level (δh).
Finally, to reduce network errors, all network weights (wi, j) are
updated in a backward manner. A step-by-step summary of the entire
diabetes classification procedure is provided by algorithm 1 for diabetes
classification.

Algorithm 14.1 Classification using MLP.

Input: Total patients, skin thickness, BP, age, BMI, HB weight
Output: A Trained Model
Method: Initialize weights with a random number
while (Weights ≤ Threshold) do
for Training samples, do
Input samples is set as input to the network
Compute the required output;
end for
for output k, do
δk = Ok(1 − Ok)(tk − Ok)
end for
An Investigation on Diabetes Using Multilayer Perceptron 221

for hidden layer h do

δh = Oh(1-Oh) Σk∈outputwh,kδk
end for
Update weight of each network wi,j;
wi,j = wi,j + Δwi,j; Δwi,j = ηδjxi,j
End

Because there are eight different variables in the MLP classification

model, eight neurons of the MLP classification model. When computing the
weights and inputs in the middle layer, the sigmoid unit will be employed
as a calculator. At the end of the process, the output layer will calculate the
results.
The hidden layers are used to compute the output data, starting with the
values and weights of the input layer as a starting point for computation.
Sigmoid activation functions are applied to each unit in the hidden layer to
reduce the vast amount of data to a more manageable range between 0 and
1. This computation can be used by every middle layer. It is also treated with
the same technique as the input layer, yielding results that can be used to
create predictions regarding the development of diabetes.

14.4 RESULTS AND DISCUSSION

To put the proposed diabetes classification system to the test, the Pima
Indian diabetes dataset is employed. A comparison study is also carried
out, utilizing the most up-to-date computational techniques, which are
also included in the package. On the basis of the experimental results, the
suggested method outperforms the currently available algorithms in terms
of performance. Included in this section are sections devoted to defining the
dataset, performance indicators, and conducting a comparison study.

14.4.1 DATASET

This study made use of the Pima Indian diabetes dataset. Create an intelligent
model for predicting if a person has diabetes based on some of the metrics
contained in this dataset using data from this dataset. When it comes to the
classification of diabetes, it is a binary classification problem to be solved.
The variables are shown in Table 14.1.
222 Computational Imaging and Analytics in Biomedical Engineering

TABLE 14.1 Description of Variables in the Dataset.

Attributes Range Mean Standard deviation
Pregnancies 0–17 3.946 3.454
Glucose 0–199 124.025 32.800
BP 0–122 70.828 19.783
Skin thickness 0–99 21.013 16.298
Insulin 0–846 81.795 117.875
BMI 0–67 32.800 8.077
Diabetes pedigree function 0.078–2.4 0.482 0.338
Age 21–81 34.030 12.095
Outcome Y/N 0.359 0.492

As shown in Figure 14.4, the dataset contains 768 records of female

patients over the age of 21 who are either healthy or diabetic, as determined
by their blood glucose levels. The distribution of feature values is depicted
in Figure 14.5. For the outcome of the target variable, there are only two
potential values: 0 and 1. This dataset was used to get an accurate diagnosis
of type 2 diabetes. Pima Indian diabetes dataset determines whether or not
a user is at risk of acquiring diabetes in the next four years based on their
behavior.
To compare the proposed classification system with existing method­
ologies, this research employs the same experimental design as the Pima
Indian dataset from the previous paper. The performance measures that were
employed and the outcomes that were obtained for classifying or predicting
are explained further below. It is also possible to make a comparison with
previous investigations.
The performance is evaluated using three extensively used performance
measures: recall, precision, and accuracy. Recall, precision, and accuracy are
three frequently used, measures to evaluate the performance of the presented
techniques.
Individuals who are not diabetic are appropriately identified as such in
true positive (TP), whereas diabetic patients are correctly identified as such
in true negative (TN). The patient has diabetes, as revealed by the false nega­
tive (FN), but a favorable result is indicated. The patient’s false positive (FP)
also suggests that he or she is a healthy individual who has been classified
as having diabetes, which is a positive sign. The classification models were
trained and evaluated using a 10-fold cross-validation approach, which was
developed by the authors.
An Investigation on Diabetes Using Multilayer Perceptron 223

FIGURE 14.2 Accuracy.

FIGURE 14.3 Precision.

The Pima dataset is used to assess three state-of-the-art diabetic clas­

sification classifiers that have been developed recently. When comparing the
accuracy of the fine-tuned MLP algorithm in Figure 14.2 to that of present
systems, the highest accuracy of 86.083% can be observed. The precision
rate has been increased than the existing method which is shown in Figure
14.3. The comparison of the recall rate with the existing method is shown in
Figure 14.4.
224 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 14.4 Recall.

FIGURE 14.5 F-Measure.

An Investigation on Diabetes Using Multilayer Perceptron 225

In light of the findings, the calibrated MLP model might be utilized to

accurately classify diabetes, as demonstrated. We believe that the classifica­
tion approach we have described here will be beneficial to our hypothetical
system in the future.
According to Figure 14.5, a proposed MLP algorithm outperforms with
an accuracy of 86.083% when compared to the current state of the art (86.6%
precision and 85.1% recall). These discoveries will have a significant impact
on the planned hypothetical method for evaluating whether a patient has type
1 or type 2 diabetes.

14.5 CONCLUSIONS

In this paper, the authors proposed a model for supporting the healthcare
business. The study developed an algorithm for the classification of diabetes
that was based on MLPs. The primary purpose of the proposed system is to
aid users in keeping track of their vital signs through the use of their mobile
phones and other mobile devices. Users will be able to recognize their elevated
risk of diabetes at an earlier stage as a result of the model projections about
future blood glucose levels, which is an extra benefit. Diabetic patients are
classified and predicted using MLP. The proposed methodologies are tested
on the Pima Indian diabetes dataset, which is available online. In terms of
accuracy, the two approaches outperform existing best practices by 86.083
and 87.26%, respectively, when compared to current best practices.

KEYWORDS

• fatal diseases
• diabetes mellitus
• prediction
• classification

REFERENCES

1. Miranda, M. A.; Macias-Velasco, J. F.; Lawson, H. A. Pancreatic β-Cell Heterogeneity

in Health and Diabetes: Classes, Sources, and Subtypes. Am. J. Physiol.-Endocrinol.
Metabol. 2021, 320 (4), E716–E731.
226 Computational Imaging and Analytics in Biomedical Engineering

2. Babalola, O. M.; Garcia, T. J.; Sefcik, E. F.; Peck, J. L. Improving Diabetes Education in
Mexican American Older Adults. J. Transcult. Nurs. 2021, 32 (6), 799–809.
3. Zaccardi, F.; Kloecker, D. E.; Buse, J. B.; Mathieu, C.; Khunti, K.; Davies, M. J. Use of
Metformin and Cardiovascular Effects of New Classes of Glucose-Lowering Agents: A
Meta-Analysis of Cardiovascular Outcome Trials in Type 2 Diabetes. Diab. Care 2021,
44 (2), e32–e34.
4. Kannan, S.; Idrees, B. A.; Arulprakash, P.; Ranganathan, V.; Udayakumar, E.; Dhinakar,
P. Analysis of Convolutional Recurrent Neural Network Classifier for COVID-19
Symptoms Over Computerised Tomography Images. Int. J. Comput. App. Technol.
2021, 66 (3–4), 427–432.
5. Saravanan, D.; Surendiran, J. A New Framework for Video Data Retrieval Using
Hierarchical Clustering Technique. Int. J. Eng. Adv. Technol. (IJEAT) ISSN: 2249-8958
2019, 8 (6S3).
6. Ojugo, A. A.; Ekurume, E. Predictive Intelligent Decision Support Model in Forecasting
of the Diabetes Pandemic Using a Reinforcement Deep Learning Approach. Int. J. Educ.
Manag. Eng 2021, 11 (2), 40–48.
7. Khunti, K.; Knighton, P.; Zaccardi, F.; Bakhai, C.; Barron, E.; Holman, N. et al.
Prescription of Glucose-Lowering Therapies and Risk of COVID-19 Mortality in
People with Type 2 Diabetes: A Nationwide Observational Study in England. Lancet
Diab. Endocrinol. 2021, 9 (5), 293–303.
8. Syed, S. A.; Sheela Sobana Rani, K.; Mohammad, G. B.; Chennam, K. K.; Jaikumar, R.;
Natarajan, Y. et al. Design of Resources Allocation in 6G Cybertwin Technology Using
the Fuzzy Neuro Model in Healthcare Systems. J. Healthcare Eng. 2022.
9. Greiver, M.; Havard, A.; Bowles, J. K.; Kalia, S.; Chen, T.; Aliarzadeh, B. et al. Trends
in Diabetes Medication use in Australia, Canada, England, and Scotland: A Repeated
Cross-Sectional Analysis in Primary Care. Br. J. Gen. Pract. 2021, 71 (704), e209–e218.
10. Vidhya, R. G.; Batri, K. Segmentation Classification of Breast Cancer Using an Krill
Herd Oprimization. Med. Imaging Health Inf. 2020, 10 (6), 1294–1300.
11. Anandaraj, S. P.; Kirubakaran, N.; Ramesh, S.; Surendiran, J. Efficient Way to Detect
Bone Cancer Using Image Segmentation. International Journal of Engineering and
Advanced Technology (IJEAT) ISSN: 2249-8958, 2019, 8 (6S3).
12. Ferrari, M.; Speight, J.; Beath, A.; Browne, J. L.; Mosely, K. The Information-
Motivation-Behavioral Skills Model Explains Physical Activity Levels for Adults with
Type 2 Diabetes Across all Weight Classes. Psychol. Health Med. 2021, 26 (3), 381–394.
13. Shah, N.; Karguppikar, M.; Bhor, S.; Ladkat, D.; Khadilkar, V.; Khadilkar, A. Impact of
Lockdown for COVID-19 Pandemic in Indian Children and Youth with Type 1 Diabetes
from Different Socio-Economic Classes. J. Pediatric Endocrinol. Metabol. 2021, 34
(2), 217–223.
14. Mansi, I. A.; Chansard, M.; Lingvay, I.; Zhang, S.; Halm, E. A.; Alvarez, C. A.
Association of Statin Therapy Initiation with Diabetes Progression: A Retrospective
Matched-Cohort Study. JAMA Intern. Med. 2021, 181 (12), 1562–1574.
15. Bouyahya, A.; El Omari, N.; Elmenyiy, N.; Guaouguaou, F. E.; Balahbib, A.; Belmehdi,
O. et al. Moroccan Antidiabetic Medicinal Plants: Ethnobotanical Studies, Phytochemical
Bioactive Compounds, Preclinical Investigations, Toxicological Validations and Clinical
Evidences; Challenges, Guidance and Perspectives for Future Management of Diabetes
Worldwide. Trends Food Sci. Technol. 2021, 115, 147–254.
An Investigation on Diabetes Using Multilayer Perceptron 227

16. Saritha, G.; Saravanan, T.; Anbumani, K.; Surendiran, J. Digital Elevation Model and
Terrain Mapping Using LiDAR. Mater. Today: Proc. 2021, 46 (9), 3979–3983. ISSN
2214-7853.
17. Vidhya, R. G.; Saravanan, G.; Rajalakshmi, K. Mitosis Detection for Breast Cancer
Grading. Int. J. Adv. Sci. Technol. 2020, 29 (3), 4478–4485; Gupta, S.; Verma, H.
K.; Bhardwaj, D. Classification of Diabetes Using Naive Bayes and Support Vector
Machine as a Technique. In Operations Management and Systems Engineering;
Springer: Singapore, 2021; pp 365–376.
18. Choubey, D. K.; Kumar, M.; Shukla, V.; Tripathi, S.; Dhandhania, V. K. Comparative
Analysis of Classification Methods with PCA and LDA for Diabetes. Curr. Diab. Rev.
2020, 16 (8), 833–850.
19. Satheeshwaran, U.; Sreekanth, N.; Surendiran, J. X-ray CT Reconstruction by Using
Spatially Non Homogeneous ICD Optimization. Int. J. Eng. Adv. Technol. (IJEAT),
ISSN: 2249-8958, 2019, 8 (6S3).
20. Ahuja, R.; Sharma, S. C.; Ali, M. A Diabetic Disease Prediction Model Based on
Classification Algorithms. In Annals of Emerging Technologies in Computing (AETiC),
Print ISSN, 2019; pp 2516–0281.
21. Mohapatra, S. K.; Swain, J. K.; Mohanty, M. N. Detection of Diabetes Using Multilayer
Perceptron. In International Conference on Intelligent Computing and Applications;
Springer: Singapore, 2019; pp 109–116.
22. Singh, N.; Singh, P. Stacking-Based Multi-Objective Evolutionary Ensemble Framework
for Prediction of Diabetes Mellitus. Biocybern. Biomed. Eng. 2020, 40 (1), 1–22.
23. Surendiran, J.; Saravanan, S. V.; Elizabeth Catherine, F. Glaucoma Detection Using
Fuzzy C- Mean (FCM). IJPT 2016, 8 (3), 16149–16163.
24. Islam, M. M.; Ferdousi, R.; Rahman, S.; Bushra, H. Y. Likelihood Prediction of Diabetes
at Early Stage Using Data Mining Techniques. In Computer Vision and Machine
Intelligence in Medical Image Analysis; Springer: Singapore, 2020; pp 113–125.
25. Malik, S.; Harous, S.; El-Sayed, H. Comparative Analysis of Machine Learning
Algorithms for Early Prediction of Diabetes Mellitus in Women. In International
Symposium on Modelling and Implementation of Complex Systems; Springer: Cham,
2020; pp 95–106.
26. Hussain, A.; Naaz, S. Prediction of Diabetes Mellitus: Comparative Study of Various
Machine Learning Models. In International Conference on Innovative Computing and
Communications; Springer: Singapore, 2021; pp 103–115.
CHAPTER 15

DERMOSCOPIC IMPLEMENTATION
AND CLASSIFICATION ON
MELANOMA DISEASE USING
GRADIENT BOOST CLASSIFIER
B. BALAKUMAR1, K. SAKTHI MURUGAN2, N. SURESHKUMAR3,
A. PURUSHOTHAMAN4
1
Centre for Information Technology and Engineering, Manonmaniam
Sundaranar University, Tirunelveli, India
2
Department of ECE, PSN College of Engineering and Technology,
Tirunelveli, India
3
Department of ECE, Muthayammal College of Engineering,
Rasipuram, India
4
Department of ECE, Hindhusthan Institute of Technology, Coimbatore,
India

ABSTRACT

Melanoma is a form of skin cancer that grows when melanocyte become out
of balance (the cells that render the skin tan or brown). Cancer starts with
cells developing out of balance in the body. Cells may become cancer in
virtually every part of the body and then spread to other parts of the body.
The melanoma of certain other skin cancers is considerably less common.
However, melanoma is harmful as it travels to certain parts of the body even
faster than detected and handled. In this study, we propose a deep learning

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
230 Computational Imaging and Analytics in Biomedical Engineering

techniques for classify a deadly disease. Dermoscopic images are used in

this study as a dataset directory. We use the classifier of gradient boost and
feature extraction by NSCT to predict accuracy and specificity value of
melanoma disease.

15.1 INTRODUCTION

Melanoma is one of the skin cancer’s deadliest types of disease. Late diag­
nosis is curable, although only professionally qualified specialists can reliably
diagnose the disease. Despite insufficient availability of resources, computer
devices that can classify pathogens can save lives, decrease unwanted
biopsies and rising prices. In order to do this, we propose a framework that
combines recent advances in profound learning with existing method of
machine learning, creates sets of methods which segment skin lesions and
analyzes the detected region and the surrounding tissue to detect melanoma.
In people’s daily life, skin diseases are very normal.1
Millions of individuals in the United States suffer various forms of
skin conditions annually. Diagnosing skin disorders also requires a strong
degree of knowledge in the different aspects of their vision. Because human
judgment is always arbitrary and hard to replicate, a computer-aided diag­
nostic device would be considered to obtain a more accurate and effective
diagnosis. In this2 article, we investigate the viability in the development
of a deep convolutionary neural network (CNN), a standardized diagnostic
method of skin diseases.
One out of five Americans was treated in their lifespan with cutaneous
malignancy. While melanomas constitute less than 5% of all skin cancers in
the United States, nearly 75% of all deaths due to skin cancer and more than
10,000 deaths are recorded in the United States alone per year. Early detec­
tion is important, as the average 5-year melanoma survival rate decreases
from 99% in the earliest stages to around 14% in the latest level. We also3
established a statistical approach that can proactively map skin lesions and
diagnose cancer sooner by physicians and patients.
Second, many AL training functions depend on model instability, but
such model vulnerability is not always reflected by profound learning
methods. In this article, we incorporate in a realistic way recent develop­
ments in Bayesian5 profound learning in the active learning sense. With
very limited current literature, we are designing an active learning system
for higher dimensional data that have been incredibly challenging to date.
Dermoscopic Implementation and Classification 231

We demonstrate our active learning strategies with images, using special

models such as Bayesian CNNs, to greatly enhance current active learning
approaches.
Dermoscopy is a technology for capturing skin images, and these images
are important for the study of different forms of skin diseases. Malignant
melanoma is a form of skin cancer that often causes death to its extent.
Previous melanoma diagnosis6 delays mortality and doctors should treat
patients in order to improve their likelihood of survival. There are several
machine learning methods to diagnose the melanoma by its characteristics.
Melanoma is a type of skin cancer that is most active and increasing. There
is a development in work exploring the usage of photographs taken by
automated cells to computerize-reported lesions of skin for malignancy. The
analysis of these images7 is usually difficult due to disturbing factors like light
variations and light reflections on the surface of the skin. The segmentation
of the lesion region from the usual skin is a significant phase in the treatment
of melanoma. In any field, a correct choice of the learning algorithm and
its statistician validation are necessary for the use of learning techniques to
improve performance. Computer-aided diagnosis8 of skin cancer is difficult
given the relative lack of labeling data for lesions, and therefore, the low
quality of available training data.
Second, there are some areas that do not take into consideration the
research domain for automated diagnosis of skin cancer in the balance
between model complexity and generalization capability of the final clas­
sification model.

15.2 METHODS AND MATERIALS

15.2.1 NSCT ALGORITHM FOR FEATURES EXTRACTION

Extraction function is the first element from which dermoscopic input

images remove the basic characteristic values. A Laplacian pyramid and
directional characteristics can be used for detecting dot discontinuities using
a directional filter bank for the curtain transition with a double-iterated
filter bank. The comparison of the provided picture to a smooth contour
at different decomposition rates is feasible.4 This displays multi-scale and
spatial characteristics while at specific sizes, this decomposes the dermo­
scopic pictures. As a consequence of the down-and-up sampler in both the
directional filter bank and the Laplacian pyramid, the contourlet transform
232 Computational Imaging and Analytics in Biomedical Engineering

will have a redundant 33%, NSCT is being used. The sample architecture
diagram is shown in Figure 15.1.

FIGURE 15.1 Architectural diagram for melanoma classification.

15.2.2 CLASSIFIER EXPLANATION

Gradient boosting is a machine learning technique that generates a predictive

model in the form of a community of low predictor models usually decision
trees for regression and classification. This constructs the model as other
boosting approaches do and generalizes it by maximizing a differentiable
loss function randomly.

15.3 RESULT AND DISCUSSION

Dermoscopic images are an easy way to capture melanoma images. In this

study, we use 60 images normal and abnormal. From the procedure of pre­
processing and feature extracting normal and abnormal images has been
shown in Figures 15.2 and 15.3.
Dermoscopic Implementation and Classification 233

FIGURE 15.2 Normal images.

FIGURE 15.3 Abnormal images.

Classification of melanoma disease is done by deep learning classifier of

gradient boost as shown in Table 15.1

TABLE 15.1 Classification of Melanoma.

Classifier Accuracy Specificity
Gradient Boost 92.4% 88.7%

15.4 CONCLUSIONS

Melanoma identification process is the one of the important things in

biomedical field. In this study, we use deep learning libraries to classify the
disease whether it is or not by analysis report of classifiers. We implement
dermoscopic images as a dataset directory. In this directory, it is split up in
to normal and abnormal images. Total of 60 dermoscopic images used in
this directory. It is preprocessing in to 40 images as training and 20 images
as testing. Then use NSCT algorithm to build a feature extraction and
gradient boosting classifier to predict a result of melanoma disease. From
234 Computational Imaging and Analytics in Biomedical Engineering

the classifier, we got accuracy of 92.4% and specificity rate of 88.7% as a

prediction result for melanoma disease classification.

KEYWORDS

• dermoscopic
• NSCT
• deep learning
• gradient boost
• accuracy
• specificity

REFERENCES

1. Codella, N. C.; Nguyen, Q. B.; Pankanti, S.; Gutman, D. A.; Helba, B.; Halpern, A.
C.; Smith, J. R. Deep Learning Ensembles for Melanoma Recognition in Dermoscopy
Images. IBM J. Res. Dev. 2017, 61 (4/5), 5–1.
2. Liao, H. A Deep Learning Approach to Universal Skin Disease Classification; University
of Rochester Department of Computer Science, CSC.
3. Esteva, A.; Kuprel, B.; Novoa, R. A.; Ko, J.; Swetter, S. M.; Blau, H. M.; Thrun, S.
Dermatologist-Level Classification of Skin Cancer with Deep Neural Networks. Nature
2017, 542 (7639), 115–118.
4. Sonia, R. Melanoma Image Classification System by NSCT Features and Bayes
Classification. Int. J. Adv. Signal Image Sci. 2016, 2 (2), 27–33.
5. Gal, Y.; Islam, R.; Ghahramani, Z. Deep Bayesian Active Learning with Image Data.
Proc. 34th Int. Conf. Mach. Learn. 2017, 70, 1183–1192.
6. Premaladha, J.; Ravichandran, K. S. Novel Approaches for Diagnosing Melanoma Skin
Lesions Through Supervised and Deep Learning Algorithms. J. Med. Syst. 2016, 40 (4),
96.
7. Jafari, M. H.; Karimi, N.; Nasr-Esfahani, E.; Samavi, S.; Soroushmehr, S. M. R.; Ward,
K.; Najarian, K. Skin Lesion Segmentation in Clinical Images Using Deep Learning.
In 2016 23rd International conference on pattern recognition (ICPR); IEEE, 2016; pp
337–342.
8. Masood, A.; Al-Jumaily, A.; Anam, K. Self-Supervised Learning Model for Skin Cancer
Diagnosis. In 2015 7th International IEEE/EMBS Conference on Neural Engineering
(NER); IEEE, 2015; pp 1012–1015.
CHAPTER 16

IMAGE PROCESSING AND DEEP

LEARNING TECHNIQUES FOR LUNG
DISEASE SEGMENTATION USING KNN
CLASSIFIER
G. SONIYA PRIYATHARSINI1, N. NAVEENKUMAR2,
BALACHANDRA PATTANAIK3, and M. KUMARASAMY4
1
Department of ECE, DR. M. G. R. Educational and Research
Institute, Maduravoyal, Chennai, India
2
Department of CSE, Muthayammal Engineering College, Rasipuram,
India
3
Department of Electrical and Computer Engineering, College of
Engineering and Technology, Wollege University, Ethiopia
4
Department of Computer Science, College of Engineering and
Technology, Wollege University, Ethiopia

ABSTRACT

Lung cancer refers to an uncontrollable breakdown of the cells in lungs.

This causes tumor development that reduces a person’s breathing capacity.
Throughout 2015, 218,527 individuals in the United States were infected
with lung cancer, according to the Centers for Disease Control. Early
warning will enable a person care in the course of the illness as quickly as
possible. However, it may be challenging to recognize the early lung cancer,
because the indications may be identical with no indicators at all to those of

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
236 Computational Imaging and Analytics in Biomedical Engineering

a coughing disorder. In this study, we implemented a segmented lung disease

prediction using deep learning techniques such as convolutional Neural
Network (CNN) and KNN classifier.

16.1 INTRODUCTION

Lung cancer is the most prevalent disease not to be missed which induces late
healthcare mortality. Now, CT is available support clinicians diagnose early
stage lung cancer.2 The test to diagnose lung cancer is also calculated by the
Doctors’ expertise, which certain patients can neglect and trigger problems.
Broad awareness has proven to be common and solid process in many testing
fields of medical imaging. Three forms of deep neural networks are included
in this paper (e.g., CNN, DNN, and SAE) are the calcification of lung cancer.
The CT images recognition function includes these networks. One of the
main methods used by pathologists for evaluating the stage, the types and
subtypes of lung cancers are the visual analysis of histopathological slides
of lung cell tissues. In this research, we learned a profound learning neural
convolutionary network (CNN) model (first v3).
This model1 was used to identify whole-slide pathological photos in
correct adenocarcinoma, squamous cell carcinomas, and typical lung tissue
from the Cancer Genome Atlas (TCGA). Deep learning in pattern recogni­
tion and classification is considered to be a popular and powerful method.
However, in the area of medical diagnostics imagery, there are not many
highly organized implementations because broad databases are not often
visible in medical images. In this3 research, we checked if deep learning
algorithms are feasible for the diagnosis of lung cancer in cases.

16.2 RELATED WORK

Lung cancer is one of the world’s main causes of death. At least, an appropriate
distinction of the clinical treatment of lung cancer types of cancer (adeno­
carcinoma, squamous carcinoma, and small cell carcinoma) is common.
Nonetheless, the quality increases yet diagnosis reliability is complicated.4
We also established an automated classification method in this analysis for
lung cancers. In microscopic photos, a big deeper learning technique is the
deep CNN (DCNN).
Late diagnosis of lung cancer will make a big difference decreased
death rate for lung cancer, and is blamed for more than 17% of overall
Image Processing and Deep Learning Techniques 237

deaths diagnosed with disease. Radiologists report a considerable number

of incidents. Diagnosis with machine support (computer-aided diag­
nostic [CAD]), radiologists can be helped by offering a second opinion
and speeding up the whole operation.5 We suggest a CAD system with
profound functionality extracted from auto encoder of be malignant or
malignant lung nodules friendly. A never-ending number of chest CT scans
will be carried out in the near future by the launch of lung cancer screening
programs, which will require radiologists to read to decide on a patient
monitoring strategy.
The function of screen-detected nodules is highly dependent on the
nodule size and form according to the existing guidelines. We present in
this6 paper, a method of deep learning, focused on multi-stream convolu­
tional network that classifies all nodule forms important to the function of
nodules automatically. In order to obtain quantitative tumor7 separation
outcomes, a traditional CAD scheme includes many image processing and
pattern recognition processes. Lung CT image segmentation is an essential
first phase in the study of the plum image, a preconditions for effective
plum CT8 examination such as the detection of lung cancer. In the society,
image segmentation plays a vital role in identification of objects, persons,
etc. The image representation9 is changed by the method of image segmen­
tation. This will quickly alter restrictions and borders.

16.3 METHODS AND MATERIALS

16.3.1 PRE-PROCESSING

The first step is pre-processing the images for the intensity measurement
of particles. The use of standard segmentation technology is to divide the
processed image. The image is, therefore, filled with cancer nodules. Other
characteristics such as area, perimeter, and eccentricities were extracted
during extraction of features such as centroid, diameter, and median intensity
of the pixels. This is accompanied by a classification module that distin­
guishes between benign and malignant tumors based on the CT scan images.
Extracted functions are used as preparation and the corresponding training
model is developed for the classification followed by model evaluation with
increasing precision, accuracy, specificity, and sensitivity for detection and
classification.
238 Computational Imaging and Analytics in Biomedical Engineering

16.4 FEATURES EXTRACTION BY CONVOLUTIONAL NEURAL

NETWORK (CNN)
A convolutionary neural network (ConvNet/CNN) is a deep learning algo­
rithm that can take an input image and assign importance to various aspects/
objects in the image (learnable weights and bias) and distinguish them
between them. The requirements for pre-processing are much lower than
other classification algorithms in a ConvNet, whereas filters are handcrafted
in primitive processes, ConvNets can learn these filters/characteristics with
adequate training.

FIGURE 16.1 Architecture diagram for CNN.

Source: Adapted from https://www.cgupta.tech/quickdraw.html

The ConvNet architecture Figure 16.1 is analogous to that of the neuron’s

communication system in human brain which was focused on the visual
cortex organization. In a small area of the visual field known as the recep­
tive zone, each human neuron reacts to stimulus only. The entire visual
area covers a collection of these fields. ConvNet measures the spatial and
temporal dependencies of a picture efficiently by utilizing the necessary
filters. The architecture fits in better with the image dataset because the
number of parameters involved is reduced and weights are reusable. This
means that the network can better understand the image’s sophistication.
Image Processing and Deep Learning Techniques 239

16.5 CLASSIFICATION TECHNIQUES USING RANDOM FOREST

CLASSIFIER

The KNN algorithm is a simple machine learning algorithm, easily managed,

which can be used to solve issues of classification as well as regression.
The KNN algorithm believes that there are related items nearby. In other
words, the KNN algorithm is based on that assumption that the algorithm
is true enough to help. KNN incorporates some of the algebra we might
have studied during our childhood—calculating the distance between points,
which is the same as size, closeness, and similarity (or similarity) in a line.
The implementation process has been shown in Figure 16.2.

16.6 RESULT AND DISCUSSION

Lung cancer is an early form of lung cancer. The lungs are two oxygen-
treating spongy bodies in the chest that inhale and release carbon dioxide as
you exhale. In this study, we implemented deep learning techniques for lung
disease prediction using lung CT images as a dataset. It is used to segment
images as shown in Figure 16.3. Also predict sensitivity, specificity, and
accuracy values using KNN classifier as shown in Table 16.1.

FIGURE 16.2 Prediction analysis of lung cancer using KNN classifier.

240 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 16.3 Lung Segmented images using CT scan images dataset.

TABLE 16.1 Analysis Report of Random Forest Classifier

Epoch Sensitivity (%) Specificity (%) Accuracy (%)
0–25 67 72 88
25–50 76 78 89.6
50–100 78.5 82.5 94.3

16.7 CONCLUSIONS

Deep learning is an algorithmic field of machine science focused on brain

structure and function, defined as artificial neural networks. In this study,
we propose a deep learning structure to predict a segmented lung disease
using CT scan images. It contains 100 images in dataset directory. It is pre­
processed to 60 images as training and 40 images as testing. Then the features
are extracted by both CNN and random boost classifier. Then the classifier
also predicts the value of sensitivity, specificity, and accuracy value. The
maximum accuracy was achieved through epoch process 50–100 for 94.3%.
For the better prediction result, we may use multiple classifier to analyze and
classification of lung disease.
Image Processing and Deep Learning Techniques 241

KEYWORDS

• deep learning
• CNN
• random boost classifier
• breathing
• lung cancer

REFERENCES

1. Coudray, N.; Ocampo, P. S.; Sakellaropoulos, T.; Narula, N.; Snuderl, M.; Fenyö, D.;
Moreira, A. L.; Razavian, N.; Tsirigos, A. Classification and Mutation Prediction from
Non–Small Cell Lung Cancer Histopathology Images Using Deep Learning. Nat. Med.
2018, 24 (10), 1559–1567.
2. Song, Q.; Zhao, L.; Luo, X.; Dou, X. Using Deep Learning for Classification of Lung
Nodules on Computed Tomography Images. J. Healthcare Eng. 2017, 2017.
3. Sun, W.; Zheng, B.; Qian, W. Computer Aided Lung Cancer Diagnosis with Deep
Learning Algorithms. In Medical Imaging 2016: Computer-Aided Diagnosis, Vol. 9785;
International Society for Optics and Photonics, 2016; p 97850Z.
4. Teramoto, A.; Tsukamoto, T.; Kiriyama, Y.; Fujita, H. Automated Classification of Lung
Cancer Types from Cytological Images Using Deep Convolutional Neural Networks.
BioMed Res. Int. 2017, 2017.
5. Kumar, D.; Wong, A.; Clausi, D. A. Lung Nodule Classification Using Deep Features
in CT images. In 2015 12th Conference on Computer and Robot Vision; IEEE, 2015;
pp 133–138.
6. Ciompi, F.; Chung, K.; Van Riel, S. J.; Setio, A. A. A.; Gerke, P. K.; Jacobs, C.; Scholten,
E. T.; Schaefer-Prokop, C.; Wille, M. M.; Marchiano, A.; Pastorino, U. Towards
Automatic Pulmonary Nodule Management in Lung Cancer Screening with Deep
Learning. Sci. Rep. 2017, 7, 46479.
7. Hua, K. L.; Hsu, C. H.; Hidayati, S. C.; Cheng, W. H.; Chen, Y. J. Computer-Aided
Classification of Lung Nodules on Computed Tomography Images via Deep Learning
Technique. OncoTargets Therap. 2015, 8.
8. Skourt, B. A.; El Hassani, A.; Majda, A. Lung CT Image Segmentation Using Deep
Neural Networks. Procedia Comput. Sci. 2018, 127, 109–113.
9. Manikandan, M. Image Segmentation and Image Matting for Foreground Extraction
using Active Contour Based Method. Int. J. MC Square Sci. Res. 2011, 3 (1), 18–38.
CHAPTER 17

DESIGN DETECTING AND

CLASSIFYING MELANOMA SKIN
CANCER USING CNN WITH K MEANS
CLUSTERING
R. SUMATHI1 and SRIDHAR P. ARJUNAN2
Department of Computer Science and Engineering, Kalasalingam
1

Academy of Research and Education, Krishnankoil, India

2
Department of Electronics and Instrumentation Engineering, SRM
Institute of Science and Technology, Kattankulathur, India

ABSTRACT

Many people in the universe are highly affected with skin-related cancer
diseases at present. Many researchers used many soft computing approaches
for detecting the cancer part. To aid the dermatologist, we propose an auto­
mated approach with Cuckoo search optimization and K-means clustering
for detecting the skin cancer and classifying the cancer as normal and
abnormal by support vector machine. For preprocessing, median filters are
used to reduce noise inference in the input image. For validating the accuracy
we utilize the IISC-DSI dataset and our proposed approach yields 98.2%
segmentation accuracy and 98.6% classification accuracy and compared
with ABC with K-means, FCM for segmentation and for classification CNN
and KNN. Our approaches take the average of 7s for processing the images
using hybrid approach for cancer detection.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
244 Computational Imaging and Analytics in Biomedical Engineering

17.1 INTRODUCTION

According to the world medical council, skin cancer is placed in the top place
of cancer. Among various skin cancers, melanoma is the most dangerous
skin cancer causing more than 80% peoples to lose their lives. It looks like
a mole and appears anywhere in the skin where the sunrays do not fall, and
it is brown in color which spreads fast and affects the blood vessels in the
skin. With the features like asymmetry, border, color, dark, and elevation
are the basic symptoms for melanoma skin cancer. Clustering is one of the
fundamental methods for understanding various objects and categorizes
it based on their similarity. Many clustering methods are used in various
applications like image processing,l–3 text mining, and so on. K-means
clustering provides fast convergence rate compared with other clustering
like hierarchical clustering. Cuckoo search optimization4 is recently devel­
oped optimization algorithm which is based on brood parasitic behaviors
of cuckoo. It is used in many medical applications for its global optimal
solution.5 Deep neural network solved many medical problems more
globally and yields the expected outcome. To overcome the local optimal
problem, this paper proposes a hybrid approach with the integration of
Cuckoo Search with K-means for detecting the cancer part and classifies
them using the DNN model. Many optimization techniques are applied to
medical images to yield accurate output; one among them is cuckoo search
optimization which imitates the swarm behavior and yields the global opti­
mization. Integrating FCSO with K-means6 for classifying the melanoma
as malignant and benign with 10 datasets utilized from online dataset for
validating the performance proves that this integration provides the result
faster than K-means and its limitation is computational time and number
of iterations. With the utilization of IDB2 and UCI dataset combination
of firefly and K-means7 is used for enhancing the exploitation and also
reduces the cluster distance measures. To detect the cancerous part nearby
unwanted noisy information are removed by applying morphological
based fuzzy threshold8 is applied for the enhancement and integration of
the K-means with the firefly optimization technique for detecting the skin
lesions with ISIC online dataset and proves its accuracy is superior than
PSO9 using K-means clustering for categorizing cancers and predicting the
cancer using total dermoscopic value (TDO) for various levels of cancer
and enhances the features using GLCM concept and proves its detection
and prediction is superior than various state-of-the-art techniques and its
Design Detecting and Classifying Melanoma Skin Cancer 245

limitation is that very few images are only used for validation. Feature
selection10 is done with the integration of BBA (Binary BAT Algorithm)
with threshold based approach for segmenting the cancer part and toclas­
sify them using Radial Basis Function Network (RBFN) and support vector
classifier (SVM) with 300 images and proved its specificity rate is far
better than CNN, KNN etc. DCNN methods11 are for classifying the tumor
part with three steps: in the first step, enhancing the color transformation
is done, in the second step lesion boundary is detected by using CNN,
and in the last step to extract the deep features and yield accurate output
DNN is used. Based on skin color12 lesions are extracted with K-means
and Multi-class Support Vector Machine for classification and obtained
96% classification accuracy with ISIR dataset.13 With the combination of
genetic algorithm with PSO enhanced K-means is used for segmenting the
skin lesions and back propagation neural network is used for classifica­
tion and yields 87.5% accuracy.14 Using hybrid techniques like dynamic
particle swarm optimization and K-means is performed for skin segmenta­
tion with high quality and it ensured that K-means is superior than K-mean
clustering.15 With 8 colors and 200×150 pixels the image is used as the
input image and with CNN skin cancers are segmented and analyzed its
performance with various existing start-of-the-art techniques.16 With the
combination of grab cut and K-means skin lesions are segmented and
obtained 0.8236 and 0.9139 dice and Jaccard coefficient.

17.2 MATERIALS AND METHODS

i. Image Description
For validating the efficiency and accuracy of skin cancer using the
hybrid approach, we collect various skin-related disease images
from online dataset like from IISC-DSI dataset. We utilized 320
cancer images and 100 normal images for this approach. Dermos­
copy or Epiluminescence Light Microscopy is an imaging method
which examines the skin lesion in detail; the working process of
dermoscopy is to place an oil immersion between sin and the optical
lens. Microscope is used to identify the pigmented surface in terms
of color, shape, and structure in detail for analyzing the lesions
intensively for prevention or for diagnosis. Flow diagram briefs our
hybrid approach and it is shown in Figure 17.1.
246 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 17.1 Flow diagram for CSO+K means.

ii. Median Filter for Preprocessing

Median filter is mainly used for enhancing the contrast based on
median pixel value, which aids to detect the boundary and lesion
parts in detail for further processing the image. The median filter
selects a random pixel from the input image and sorts the pixels in an
increasing order and then selects the middle value and replaces the
existing pixel value with the obtained median value.

f ( x, y ) = median ( s ( x, y ) ) (17.1)

where F(x,y) is the median filtered output and s(x,y) is the input
image for preprocessing. To remove the hair, air bubbles median
filtering is used and also aids in shaping and sharpening the edges of
images for extracting the affected part in accuracy.
Design Detecting and Classifying Melanoma Skin Cancer 247

iii. Cuckoo Search Optimization

Cuckoo search optimization algorithm was developed by Yang and
with three golden rules and levy distribution CS optimization is
created and applied in many medical image processing applications17
Cuckoo search optimization pseudo code is as follows:
Step 1: To define the objective function f(x)
Step 2: To generate the initial solution with ‘‘n” nests
Step 3: While stop criteria are met do
3.1 Get a random cuckoo and compute levy flight
3.2 Compute the fitness function
3.3 Randomly choose a nest j among ‘‘n”
3.4 Check if fi > fj then
Replace the current solution
End if
3.5 Worst nests portion pa will be eliminated and new nests to be
created and keep the best Solution
Step 4: End the while loop
iv. K-Means Clustering
K-Means is a clustering method which categorizes the classes into
k clusters for classification based on their characteristics. K-means
integrated with CS highly aids to detect the accurate part of skin
cancer. The following steps are used to brief the K-means clustering
method.
Step 1: Randomly choose number of cluster K
Step 2: Randomly choose K pixels from various intensities as
centroids
Step 3: Find the distance between two points and assign clusters
based on the minimum distance
Step 4: Compute the mean of each cluster and centroid by calculating
k clusters using the following formula:
∑ 1( C = j ) x
m i
i =2 i
(17.2)
∑ 1( C = j )
m
i =1 i

Step 5: Perform the above step repeatedly until all the pixels in each
cluster is converged
Cuckoo search optimization is integrated with K-means to yield global
optimal solution, K-means aids in finding the initial cluster and to
update the centroid CS is integrated for yielding best outcome of this
hybrid approach. Features set (12) are extracted from the segmented
248 Computational Imaging and Analytics in Biomedical Engineering

result and fed to SVM for classification with 70% training set and
30% testing images. Various features like energy, autocorrelation,
contrast, skewness, Kurtosis, Difference Entropy, Variance, Standard
deviation, Correlation, Homogeneity, dissimilarity are computed
for classification. The formulas for these features are mentioned in
Table 17.1:

TABLE 17.1 Features Extraction Formula Table.

Name of the feature Formula
Cluster prominence g −1 (17.3)
∑ h (i − f
i , j =0
ij i + j − fj) 3

Energy g −1 (17.4)
∑ − log ( h )
2
ij
ij =0

Autocorrelation g −1 (17.5)
∑ ( ij ) h ( ij )
ij =0

Contrast g −1 (17.6)
∑ − hlj ( i − j )
2

ij =0

Difference Entropy g −1 (17.7)

−∑hi + j ( i ) log ( hi + j ( i ) )
i =0

Skewness g −1 (17.8)

∑(oi − mean)3 h ( oi )
h=0

Kurtosis g −1 (17.9)

∑(o − mean) h ( oi )
h=0
i
4

Variance g −1 (17.10)

∑(o − mean) h ( oi )
h=0
i
2
Design Detecting and Classifying Melanoma Skin Cancer 249

TABLE 17.1 (Continued)

Name of the feature Formula

Standard deviation (17.11)
g −1

∑(o − mean) h ( oi )
h=0
i
2

Correlation (17.12)
g −1
( i − µ )( j − µ )
∑ hij
i , j =0 1+ ( i − j )
2

Homogeneity (17.13)
g −1
hij
∑ 1+
(i − j )
2
i , j =0

dissimilarity g (17.14)

∑ i − j h ( ij )
ij

v. SVM Classifier
An SVM classifier classifies the skin cancer into benign and malig­
nant based on their illumination shape and various features extracted
from the segmented image. The following steps brief the working
principle of the SVM classifier
Step 1: Collect the images from dataset and partition the data into
training (70%) and testing (30%).
Step 2: All features extracted from segmented images are classified
and label it.
Step 3: Compute support value and estimate it.
Step 4: Iterate the following steps until instance value is null.
Step 4.1: Check if support value is equal to similarity between each
instance, if so means, find the total error value.
Step 5: Check the instance if it is less than zero and then calculate the
FA where
Support _ value
FA = (17.15)
Total Error

Repeat step 5 until it becomes empty else stop the process.

In total 320 tumor and 100 normal images are collected from IISC-DSI
dataset for validating the proposed CSO + K means and the SVM classifier.
250 Computational Imaging and Analytics in Biomedical Engineering

The optimized features are taken from segmented output and partitioned
as 70% for training and 30% for testing ratio. The performance measures
ensured the efficiency of the proposed approach. Sensitivity, specificity, and
accuracy measure ensure both segmentation and classification accuracy and
proved the efficiency of the proposed approach.
Specivicity = 1− FPR *100 (17.16)

TP
Sensitivity = *100 (17.17)
TP + FN

TP + TN
Accuracy = *100 (17.18)
TP + FN + TN + FP
where TP represents the correctly detected part, FN represents the incorrectly
detected part, FPR represents the total number of wrongly detected parts, and
the classification comparison diagram is shown in Figure 17.2.

FIGURE 17.2 Comparison of classification accuracy.

From Figure 17.2 it was found that our SVM classifier yields good accu­
racy measures compared with existing approaches.17–19 From the segmenta­
tion result it was proved that the CSO with K-means is highly suitable for skin
cancer detection and its segmentation accuracy is compared with existing
approaches like FCM, K-means, adaptive K-means, and its comparison chart
is shown in Table 17.2.
Design Detecting and Classifying Melanoma Skin Cancer 251

TABLE 17.2 Segmentation Accuracy Comparison.

Techniques Segmentation accuracy (%) Computational time (S)
Fuzzy C-means 97 11
K Means 95.3 9
Adaptive K-means 967 8
Proposed CSO + K means 98.2 7

FIGURE 17.3 Proposed CSO + K means segmented results.

252 Computational Imaging and Analytics in Biomedical Engineering

Figure 17.3 represents the segmented result using CSO + K means

with median filters, where the first column represents the gray scale input
images, the second column represents the median filtered images, and the
third column represents the proposed output. From the integration of CSO +
K –means it was observed that based on cluster and number of iterations the
processing and accurate result will be modified. In our proposed approach
number of clusters is chosen as 2 and maximum number of iterations for
segmenting the tumor part is 90. It was observed that few hair and skin arti­
facts are detected in some benign images; to reduce that problem we need to
include morphological reconstruction filter20 for exact detection.

17.3 CONCLUSIONS

Integration of CSO with K-Means Clustering approach is suitable for

segmenting the skin cancers with 98.2% segmentation accuracy and with
SVM classifier it ensures 98.6% classification accuracy. Efficiency of
segmentation is proved with existing techniques like PSO with K-means,
Fuzzy C-Means, GA with K-Means, and ABC with K-Means. From the
computational time comparison it was proved that classification accuracy is
far superior to the existing techniques. In the future DNN models are used
for skin cancer classification.

KEYWORDS

• dermoscopy imaging
• median filter
• cuckoo search optimization
• K-means clustering
• performance measures

REFERENCES

1. İlkin, S.; Gençtürk, T. H.; Gülağız, F. K.; Özcan, H.; Altuncu, M. A.; Şahin, S. SVM:
Bacterial Colony Optimization Algorithm Based SVM for Malignant Melanoma
Detection. Eng. Sci. Technol. 2021, 24, 1059–1071.
Design Detecting and Classifying Melanoma Skin Cancer 253

2. Chitra, A.; Rajkumar, A. Paraphrase Extraction Using Fuzzy Hierarchical Clustering.

Appl. Soft Comput. 2015, 34, 426–437.
3. Lakshmi, S. P.; Karthikeyani, K.; Visalakshi, N.; Shanthi, S. Clustering Categorical
Data Using K-Modes Based on Cuckoo Search Optimization Algorithm. ICTACT J. Soft
Comput. 2015, 8, 1561–1566.
4. Esteva, A.; Kuprel, B.; Novoa, R. Dermatologist-Level Classification of Skin Cancer
with Deep Neural Networks. Nature 2017, 542, 115–118.
5. Sumathi, R.; Venkatesulu, M.; Arjunan, S. P. Segmenting and Classifying MRI
Multimodal Images Using Cuckoo Search Optimization and KNN Classifier. IETE J.
Res. 2021, 1–8.
6. Gaonkar, R.; Singh, K.; Prashanth, G. R.; Kuppili, V. Lesion Analysis Towards
Melanoma Detection Using Soft Computing Techniques. Clin. Epidemiol. Glob. Health
2020, 8, 501–508.
7. Girsang, A. S.; Yunanto, A.; Aslamiah, A. H. A Hybrid Cuckoo Search and K-Means for
Clustering Problem. Int. Conf. Electr. Eng. Comput. Sci. (ICECOS), 2017; pp. 120–124.
8. Xie, H.; Zhang, L.; Lim, C. P.; Yu, Y.; Liu, C.; Liu, H.; Walters, J. Improving K-Means
Clustering with Enhanced Firefly Algorithms. Appl. Soft Comput. 2019, 84, 105763.
9. Ali, Md. S.; Miah, Md. S.; Haque, J.; Rahman, Md. M.; Islam, Md. K. An Enhanced
Technique of Skin Cancer Classification Using Deep Convolutional Neural Network
with Transfer Learning Models. Mach. Learn. App. 2021, 5, 100036.
10. Dorj, U.-O.; Lee, K.-K.; Choi, J.-Y.; Lee, M. The Skin Cancer Classification Using
Deep Convolutional Neural Network. Multimedia Tools App. 2018, 77, 9909–9924.
11. Krishna Monika, M.; Arun Vignesh, N.; Usha Kumari, Ch.; Kumar, M. N. V. S. S.;
Laxmi Lydia, E. Skin Cancer Detection and Classification Using Machine Learning.
Mater. Today: Proc. 2020, 33, 4266–4270.
12. Aljawawdeh, A.; Imraiziq, E.; Aljawawdeh, A. Enhanced K-Mean Using Evolutionary
Algorithms for Melanoma Detection and Segmentation in Skin Images. Int. J. Adv.
Comput. Sci. App. 2017, 8 (12), 447–483.
13. Lam, Y. K.; Tsang, P. W. M.; Leung, C. S. PSO-Based K Means Clustering with
Enhanced Cluster Matching for Gene Expression Data. Neural Comput. App. 2013, 22,
1349–1355.
14. Li, H.; He, H.; Wen, Y. Dynamic Particle Swarm Optimization and K-Means Clustering
Algorithm for Image Segmentation. Optik Int. J. Light Electron Opt. 2015, 126,
4817–4822.
15. Setiawan, W.; Faisal, A.; Resfita, N. Effect of Image Downsizing and Color Reduction
on Skin Cancer Pre-screening. In 2020 International Seminar on Intelligent Technology
and Its Applications, 2020; pp 148–151.
16. Jaisakthi, S. M.; Mirunalini, P.; Aravindan, C. Automated Skin Lesion Segmentation of
Dermoscopic Images Using Grab Cut and K-Means Algorithms. IET Comput. Vis. 2018,
12, 1088–1095.
17. Hoshyar, A. N.; Al-Jumaily, A. Comparing the Performance of Various Filters on Skin
Cancer Images. Procedia Comput. Sci. 2014, 42, 32–37.
18. Mehta, A.; Parihar, A. S.; Mehta, N. Supervised Classification of Dermoscopic Images
Using Optimized Fuzzy Clustering Based Multi-Layer Feed-Forward Neural Network.
In 2015 International Conference on Computer, Communication and Control (IC4),
2015; pp 1–6.
254 Computational Imaging and Analytics in Biomedical Engineering

19. Sreelatha, T.; Subramanyam, M.; Prasad, M. G. Early Detection of Skin Cancer Using
Melanoma Segmentation Technique. J. Med. Syst. 2019, 43, 190–205.
20. Sumathi, R.; Arjunan, S. Towards Better Segmenting Low Depth of Filed Images Using
Morphological Reconstruction Filters. Int. J. Signal Syst. Eng. 2014, 7, 189–194.
CHAPTER 18

DETECTION OF LUNG CANCER

USING FUSION METHODS FOR CT
AND PET IMAGES
V. RAMESH BABU1, G. GUNASEKARAN1, and SREENITHI R.2
1
Department of Computer Science and Engineering, Dr. M. G. R.
Educational and Research Institute, Maduravoyal, Chennai, India
Department of Computer Technology, Madras Institute of Technology,
2

Chennai, India

ABSTRACT

Lung cancer is caused due to anomalous growth of cells that develop into
a tumor. Various researches report that the death rate of lung cancer is the
highest among all other types of cancer. In the first part of work segmenta­
tion of lungs, a tumor in CT image is used for Spatially Weighted Fuzzy C
Means Clustering (SWFCM) techniques. The overall accuracy, sensitivity,
and predictive values achieved are 86.082, 85.636, and 92.673%, respec­
tively. In the second part of the work segmentation of lungs, a tumor in PET
image is used for Spatially Weighted Fuzzy C Means Clustering (SWFCM)
techniques. The overall accuracy, sensitivity, and predictive values achieved
are 89.31, 87.27, and 95.88%, respectively. In the third part of the work, the
diagnosis is strengthened for mass screening; the CT and the PET images
are fused effectively. The four fusion methods namely Wavelet Transform,
Curvelet Transform, Non Subsample Contourlet Transform (NSCT), and
Multimodal image fusion are applied. The performance analysis Entropy,
Peak Signal Noise Ratio (PSNR), Standard Deviation (SD), Structural

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
256 Computational Imaging and Analytics in Biomedical Engineering

Similarity Index Measure (SIM), and Root Mean Square Error (RMSE) are
computed.

18.1 INTRODUCTION

The initial stage of cancer is when cells unusually grow out of control
anywhere in the body. The growth of cancer cell and normal cell differs.
The cancer cells do not die but instead they grow and form several
abnormal cells whereas normal cells die. Cancer cells can also possibly
invade into other organs, but normal cells cannot. The process of cells
abnormally growing without control and affecting other tissues leads to the
formation of a cancer cell. Utilization of anatomical priors in the process
of segmentation of PET lung tumor images was proposed by Ref. [1]. This
method combines both the anatomical and functional images and hence
provides effective tumor staging and improved treatment planning. PET/
CT detects tumor invasion into adjacent tissues successfully as well as
leads to precise localization of lesions, even though there are no morpho­
logical changes found in CT2 as presented in a PET and CT images based
automated lung nodule detection. It was an automatic method to detect
lung nodules in PET and CT. Here, the nodules that are in proximity region
and similar are combined into one by a split-up postprocessing step, and
the time of the localization can be minimized to a greater extent from more
than one hour to maximum five minutes. This method when executed and
authenticated on real clinical cases in Interview Fusion clinical evaluation
software (Mediso) proved successful in detecting lung nodules and may
be a valuable aid for physicians for the daily routine of oncology. Ref. [3]
presented a method of Statistical Texture Features analysis for Automatic
Lung Cancer Detection with the PET/CT images. The overall endurance
rate of lung cancer patients is observed to be only 14%.4 Early detections
increase the chance of choosing the apt treatment for the cancer patients.
Computational systems diagnoses are highly useful for radiologists in the
elucidation of images. Image preprocessing methods namely Contrast
Limited Adaptive Histogram Equalization (CLAHE) and Wiener filtering
have been performed to remove the artifact due to contrast variations
and noise. Haralick statistical texture features were chosen as they could
extract more texture information from the cancer regions than the visual
assessment. In order to classify the regions as normal or abnormal, Fuzzy
C-Means (FCM) clustering was used.5
Detection of Lung Cancer Using Fusion Methods for CT and PET Images 257

18.2 DETECTION OF LUNGS TUMOR USING CT IMAGES

Figure 18.1 shows the detection of lung tumor from CT image. The lungs’
tumor from CT image is segmented by using Spatially Weighted Fuzzy–C
Means Clustering.6

FIGURE 18.1 Flow chart: lung tumor detection from CT image.

18.2.1 PREPROCESSING REMOVAL OF BONE REGION

The first step is the removal of bone region from the lung CT image since
the bone region affects the segmentation accuracy. Toward this, R-plane,
G-plane, and B-plane are separated from the RGB image. In this entire plane,
the bone region is detected. By subtracting all these images, the resultant
image is obtained. Figure 18.2 shows R-plane, G-plane, and B-plane images.
Figure 18.3 shows input CT image.7

FIGURE 18.2 R-plane, G-plane, and B-plane.

258 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 18.3 Input lung CT image.

Figure 18.4 shows the enhanced and bone removal image. T is added
with difference between G-plane and B-plane as given in eq 3.1 to find the
affected area of the disease. S=T+[G-B]

FIGURE 18.4 Enhanced and bone-removed image.

Detection of Lung Cancer Using Fusion Methods for CT and PET Images 259

18.2.2 LUNG CANCER DETECTION USING SPATIALLY WEIGHTED

FUZZY C-MEAN ALGORITHM

The quality of the CT image will not be very good, so the SWFCM method
is adopted for the corresponding region.8
In SWFCM, to exploit the spatial information, a spatial function is
defined as eq 18.1
hij = ∑ k∈NB(x j ) uik (18.1)

where NB(xj) represents a square window centered on pixel xj in the spatial

domain. The spatial function is incorporated into membership function as eq
18.2
uijp hijq
uij′ = (18.2)
∑ ck =1 ukjp hkjq

where p and q are controlling parameters of both functions.

SWFCM—Spatially Weighted Fuzzy C-Means Clustering Algorithm
Step 1: Create the random number with the range from 0 to1 to be the initial
memberships. Let us consider the number of clusters as N and then calculate
Vi using (18.3)
∑ Nj =1 uijm x j
Vi = (18.3)
∑ Nj =1 uijm
where
Vi = ith cluster center, m = fuzziness parameter m = 2
where uij is by using eq 18.4
1
uij =
 x j − vi 
∑ ck =1  2
 x j − vk  (m −1) (18.4)
 
Step 2: Map uij into the pixel position and calculate the modified member­
ship uij using (3.19). Compute objective function J using eq 18.5

2
∑ Nj 1 =
J== ∑Ci 1 uijm x j − υi (18.5)

Step 3: Update the cluster center using (18.6)

260 Computational Imaging and Analytics in Biomedical Engineering

Step 4: Repeat step 2 to step 4 until the following termination criterion is

satisfied:

J new − J old < ε (18.6)

Figure 18.5 shows the SWFCM output image. The image with three
clusters is shown in Figure 18.5.

FIGURE 18.5 SWFCM image.

The segmented image has three clusters, namely the background region
of interest and small regions. The background and small regions are elimi­
nated and only large size of cluster is considered. Figure 18.6 shows tumor-
affected area in the CT image. Figure 18.7 shows the tumor-affected area in
identified cluster superimposed with the CT input image. Table 18.1 shows
the data sets of cancer-affected area of CT image.

FIGURE 18.6 Identified clustered image.

Detection of Lung Cancer Using Fusion Methods for CT and PET Images 261

FIGURE 18.7 Super imposed image.

TABLE 18.1 Data Sets of Cancer-Affected Area of CT Image.

Image T R M TP FP FN S (%) PV (%)

1 1102 1067 87.31 1011 56 90 91 94.57
2 1151 1104 91.92 1080 24 71 93.83 94.75
3 1053 1071 91.01 1012 59 41 96.11 94.49
4 1104 1062 86.89 1007 55 97 91.21 94.82
5 1162 1082 86.89 1007 55 97 91.29 98.24
6 5324 4864 88.81 4792 72 532 90.01 98.52
7 1151 1052 81.92 992 60 159 86.39 94.30
8 517 520 89.18 486 37 34 93.42 92.89
9 1322 1288 70.55 1340 149 183 86.23 88.51
10 1673 1518 89.94 1511 8 163 90.31 99.55
11 1214 1183 93.62 1159 25 56 95.38 97.98
12 2131 2769 76.24 2117 647 15 99.35 76.62
13 1175 1156 84.56 1062 90 108 90..90 92.48
14 1565 1372 86.19 1359 19 207 86.93 99.05
15 1974 1378 86.20 1365 984 138 93.08 73.33
T, ground truth region; R, region is proposed method; M, accuracy; TP, true positive, FP,
false positive; FN, false negative; S, sensitivity; PV, predictive value
262 Computational Imaging and Analytics in Biomedical Engineering

Figure 18.8 shows the Bar Graph of the above-obtained result.

FIGURE 18.8 Bar graph-comparison of lung cancer-affected area for the CT image.

18.3 DETECTION OF LUNGS TUMOR FROM PET IMAGES

The proposed method is presented in a flowchart as shown in Figure 18.9. In

this thesis Fuzzy Local Information C-means Clustering is used for segmen­
tation of lungs tumor from PET images.9,10. The size of the input PET images
is 512*512.

FIGURE 18.9 FLICM flowchart.

18.3.1 LUNG CANCER DETECTION USING FLICM METHOD

The FLICM is applied to the PET image. The major characteristic of FLICM
is that it guarantees noise immunity and preserves image detail and it is free
of any parameter selection.11
Detection of Lung Cancer Using Fusion Methods for CT and PET Images 263

FLICM incorporates local spatial and gray level information into its
objective function as defined in eqs 18.7 and 18.8
2
Jm =∑iN 1 =
∑ ck 1[U mki xi − vk + G ki ] (18.7)
=

where the ith pixel is the center of the local window, k is the reference cluster,
and the jth pixel belongs to the set of the neighbors falling into a window
around the ith pixel (Nj). di,j is the spatial Euclidean distance between pixels
i and j, Ukj is the degree of membership of the jth pixel in the kth cluster, m
is the weighting exponent on each fuzzy membership shown in eq 18.9, and
Vk is the prototype of the center of cluster k shown in eq 18.10
Here G =fuzzy factor
1
(1−U kj ) X j − Vk
m 2
Gki =
∑ j∈ N i (18.8)
 d +1
i≠ j ij

1
U ki = 1
 2  m−1
xi − vk + Gki 
∑Cj =1  (18.9)
 2 
x − vj + G ji 
 i 

∑iN=1 U kim xi
Vk = (18.10)
∑iN=1 U kim

The details of the algorithm are given in eq 18.7.

Algorithm: FLICM algorithm
Step 1. Set the cluster prototypes c, fuzzification parameter m,
and the ending condition .
Step 2. Set randomly the fuzzy partition matrix.
Step 3. Set the loop counter b = 0.
Step 4. Determine the cluster prototypes using 20
Step 5. Calculate membership values using 19
Step 6. If max {U (b)−U (b+1)}< ɛ then stop, otherwise
set b = b + 1 and go to step 4.
When the algorithm has converged, a defuzzification process takes place
to convert the fuzzy partition matrix U to a crisp partition FLICM.
Figure 18.10 shows the result of FLICM clustering output. From the
clusters, the tumor is identified by selecting the cluster index at the location
of the darkest point in the PET image because the tumor belongs to lower
264 Computational Imaging and Analytics in Biomedical Engineering

intensity regions. Figure 18.10 shows the segmented tumor region from the
FLICM output. From the FLICM output corner cluster is removed, the zero
pixels are tumor regions. Figure 18.11 shows the segmented tumor region
from the FLICM output. Table 18.2 illustrate data sets of cancer-affected
area of the PET image.

FIGURE 18.10 The result of FLICM with white background.

FIGURE 18.11 Detection of tumor in PET image.

Detection of Lung Cancer Using Fusion Methods for CT and PET Images 265

TABLE 18.2 Data Sets of Cancer-Affected Area of PET Image.

Image T R M% TP FP FN S (%) Pv (%)
1 1152 1092 92.10 1101 59 95 91.52 95.56
2 1110 1108 94..32 1086 29 82 94.95 95.85
3 1192 1065 94.62 1020 59 41 96.01 94.32
4 1164 1083 95.52 1007 55 97 91.55 96.85
5 1180 1170 97.32 1069 20 103 92.48 98.24
6 5355 5325 98.52 4856 79 562 93.35 98.52
7 1168 1142 96.25 900 70 168 90.37 94.3
8 560 530 96.23 486 40 56 93..86 93.89
9 1352 1340 97.23 1131 150 192 92.42 93.56
10 1690 1562 95.23 1510 15 172 95.36 98.54
11 1252 1199 96.32 1170 34 65 95.37 97.21
12 2146 2956 97.01 2119 676 45 99.34 96.52
13 1196 1175 94.55 1098 102 108 91.23 93.58
14 1582 1499 92.16 1468 25 252 92.13 99.12
15 2001 1965 96.54 1954 978 152 93.01 92.18
T, ground truth region; R, region is proposed method; M, accuracy; TP, true positive; FP, false
positive; FN, false negative; S, sensitivity; PV, predictive value

Figure 18.12 shows the Bar Graph of the above obtained result.

FIGURE 18.12 Bar graph: comparison of lung cancer-affected area for PET image.
266 Computational Imaging and Analytics in Biomedical Engineering

18.4 IMAGE FUSION OF CT AND PET IMAGES

Image fusion is defined as the process of combining the relevant information

from a set of images into a composite image, where the resultant fused image
will be more informative and complete than any of the input images. Image
fusion techniques can improve the quality and increase the application of
these data. Image fusion finds applications in the area of navigation guid­
ance, object detection and recognition, satellite imaging for remote sensing
and civilian etc. In this paper four model works are proposed: the wavelet
curvelet, NSCT and multi-model image method, and two fusion rules namely
PCA and maximum absolute rule.12–21

18.5 RESULT AND DISCUSSION

The first column in Figure 18.13 shows CT images, the second column
shows the PET images, the third column shows the fused image for wavelet
transform based maximum absolute fusion rule, and the fourth column shows
the PCA fusion rule

FIGURE 18.13 The result for fused image in wavelet transform.

The first column in Figure 18.14 shows CT images, the second column
shows the PET images, the third column shows the fused image for Fast
Discrete Curvelet Transform based maximum absolute fusion rule, and the
fourth column shows the PCA fusion rule.
Detection of Lung Cancer Using Fusion Methods for CT and PET Images 267

FIGURE 18.14 The result for fused image in curvelet transform.

The first column in Figure 18.15 shows CT images, the second column
shows the PET images, the third column shows the fused image for Non
Sub-sampling Contourlet Transform (NSCT) based maximum absolute
fusion rule, and the fourth column shows the PCA fusion rule.

FIGURE 18.15 The result for fused image in NSCT.

The first column in Figure 18.16 shows CT images, the second column
shows the PET images, the third column shows the fused image for multi-
modal based maximum absolute fusion rule, and the fourth column shows
the PCA fusion rule.
268 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 18.16 The result for fused image in multi-model image fusion based.

Evaluating the performance of the fusion algorithm can be ascertained

effectively via Image Quality Assessment (IQA) of the fused image (Sudeb
Das and Malay Kumar Kundu, 2013). Image fidelity metrics based on
error estimation, that is, peak signal-to-noise ratio (PSNR) is commonly
employed for objective evaluation of fused image quality. Other than this,
fusion metrics based on entropy (E), standard deviation (SD), and root mean
square error (RMSE) account for the restored information content in the
fused image. Table 18.3 shows the performance analysis of different fusion
methods. Figures 18.17, 18.18, 18.19, 18.20, and 18.21 show bar charts for
various parameters. The graph indicates that the proposed method is better
than other methods.

TABLE 18.3 Performance Analysis for Different Fusion Methods.

Transform Fusion rule Entropy PSNR SD Similarity RMSE
index
measure
Wavelet Maximum absolute 5.832 25.96 34.7321 0.421 3.256
PCA 5.956 26.72 36.888 0.432 3.168
Curvelet Maximum absolute 6.328 28.47 54.1765 0.521 2.527
PCA 6.786 29.56 56.192 0.556 2.496
NSCT Maximum absolute 6.543 22.78 55.682 0.435 2.478
PCA 7.498 23.16 56.182 0.479 1.986
Multi Model Maximum absolute 7.568 49.58 60.282 0.623 0.978
Image Fusion and PCA
Detection of Lung Cancer Using Fusion Methods for CT and PET Images 269

FIGURE 18.17 Performance analysis of entropy.

FIGURE 18.18 Performance analysis of PSNR.

270 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 18.19 Performance analysis of SD.

FIGURE 18.20 Performance analysis of SIM.

Detection of Lung Cancer Using Fusion Methods for CT and PET Images 271

FIGURE 18.21 Performance analysis of RMSE.

18.6 CONCLUSIONS

In this work, first phase lung cancer is detected by CT images by using

SWFCM clustering. In second phase lung cancer is detected by PET image
by using FLICM algorithm. In third part of work image fusion is done by
CT and PET images. The values of entropy, PSNR, SD, SIM, and RMSE for
wavelet transform under fusion rule of maximum absolutes are 5.832, 25.96,
34.7321,0.421, and 3.256 and principal component analyses are 5.956,
26.72, 36.888, 0.432, and 3.168, respectively. The values of entropy, PSNR,
SD, SIM, and RMSE for curvelet transform under fusion rule of maximum
absolutes are 6.328, 28.47, 54.1765, 0.521, and 2.527 and principal compo­
nent analyses are 6.786, 29.56, 56.192, 0.556, and 2.496, respectively. The
values of entropy, PSNR, SD, SIM, and RMSE for NSCT under fusion rule of
maximum absolutes are 6.543, 22.78, 55.682, 0.435, and 2.478 and principal
component analyses are 7.498, 23.16, 56.182, 0.479, and 1.986, respectively.
The values of entropy, PSNR, SD, SIM, and RMSE for multi-model image
272 Computational Imaging and Analytics in Biomedical Engineering

fusion method under fusion rule of maximum absolute and PCA are 7.568,
49.58, 60.282, 0.623, and 0.978, respectively. Higher resolution of CT and
PET image may be used for future work to increase the level of accuracy for
the result.

KEYWORDS

• CT
• PET
• SWFCM
• FLICM
• RMSE
• PSNR

REFERENCES

1. Aggarwal, P. et al. Sardana, Semantic and Content-Based Medical Image Retrieval for
Lung Cancer Diagnosis with the Inclusion of Expert Knowledge and Proven Pathology.
In Proceedings of the IEEE Second International Conference on Image Information
Processing ICIIP'2013, 2013; pp 346–351.
2. Zsoter, N.; Bandi, P.; Szabo, G.; Toth, Z.; Bundschuh, R. A.; Dinges, J.; Papp, L. PET-CT
Based Automated Lung Nodule Detection. In Annual International Conference of the
IEEE Engineering in Medicine and Biology, 2012; pp 4974–4977.
3. Punithavathy, K.; Ramya, M. M.; Poobal, S. Analysis of Statistical Texture Features for
Automatic Lung Cancer Detection in PET/CT Images. In International Conference on
Robotics, Automation, Embedded & Control RACE 2015, 2015.
4. Akram, S.; Nicolini, S.; Carolia, P.; Nannia, C.; Massarob, A.; Marzolab, M.-C.;
Rubellob, D.; Fantia, S. PET /CT Imaging in Different Types of Lung Cancer: An
Overview. Eur. J. Radiol. 2013, 81, 988–1001.
5. Akram, S.; Javed, M.-Y.; Qamar, U.; Khanum, A.; Hassan, A. Artificial Neural Network
based Classification of Lungs Nodule Using Hybrid Features from Computerized
Topographic Images. Appl. Math. Inf. Sci. 2015, 9 (1), 183–195.
6. Sharma, D.; Jindal, G. Identifying Lung Cancer Using Image Processing Techniques.
Int. Conf. Comput. Techniq. Artif. Intell. (ICCTAI’2011) 2011, 17, 872–880.
7. Tong, J.; Da-Zhe, Z.; Ying, W.; Xin-Hua, Z. Xu, W. Computer-Aided Lung Nodule
Detection Based on CT Images. In IEEE/ICME International Conference on Complex
Medical Engineering, 2007; pp 816–819.
8. Ganeshbabu, T. R. Segmentation of Optic Nerve Head for Glaucoma Detection using
Fundus images. Biomed. Pharmacol. J. 2014, 7 (2), 1–9.
Detection of Lung Cancer Using Fusion Methods for CT and PET Images 273

9. Kanakatte, A.; Mani, N.; Srinivasan, B.; Gubbi, J. Pulmonary Tumor Volume Detection
from Positron Emission Tomography Images. Int. Conf. Bio Med. Eng. Inf. 2008, 2 (8),
213-217.
10. Kumar, A.; Kim, J.; Wen, L.; Dagan, D.; Feng, F. A Graph-Based Approach to the
Retrieval of Volumetric PET-CT Images. In International Conference of the IEEE
Engineering in Medicine and Biology, 2012; pp 5408–5411.
11. Ganesh Nabu, T. R.; Tamil Thendral, M.; Vidhya, K. Detection of Lung Cancer
Tumorusing Fuzzy Local Information C- Means Clustering. Int. J. Pure Appl. Math.
2018, 118 (17), 389–400.
12. Ahmed, H.; Hassan, E. N.; Badr, A. A. Medical Image Fusion Algorithm Based on Local
Average Energy-Motivated PCNN in NSCT Domain. Int. J. Adv. Comput. Sci. App.
2016, 7 (10), 269–276.
13. Patel, K. Fusion Algorithms for Images Based on Principal Component Analysis and
Discrete Wavelet Transform. Int. J. Innov. Res. Sci. Technol. 2015, 1, 180–182.
14. Nahvi, N.; Sharma, O. C. Comparative Analysis of Various Image Fusion Techniques
for Biomedical Images: A Review. Int. J. Eng. Res. App. 2015, 4 (5), 81–86.
15. Rajesh, K.; Ravichandran, C. G. Curvelet Transform Based Image Fusion With Noise
Reduction Using Gaussian Filter. In Australian J. Basic Appl. Sci. 2015, 9 (35), 161–166.
16. Rajalingam, B.; Priya, R. A Novel Approach for Multimodal Medical Image Fusion
using Hybrid Fusion Algorithms for Disease Analysis. Int. J. Pure Appl. Math. 2017,
117 (15), 599–619.
17. Wang, W.; Chang, F. A Multi-Focus Image Fusion Method Based on Laplacian Pyramid.
J. Comput. 2011, 6 (12), 2559–2566.
18. Sun, Y.; Zhao, C.; Jiang, L. A New Image Fusion Algorithm Based on Wavelet Transform
and the Second Generation Curvelet Transform. In IEEE International Conference on
Image Analysis and Signal Processing, 2010; pp. 438–441.
19. Fei, Y.; Wei, G.; Zongxi, S. Medical Image Fusion Based on Feature Extraction and
Sparse Representation. Int. J. Biomed. Imaging 2017.
20. Yang, Y.; Huang, S.; Gao, J.; Qian, Z. Multi-Focus Image Fusion Using an Effective
Discrete Wavelet Transform Based Algorithm. Measurement Sci. Rev. 2014, 14 (2),
102–108.
21. Liu, Z.; Chai, Y.; Yin, H.; Zhou, J.; Zhu, Z. A Novel Multi-Focus Image Fusion Approach
Based on Image Decomposition. Inf. Fusion 2017, 35, 102–116.
CHAPTER 19

A FRAMEWORK PROMOTING
POSITION TRUST EVALUATION
SYSTEM IN CLOUD ENVIRONMENT
S. R. SRIDHAR, S. PRAGADEESWARAN, and M. GANTHIMATHI
Department of CSE, Muthayammal Engineering College, Namakkal,
Tamil Nadu, India

ABSTRACT

Trust management is one of the biggest obstacles to the adoption and growth
of cloud computing. Both consumers and corporations commonly use cloud
computing. Protecting the customer’s privacy would not be a simple task due
to the sensitive data contained in the interaction between the customer and
the trust evaluation service. It might be challenging to defend cloud services
against malicious users. The design and development of CloudArmor, a
legacy trust evaluation framework that offers a set of functions to provide
Trust as a Service, will be discussed in this chapter (TaaS). This chapter also
discusses the problems with calculating trust based on input from cloud users.
This technology effectively protects cloud services by detecting hostile and
inappropriate behavior by the use of trust algorithms, which may recognize
on/off assaults and colluding attacks by using different security criteria. In
conclusion, the findings demonstrate that the suggested trust model system
may deliver high security by lowering security risk and enhancing the
decision-making capabilities of cloud users and cloud operators.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
276 Computational Imaging and Analytics in Biomedical Engineering

19.1 INTRODUCTION

Two major players have emerged in the new computing method known as
cloud computing. cloud-end consumers and cloud service providers. There
are multiple definitions given forth by various authors to precisely clarify
what cloud computing is. In the realm of computers, cloud computing is a
new business model. According to NIST’s official definition, “public cloud
is a platform which enables widespread, comfortable, on-demand access
to a common cloud computing model” (i.e., connections, servers, storage,
applications, and services) that can be quickly provisioned and released with
little interaction from service providers.
Cloud computing adoption raises privacy problems. Customers and cloud
service providers may engage in dynamic exchanges that include sensitive
data. There have been a number of instances of privacy violations, including
the disclosure of behavioral data or sensitive information (such a person’s
address and birth date). Services that use customer information (such as
interaction history) should unquestionably protect that information.
Collusion attempts become a threat whenever an individual or group of
evil elements tries to undermine the system. A reputational system’s cred­
ibility is typically put at greater risk when several evil actors work together
than when they act maliciously alone. Here are a few instances of particular
cooperation attacks.
Collusive slandering attacks, also known as corrupt business badmouthing
attacks, take place when dishonest users team up to disseminate negative
testimonials about a reliable person in an effort to badly damage that user’s
reputation. They also hope to boost their own reputations by complimenting
one another.
Several industries, including e-commerce, human sociology, wireless
systems, and others employ trust management extensively. Finding a service
provider you can trust in the cloud environment requires careful consider­
ation of your level of trust. When evaluating a service provider’s reliability,
cloud service user evaluations play a big role. This study looks at a number
of risks that may arise when trust is determined by user input from the cloud.
The link among cloud user’s services is offered by a trust management
system (TES) for efficient trust management. However, because to the unpre­
dictability of user numbers and the extremely dynamic character of the cloud
environment, ensuring TES uptime is a challenging task. It is unsuitable for
cloud systems to evaluate user preferences and capabilities using indicators
A Framework Promoting Position Trust Evaluation 277

of success or functional durability measures. TES must be extremely scal­

able and adaptive in order to function in cloud settings.

19.2 TRUST EVALUATION

19.2.1 EXISTING SYSTEM

One of the top 10 related to the implementation of cloud computing,

according to Berkeley researchers, is trust and security. Service-Level Agree­
ments do exist (SLAs). Feedback from customers is a reliable source for
determining how trustworthy cloud services are overall. Several researchers
have acknowledged the need of managing trust and have put forth strategies
to gauge and regulate effect on participant feedback.
Two distinct types of trust exist. (1) Indirect Trust and (2) Direct Trust.
Indirect trust states that when a person lacks any direct experience, he must
rely on the direct trust of others. Direct trust is approach on personal expe­
rience. Indirect trust is the name for this kind of trust. There are several
service providers available in a cloud environment. Finding a reliable service
provider is so crucial. Since trust is more irrational, dependent on context,
asymmetric, and uncertain than other cloud entities, it is a complex interac­
tion between them.
Given that cloud solutions are very dynamic, dispersed, and opaque,
maintaining confidence in public clouds is a significant challenge. Feedback
from customers is a reliable source for determining how trustworthy cloud
services are overall. Several researchers have acknowledged the value of
managing trust and have put forth strategies to gauge and maintain trust situ­
ated on participant feedback. In practice, it is not unusual for a cloud service
to encounter malevolent user actions (such as collusion or Sybil assaults).
There are several methods for measuring trust. By reviewing user reviews
posted to the cloud, we may assess the trustworthiness in the e-commerce
sector. There is a risk for different evaluation base attacks when trust is
determined via cloud user feedback ratings. The potential attacks on the
feedback-based trust assessment are the focus of the next section. Users of
cloud solutions may be handling remarkably confidential material during
using trust maintenance services, so it is imperative that the highest level of
security be offered to them; obtaining cloud services by efficiently identi­
fying fraudulent and unsuitable behaviors by the use of trust methodologies
that can recognize on/off and collusion threats.
278 Computational Imaging and Analytics in Biomedical Engineering

FIGURE 19.1 System architecture.

19.2.2 PROPOSED SYSTEM

The opinions of cloud service consumers are a reliable source for deter­
mining how trustworthy cloud services are overall. As demonstrated in
Figure 19.1, unique methodologies are introduced in this paper that assist
in identifying reputation-based assaults and enable users to quickly select
reliable cloud services. It presents a version of success that not only
recognizes deceptive trust feedbacks arising from differential attack but
also recognizes Sybil attacks whether they occur over a lengthy or brief
period of time, also provides a model of availability that keeps the trust
comprehensive suite operating at the desired level; it also provides a model
of availability that keeps the trust management solutions operating at the
desired level.
A model of credibility. The effectiveness of the trust management is
significantly influenced by the veracity of feedback. Therefore, suggest a
number of metrics, such as Output Volume and Frequent Feedback Colluding,
for the detection of feedback collusion. These measurements separate decep­
tive user feedback from malevolent user behavior.
A Framework Promoting Position Trust Evaluation 279

Benefits of the Proposed System

• TES will validate user comments without knowing the user’s
credentials.
• Processes credentials without storing sensitive information.
• Uses anonymized data and consistently hashes it to find Sybil assaults,
whether they last for a long time or just a short while.

19.3 DESIGN METHODOLOGIES

19.3.1 TRUST EVALUATION SYSTEM (TES)

The trust model has additional layers added to it to improve the system’s
overall efficacy. The following section provides descriptions of the various
TES subsections.
The central repository serves as the place where interactions are stored.
For later use by the deciding engine trust for determining the values of the
tasks and roles, it maintains all types of trust data and interaction histories
produced by interaction skills and duties. Elements not found in the TES are
inaccessible to the central repository.
• Role Behavior Analyzer: This part examines how the simplest levels
of trust laws apply to shared resources, including roles and functions.
Based on the feedback provided by the service in the central reposi­
tory, it assesses those regulations that have been recognized in the
level of trust. The role behavior checker links the roles in order to
learn more about them and find any leaks that might exist. In order
to easily follow unauthorized users or attackers and provide proof of
any type of data loss, it is crucial to identify the user and keep track of
every actions they do.
• Task Behavior Analyzer: The task behavior analyzer is in charge
of assessing tasks and functions in light of minimal trust level laws
while gaining access to shared resources. By computing the trust
value and storing it in the central repository, the tasks indicated only
within confidence level are examined in terms of such feedback of
owners. When determining the histories of users with relation to the
stored data, it gathers data from the channels, though there are two
in this case reports from development comes information leakage
and data from the role behavior analyzer. The task behavior analyzer
can help identify customers, and it’s important to keep track of the
280 Computational Imaging and Analytics in Biomedical Engineering

activities completed. This makes it simpler to locate attackers or

unwelcome clients and to provide evidence of data leakage, should
it have happened. Updated client accounts that have been registered
are checked to see if a customer account was responsible for the
occurrence.
• Feedback Collector: Before being automatically assigned, feedback
from service owners to a depository headquarters must be managed
by the feedback collector. The feedback provided on duties and tasks
reveals the user’s credibility. Its integrity is protected by the job and
role feedback collector integrity to guarantee security. This element
makes sure that the person submitting feedback further into system is
authorized. It is able to identify false feedback and remove it off the
system. Prior to uploading the data to the central repository, the role
feedback collector also gathers information regarding the job and role
assignments.
• Trust Decision Engine: In this section, the trustworthiness of the cloud
users, their responsibilities, and the entity’s values are examined and
identified. Before deciding what sort of answer the system should
provide, it gathers all types of data about the contact records that are
contained in the repository center and the respect values of a given
consumer.
• Detection of service: This layer is made up of many cloud service
consumers. For instance, a young company with meager funding can
use cloud services. This layer’s interactions consist of: (1) service
discovery, which allows customers to hunt up more cloud services and
other services online. (2) Trust and counter, where users can provide
feedback or access a cloud service’s trust score; and (3) registration,
where users confirm their identity by logging in to IdM before using
TES.
• IDM Registration: The system suggests utilizing the Identity Manage­
ment Service (IdM) to assist TES in determining the validity of a
customer review. Processing the IdM data, however, has the potential
to violate user privacy. Using cryptographic encryption techniques
is one way to protect privacy. However, processing encrypted data
cannot be done effectively. Another option to process IDM data
without violating users’ privacy is by using anonymization techniques.
It is obvious that utility and great anonymity are trade-offs.
• Service Announcement and Communication: This layer comprises
of several cloud service providers who openly provide IaaS
A Framework Promoting Position Trust Evaluation 281

(Infrastructure services), PaaS (Platform as a Service), and SaaS

(Software as a Service) cloud services, as well as combinations of
these services, over the Internet Web portals provide access to these
cloud services, which are also listed on search engines like Google,
Yahoo, and Baidu. This layer’s interactions with users and TES are
regarded as cloud service interactions.

FIGURE 19.2 Trust model system architecture.

19.4 FEEDBACK BASE TRUST EVALUATION

A. Feedback Storage and Evaluation

In current methodologies, a service’s trustworthiness is largely
evaluated centrally, whereas feedback is provided by dispersed trust
participants. The centralization of power may cause scalability and
security problems. On a customized virtualization that protects the
processor architecture, a suggested framework for safe application
execution is based suggested that trust paradigm for private cloud
environment data security proposed a framework for managing trust
282 Computational Imaging and Analytics in Biomedical Engineering

that uses credibility and feedback to determine the value of trust. The
percentage of recognized consensus feedback can help determine the
veracity of other feedback.
B. Accuracy of Trust Result
Because the cloud environment is changing, selecting the reliable
feedbacks from among the many supplied feedbacks is the most
difficult problem. Accuracy of Trust and Trustworthiness are signifi­
cantly correlated with others. By minimizing potential threats, we
can obtain an accurate trust result.
C. Collusion Attacks
When several users band together to submit false feedback in an
effort to boost or lower the service provider’s trust score, the attack
takes place. This conduct is referred to as tacit collusion malicious
feedback behavior. Collusion assaults of three different types are
feasible.
a) A promotion-focused attack
Entire group Enter all encouraging comments to help the cloud
service provider grow.
b) Slanderous Assault
The entire group entered all critical comments for the cloud
service provider.
Occasional Feedback Collusion.
c) Oblique Feedback Collusion
Cloud services occasionally experience collusion assaults. Time
is a crucial factor in recognizing sporadic collusion attacks.
Irissappane suggested a clustering strategy that would differ­
entiate submitted feedback from malicious feedback presented
clustering approach to increase the performance of the trust
system. This approach creates clusters based on the variances
in all the ratings. The value of trust is increased by combining
reliable weighted ratings. We can calculate the irregular change
in submitting all feedback loops within the overall feedback
behavior to identify Occasional Feedback Collusion Sybil
Attacks.
d) Such an attack takes place when malicious people leave several
false reviews while using different identities (i.e., producing
numerous false ratings while making a limited number of
different purchases in a brief period of time) in an effort to boost
or lower the trust rating.
A Framework Promoting Position Trust Evaluation 283

e) These remarks are made for many different reasons, including

user ineptitude, deception, or carelessness, among others.
Attackers might assume several different identities to cover up
their previous bad reputations. Sybil can attack in one of three
ways. Attack That Promotes Itself Users provide encouraging
comments to promote cloud service providers. This assault is
sometimes referred to as a ballot-stuffing assault.
f) A Slandering Attack
Users can promote a cloud service provider by leaving unfavor­
able reviews. This attack is also known as a verbal assault.
g) Occasional Sybil Attacks
By making numerous accounts and submitting false comments
in a short amount of time, malicious users can influence the
outcome in terms of trust. By employing previously saved
identity records for strong identity authentication, we might
be able to prevent the Sybil attack. By calculating the sporadic
variations in all submitted IDs and identity behavior, it can be
detected.
D. Intoxication Attacks
In this strategy, a member first acts appropriately for a set amount of
time to build up a high reputation before beginning to act inappropri­
ately. Due to their high reputation, intoxication makes it challenging
for the system to locate such misbehaving users.
On-off attack is another name for this assault. A member first exhibits
positive behavior, but as time goes on, the user starts to behave badly.
Such members are challenging to spot since they have historically
maintained a positive reputation. This attack is referred to as the hold
a positive attitude of peers in P2P network systems.
The forgetting factor is one of the most common tactics used to stop
this onslaught. However, malevolent people frequently took advan­
tage of forgetting. Forgetting factor, put out by Sun and Liu states
that when trust results go below.
E. Discrimination Attacks
This attack occurs when a service provider offers customers better
services in some geographical regions while offering subpar services
in others. These guaranteed conflicting feedbacks between these
regions, which may have resulted in a coordinated attack. No defense
against such an attack has been discovered.
284 Computational Imaging and Analytics in Biomedical Engineering

F. Newcomer Attacks
If a person can readily register under a new identity even when the
member already has registered with the service provider and displayed
some unsavory behavior in the past, the member can launch a newbie
assault. Re-entry attack is another name for this assault. It succeeded
in Sybil’s attack in the end. By matching the credential recodes with
various parameters, such as location and unique id, we can lessen
newbie attacks.

The system is launched by the administrator, and defines the hierarchy of

roles and responsibilities for it. Channel 1 facilitates uploading to the cloud
of the roles and task parameters established by the system;
Customers must first make an access permission via Port 2 depending on
their responsibilities and roles if they want access privileges to cloud data.
The role entity transfers the demand to the task's entity via Channel 5 if the
customer request is approved. In reply, the cloud offers users an assignment
access control (T-RBAC) scheme that is encrypted. The owner can only
authorize encryption and data uploading through Channel 3 if they believe
the job or task can be respected. The proprietor also discloses Owners should
contact the feedback collectors through Channel 14 if they discover that one
of their customers has leaked their data. To preserve each trust assessment
and interaction record created whenever the roles interact, the input collector
sends new feedback from an approved owner to the central database using
Channel 11.
Then, utilizing Channel 10, the confidence decision engine uses the
interaction logs to calculate the trust level of roles and tasks; the database is
then centrally located.
The roles’ entity may at any time approve the TMS's confidence assess­
ments for the roles, at which point it responds to the TMS through Channel
13 to get feedback from the TMS. Until the input from the trusted decision
engine is received, the role entity evaluates the role parameters that influence
a consumer’s cloud role membership, and any harmful consumer member­
ship is removed; When a data breach results in negative feedback from an
owner regarding a role, the role entity sends information about the leak to
the role behavior the responsibilities entity may at any moment accept the
TMS’s confidence evaluations for the tasks through Channel 4, at which
point it responds to the TMS through Channel 12 to get input from of the
trusted decision engine.
A Framework Promoting Position Trust Evaluation 285

Until input from the confidence decision engine is received, the task
entity analyses the task’s parameters that determine a consumer's cloud task
membership, and any harmful consumer membership is removed.
When a task’s owner complains about it because of information leaks, the
task entity sends Channel 7 information about the leak to the task behavior
analyzer. Then, the analyzers utilize Channels 6 and 8 to continuously update the
trust information again for roles and responsibilities in the centralized database;
The TMS does a trust analysis whenever an owner desires that his data be
transferred and protected in the cloud. Following receipt of the request, the
TMS contacts the proprietors through Channel 9; the trusted decision engine
informs the owners of the results of the trust management for their respective
responsibilities. Based on the findings, the data owners decide whether or not
to grant customers access to their services.

19.5 SIMULATION RESULTS

In C#.net, we created a Windows Forms application.net to test the outcomes

of our system, and we used big data to verify all the precautions taken to
prevent these attacks. To ascertain the trust model's resilience to reputational
threats, experiments were used. The TMS predicated the penalty for on attack
criteria: whether the connection significance was greater than or equivalent
to the risk rate (DR), or whether the recommender’s feedback (F) was less
acceptable than interface importance (II). The trust model computed the
confidence decline penalty PTD when the recommender’s feedback (F) was
less significant than the importance of the interaction (II). The contact trust
values are affected by the cost of the on attack and trust deterioration.

FIGURE 19.3 Penalties for on/off attack and declining confidence.

286 Computational Imaging and Analytics in Biomedical Engineering

The trust model uses the penalties for bad behavior to establish the worth
of contact trust for malicious users. Figures 19.4 and 19.5 show how fresh
feedback affects interactions between trustworthy users.

FIGURE 19.4 Values of interaction trust for 100 consumers.

FIGURE 19.5 Trust scores for interactions with 100 malevolent customers.
A Framework Promoting Position Trust Evaluation 287

The resilience of the trust model against reputational attacks is investi­

gated in this section through experimentation. To stop collusion attacks, the
TMS determined the cooperation attack frequency (CAF). Here, feedback
frequency had an inverse connection with credibility and a direct link with
collusion in the feedback. The amount of recommendation systems feedback
items and the overall number of input pieces within the suspect set were used
to calculate the feedback frequency (SS).
Seven questionable recommenders’ feedback frequency is shown
Figure 19.6 shows how, as we presume, in Table 19.1. The fact that five
suspected recommenders’ feedback frequency exceeded the feedback limit
(FL) suggests that the TMS will move the feedback they provided to the
collusion set (CS) or that the trust. The suspected feedback will be moved
by the model. provided by a specific recommender for a specific user to the
feedback set (FS).

FIGURE 19.6 Collusion attack frequency.

TABLE 19.1 Collusion Attack Frequency.

FN(SRi, CR) 7 20 32 13 36 1 17
FN(SS, CR) 126 126 126 126 126 126 126
CAF(SRi, CR) 0.06 0.18 0.24 0.12 0.3 0.02 0.13
288 Computational Imaging and Analytics in Biomedical Engineering

Figure 19.7 demonstrates how the attacker scale (AS), which is the size
of the attack scale for distinct collusion sets, is determined using the trust
model. To assault and undermine the trust model, the harmful recommenders
in a recommender’s community must make up a sizeable portion of all
recommenders. This is known as the collusion set (CS).

FIGURE 19.7 Attack scale (AS).

Figure 19.8 gives the result of the attack target scale (ATS), which
provides the destructive feedback rate from one collusion set (CS) for a
certain user.

FIGURE 19.8 Attack target scale (ATS).

A Framework Promoting Position Trust Evaluation 289

Figure 19.9 gives the result of a collusion attack strength (CAS), which
is calculated by calculating the rate of all harmful feedback from various
collusion sets (CS) for a certain user.

FIGURE 19.9 Collusion attack strength (CAS).

19.5.1 COMPARISON OF SECURITY AND ACCURACY

A number of assaults can be launched against any kind of reputation manage­

ment service. These assaults have the power to either improve or harm a
particular entity’s reputation. We concentrate on applying different measures
to stop these attacks in order to develop a stable, dependable, and correct trust
model framework, which helps us to develop an efficient and reliable trust
model system. Table 19.2 compares our suggested TMS to those mentioned
in similar studies.

TABLE 19.2 Difference of Accuracy and Security.

Addressed metrics [1] [2] [3] [4] [5] Ours
Collaboration importance  
On/off the attack    
Trust failure   
Collusion outbreak     
Conspiracy attack frequency  
290 Computational Imaging and Analytics in Biomedical Engineering

Our extensive research project, Cloud Armor, which gives a framework

for prestige trust administration of cloud services, includes the deployment
of the trust management service. Users of the platform can submit sugges­
tions and ask for a trust assessment of a specific cloud service in a safe
environment provided by the platform. Specifically, the management of trust
The Trusted Data Provision and the Trust Evaluation Function are the two
main parts of the Trust Management Service (TMS).
The Trusted Data Provisioning is in charge of gathering trust data and
cloud service information. To enable the system to automatically find clouds
services on the Internet, we created the Cloud Computing Crawl which
ensures the right on the Open-sourced Crawler for Java (crawler4j). To
make the crawling process easier and the data more thorough, we provided a
number of capabilities (such as add Seeds (), pick Crawling Domain (), and
add Crawlers Time ()). Additionally, we created the Confidence Feedbacks
Collection module to gather user feedback directly and store it in the Trust
Legit reviews Database as historical records: In fact, users often need to
register in order to verify their identities the very first time they want to use
submitting their credentials for storage in the Trust at the Access Control
Service (IdM).
Identity Registry: Furthermore, we established the Identification Info
Recycler subsystem to gather the overall number of founded personalities
among the whole identification behavior (i.e., all founded personalities for
customers who tried to give feedback loops to a specific cloud service).
The Trust Assessment Function is in charge of processing user requests
for trust assessments, which compare the reliability of cloud services and
compute the trust feedback factors (i.e., the credibility factors). We established
the Variables Calculations for proactive actions based on the set of variables
(more specifics on how well the believability variables are measured can be
found). In addition, we developed the Confidence Assessment to compare
the trust in cloud services by acquiring the aggregated weighted factors from
the Components Estimator to weigh feedbacks, and then figuring out the
mean of all the feedbacks given to each cloud service. The trust outcomes for
each cloud service are stored in the trust conclusions and component weights
storage, together with the component weights for trust feedback.

a) Experimental Evaluation:

We paid special attention to validating and researching the suggested cred­

ibility model’s resistance to various harmful behaviors, such as collusion
A Framework Promoting Position Trust Evaluation 291

and Sybil assaults under various behaviors, as well as the functionality of

our available model.

b) Credibility Model Experiments:

We put our credibility model to the test using actual customer reviews of cloud
services. We specifically crawled a number of review websites, including
CloudHostingReviewer.com, fog. findthebest.com, cloud storage supplier
sreviews.com, and cloud computing. findthebest.com, where consumers
submit feedback on cloud services they have utilized. The gathered informa­
tion is shown as a field H, where the feedback corresponds to the various
QoS criteria stated previously and is supplemented with a credential for each
associated customer. We were able to compile 10,076 comments submitted
by 6982 people about 113 actual cloud services. The project website has
made the gathered dataset available to the scientific community. The gath­
ered information was split into six categories of cloud computing, three of
which were used for experimental reasons.
Where each group, consisting of 100 users, was used to validate the
criteria needed against differential attack, as well as the other three catego­
ries have been used to model validation against Sybil assaults. Each cloud
storage group served as a representation of a different type of assaulting
behavior, including Waves, Uniform, and Peaks. The behavior models show
how many harmful feedbacks were introduced overall during a specific time
instance, for example, |V(s)| = 50 malicious feedbacks.
When testing against collusion attacks Ti = 50. When testing against
Sybil assaults, the behavior models also show the total number of names
created by attackers over a period of time (for instance, |I(s)| = 78 malevolent
individuals when Tj = 30. We modeled malicious feedback to improve the
trust results for cloud services in collusion attacks (i.e., a self-promotional
attack), while we modeled hostile feedback to decrease the trust results for
Sybil attacks (i.e., slandering attack). To assess how resilient our credibility
model is to malevolent conduct (such as conspiracy and Sybil assaults),
We carried out two sets of experiments: (1) testing the robustness of the
credibility model using a standard model Con(s, t0, t) (i.e., setting Cr(c, s,
t0, t) to 1 for all confidence feedbacks); and (2) assessing the performance of
the model using accuracy and recall (i.e. how well TMS detected attackers)
(i.e., how many detected attacks are actual attacks). In our trials, TMS started
paying cloud services that engaged in harmful behavior once the attack
percentage reached 35% (i.e., et(s) = 35%), so the awarding procedure would
only occur when there was a significant drop in the trust result.
292 Computational Imaging and Analytics in Biomedical Engineering

Six of the 12 experiments we ran were to test how well our credibility
model stood up to collusion assaults, while the other six were to test how
well our model stood up to Sybil attacks. According to Table 19.3, each
study is identified by a letter ranging from A to F.

TABLE 19.3 Behavior Experimental Design.

Malicious behaviors Experimental setting Waves Uniform Peaks
Collusion I A B C
outbreaks II A′ B′ C′
Sybil I D E F
outbreaks II D′ E′ F′

FIGURE 19.10 Robustness against collusion attacks.

A Framework Promoting Position Trust Evaluation 293

High resistance against Collusion Attacks: To raise the results of cloud

service trust (i.e., a self-promotional attack), we simulated hostile users who
gave input in the [0.8, 1.0] range. The assessment of six tests, which was
done to determine how resilient our model was against collusion attacks,
is shown in Figure 19.10. The trust results for experimental environment I
are shown in Figure 19.10 as A, B, and C, whereas the findings for experi­
ment setting II are shown as A′, B′, and C′. We observe that the stronger the
trust, the closer the temporal instance is to 100. which was determined using
the standard technique and the results are produced. This happens due to
dishonest customer feedback that is offered in an effort to increase the trust
rating of the cloud service. However, there is essentially no change when
the proposed credibility model is applied to estimate the outcomes of trust
Figures 19.10A, 19.10B, and 19.10C. These facts prove that our reputational
model is responsive to differential assault and capable of recognizing such
malicious behaviors. Additionally, we may make the exciting discovery that
the Normal behavior model works best in terms of recall while our feature of
sustainability perform great in terms of accuracy when the Waves approach
is utilized

FIGURE 19.11 Robustness against Sybil attacks.

294 Computational Imaging and Analytics in Biomedical Engineering

Structural rigidity against Attacks: For the experiments involving Sybil

attacks, we designed to simulate malicious users who sought to lower the
trust ratings of cloud services (i.e., a slandering attack) by creating multiple
identities and disseminating one malevolent feedback with a level of [0,
0.3] per individuality. The assessment of six tests that were run to test our
model’s resistance against Sybil assaults is shown in Figure 19.11, which
illustrates the trust results for the experimental setting I, whereas Figure
19.11 illustrates the results for the experimental setting II.
As the time instance approaches 100, as seen in Figure 19.11, the trust
results obtained using the standard model decline. This is due to malevolent
users that provide false feedback in an effort to lower the cloud service’s
trust score.
On the other hand, the findings of our suggested credibility model in
terms of trust are better than those of the traditional approach Figures 19.11D,
19.11E, and 19.11F. Here this is so because when the attacks took place, the
cloud service received a payment.
Actual Availability Compared to (a) Estimated Trust result accessi­
bility (b) Rate of caching errors.

FIGURE 19.12 Availability prediction and caching accuracy.

The greatest amount of liquid is documented when the Peaks behavior

model is applied (i.e., we can see five Figure 19.11F, which shows drops
which matches exactly the drops in the Peak position behavior as shown
in Figure 19.11. We can also see that some dramatic drop in the confidence
results obtained by taking into account our criteria are needed. This occurs as
A Framework Promoting Position Trust Evaluation 295

a result of TMS’s requirement that the percentage of assaults over the same
period of time exceeds a certain threshold before rewarding the impacted
cloud services (i.e., which is set to 25% in this case). This indicates that TMS
has given the impacted cloud service a reward based on the factor for the
change rate of trust outcomes. Additionally, Figure 19.11D′, 19.11E′19.11D′,
and 19.11F′ show that our credibility model performs best in terms of
accuracy when the Waves behavior model is applied (i.e., 0.48; see Fig.
19.11D′), whereas the maximum recall score is obtained when the Uniform
behavior model is applied (i.e., 0.75; see Fig. 19.11A′). The ability of TMS
to commend the impacted cloud service that uses the rate of confidence
results factor shows that our model can identify Attacks (i.e., either corporate
strategy attacks like those in the Waves and Uniform behavior models or
infrequent attacks like those in the Peaks behavior model) successfully.

c) Availability Model Experiments:

Using the same dataset, we used to verify the credibility model, we evaluated
our availability model. However, we concentrated on verifying the avail­
ability forecast accuracy, trust outcomes cache accuracy, and re-allocation
efficiency of the available model for the availability tests. Accuracy of Avail­
ability Prediction: We simulated 500 nodes housing TMS instance and set
the probability of failure for the nodes at 3.5%, which is consistent with the
results to test the predictive performance of the availability model.
This experiment was designed to investigate the applicability of our
method for estimate. For 100 time steps, we tracked the availability fluctua­
tions of TMS nodes that we had simulated. We gathered and compared the
anticipated accessibility of TMS nodes produced by our particle filter tech­
nique with their real availability. The outcome of one specific TMS node is
depicted in Figure 19.12. This figure shows that the projected availability and
the TMS node’s actual availability are extremely similar. This indicates that
our method is effective at tracking and forecasting TMS node availability.
Results of Trust Accurate Caching: To assess the accuracy of the avail­
ability model’s prefetching, we varied the cache management threshold to
identify the optimum number of new trust feedback mechanisms that TMS
actually received to recalibrate the trust result for a particular cloud storage
service without having experienced a significant error in the trust results. By
assessing the difference between the predicted trust values and actual trust
result of a particular cloud service, or the underlying cause error (RMSE),
also known as the cache management error, the accuracy of the trust result
caching is determined. The trustworthiness result caching is more accurate
296 Computational Imaging and Analytics in Biomedical Engineering

than the smaller RMSE value. The trustworthiness outcome caching reli­
ability of one specific cloud service is displayed in Figure 19.13. The chart
shows that when the cache threshold rises, the caching error grows approxi­
mately linearly.

FIGURE 19.13 Re-allocation performance.

Based on a tolerable caching error rate, the results enable us to select

the best caching threshold. If a 10% error margin, for instance, is accept­
able, the cache criterion can be set at 50 feedbacks. It is important to note
that the cache error was calculated using real users’ reviews of actual cloud
services.
The outcomes let us choose the ideal caching threshold based on an
acceptable caching error rate. The cache criteria can be set at 50 feedbacks,
for example, if a 10% error margin is acceptable. It is significant to high­
light that the cache fault was determined using assessments made by actual
customers of cloud services.
Reallocation Effectiveness: When the activity flow threshold ew (stms)
= 25%, we calculated the number of TMS nodes using distribution of wealth
of competence positive feedbacks but without re-allocation while increasing
the quantity of feedbacks, and when using reorientation of competence
feedback loops and without re-allocation while varying ew (stms). The more
TMS nodes there are, the more cost-effective TMS becomes. Figure 19.13
shows the outcomes of
I. The experimental settings show that the total number of TMS nodes
when utilizing the redistribution of trusted feedback systems method
A Framework Promoting Position Trust Evaluation 297

is fairly low and much more reliable than the overall number of TMS
nodes when reallocation is not employed. The outcomes of experi­
ments II are displayed in Figure 19.13. The graphic shows that the
number of TMS nodes decreases as the workload threshold increases.
The number of TMS nodes, however, is lower when the reshuffling of
trust feedback systems technique is used than when reshuffling is not
taken into account. This means that by lowering the number of TMS
nodes, our solution provides advantages in cutting the bandwidth
cost.

19.6 CONCLUSIONS

Accessing cloud storage for users is severely hampered by authorization

problems, especially when it comes to significant data, which really is
frequently highly sensitive. The trust architecture presented in this study
enables reliable defences against on/off and collaboration assaults, which are
key security issues that users of cloud services must contend with. Specifi­
cally, this paper introduces a criteria needed that recognizes Sybil attacks in
addition to false trust feedback loops from collusion attempts. Whether they
take place over a long or short period of time make an availability model
that maintains the appropriate degree of performance for the TES. The
experiment’s results demonstrate the viability of this tactic and demonstrate
the capacity to spot such malevolent activities. Controlling models should
be combined with trust evaluation for decentralized systems through the
suggested trust algorithms, which can spot on/off and collusion assaults and
guarantee the maximum degree of anonymity for cloud service customers, in
order to appropriately address these concerns.

KEYWORDS

• trust evaluation
• CloudArmor
• colluding attacks
• trust management
• cloud computing
298 Computational Imaging and Analytics in Biomedical Engineering

REFERENCES

1. Noor, T. H.; Sheng, M.; Alfazi, A. , In Proceedings of the 12th IEEE International
Conference on Trust; Security and Privacy in Computing and Communications,
Melbourne, Australia, July 16–18, 2013; pp 469–476.
2. Chang, E. , In Proceedings of the IEEE International Conference on Computer
Communications, Valencia, Spain, July 29–August 1, 2019.
3. Mahajan, S.; Mahajan, S.; Jadhav, S.; Kolate, S. Trust Management in E-commerce
Websites. 2017, , 2934–2936.
4. Habib, S. M.; Hauke, S.; Ries, S.; Muhlhauser, M. Trust as a Facilitator in Cloud
Computing: A Survey. 2012, , 19.
5. Khan, K.; Malluhi, Q. Establishing Trust in Cloud Computing, Qatar University, IEEE
IT Professional, 2010; vol (5).
6. Manuel, P.; Somasundaram, T. S. A Novel Trust Management System for Cloud
Computing – IaaS Providers. 2011, , 3–22.
7. Chong, S. K.; Abawajy, J.; Hamid, I. R. A.; Ahmad, M. A Multilevel Trust Management
Framework for Service Oriented Environment. 2013, , 396–405.
8. Wang, D.; Mullerr, T.; Liu, Y.; Zhang, J. Towards Robust and Effective Trust Management
for Security: A Survey. 2014.
9. Kotikela, S. N. S.; Gomathisankaran, M. In , International Conference on Cyber
Security, 2012.
10. Muchahari, M. K.; Sinha, S. K. In , IEEE International Symposium on Cloud and
Services Computing (ISCOS0), 2012.
11. Canedo, E. D.; de Sousa, R. T.; de Carvalho, R. R.; de Oliveira, A. R. In , IEEE International
Conference on Cyber Security; Cyber Warfare and Digital Forensic(CyberSec), 2012.
12. Noor, T. H.; Sheng, Q. Z.; Yao, L.; Dustdar, S.; Ngu, A. H. H. CloudArmor: Supporting
Reputation-Based Trust Management for Cloud Services. 2014.
13. Xiong, L.; Liu, L. Peertrust: Supporting Reputation-Based Trust for Peer-to-Peer
Electronic Communications. 2004, (7), 843–857.
14. Irissappane, A. A.; Jiang, S.; Zhang, J. In , Proceedings of the 2014 International
Conference on Autonomous Agents and Multi-Agent Systems, 2014; pp 1385–1386.
15. Liu, S.; Zhang, J.; Miao, C.; Theng, Y. L.; Kot, A. C. In , Proceedings of the 10th
International Conference on Autonomous Agents and Multi agent System, 2011; vol 3,
pp 1151–1152.
CHAPTER 20

EFFICIENT MACHINE LEARNING

TECHNIQUES FOR MEDICAL IMAGES
GURURAMA SENTHILVEL P., K. SAMPATH KUMAR, and
T. GANESH KUMAR
Department of Computing Science and Engineering, Galgotias
University, Greater Noida, Uttar Pradesh, India

ABSTRACT

Medical image analysis plays a pivotal role in modern healthcare,

aiding in diagnosis, treatment planning, and disease monitoring. With
the exponential growth in medical data, there is an increasing demand
for efficient machine learning techniques to extract valuable insights
from medical images while minimizing computational resources. This
abstract provides an overview of recent advancements in the realm
of efficient machine learning techniques for medical image analysis.
Efficient machine learning techniques for medical images are not only
crucial for improving patient care but also for overcoming challenges
posed by limited computational resources, data privacy concerns, and the
ever-increasing volume of medical data. Future research in this field will
likely continue to focus on enhancing the efficiency, interpretability, and
generalizability of machine learning models for medical image analysis.
Deep Learning Architectures: Convolutional Neural Networks (CNNs)
have revolutionized medical image analysis by providing state-of-the­
art results. Efficient architectures like Mobile Nets, Efficient Nets, and
Squeeze-and-Excitation networks have been adapted to reduce model
size and computational demands while maintaining high accuracy.

Computational Imaging and Analytics in Biomedical Engineering: Algorithms and Applications.

T. R. Ganesh Babu, PhD, U. Saravanakumar, PhD & Balachandra Pattanaik, PhD (Eds.)
© 2024 Apple Academic Press, Inc. Co-published with CRC Press (Taylor & Francis)
300 Computational Imaging and Analytics in Biomedical Engineering

Multi-modal fusion: Integrating information from various imaging

modalities (e.g., MRI, CT, X-ray) through fusion techniques enhances
diagnostic accuracy and reduces reliance on single-modal data. Explain­
able AI: Interpretable machine learning models provide insights into why
a particular decision was made, fostering trust and acceptance among
medical professionals

20.1 INTRODUCTION

AI may be a system for seeing plans that may be applied to medical images.
The same manner that associate necessary resource will facilitate in transfer
clinical findings, it will normally be twisted. AI often begins with the
machine learning calculation system schemingthe image remembers that are
conceded to be of significance for creating the supposition or examination of
interest. The Machine Learning calculation system additionally, at that time,
perceives the fashionable mixture of these image options for requesting the
image or enrolling some estimation for the given image space. There are
a handful of designs that may be used, every with totally different charac­
teristics and scarcities. There are open-supply metamorphoses of utmost of
those machine learning systems that alter them to endeavor to use to film
land. A handful of estimations for assessing the donation of a calculation
live; anyhow, one ought to be apprehensive of the doable connected snares
that may succeed in deceiving estimations. So a lot of recently, important
accomplishments have begun to be used; this fashion enjoys the profit that it
does not bear image purpose ID-associated computation as an underpinning
advance; rather, options are worthy as a region of the accomplishment frame.
AI has been employed in clinical imaging and can have an extra clear
impact from then on out. Those operating in Medicalpictures ought to be
apprehensive of however machine accomplishment capacities.

20.2 IMAGE PROCESS

Taking care of image may be a methodology to play out specific strategy on a

picture, to urge a superior image or to get rid of several accommodating data
from it.1 It is a quite sign taking care of whereby input is and result could also
be picture or characteristics options connected thereupon image. Currently,
image addressing is among swiftly making marches.2 It structures concentrate
assessment locus within coming up with and programing disciplines too.
Efficient Machine Learning Techniques for Medical Images 301

• Image addressing in an exceedingly general sense consolidates the

going with three phases;
• Acquiring the image through image obtaining contrivances;
• Taking piecemeal and dominant image.
Yield, during which result will be altered image or report that depends
upon image examination. There are two sorts of procedures used for picture
dealing with specifically; straightforward and progressed picture taking care
of. Basic picture dealing can be used for the printed duplicates like printouts
and photographs.3–6 Picture agents use various rudiments of interpretation
while using these visual techniques. Progressed picture taking care of meth­
odologies help in charge of the automated pictures by using PCs.8 The three
general stages that a wide scope of data need to go through while using
progressed strategy are pre- taking care of, redesign, and show information
extraction (Fig. 20.1).

20.3 KEY PHASES OF DIGITAL IMAGE PROCESSING

FIGURE 20.1 Key phases of digital image processing.

20.4 WHAT IS DIGITAL IMAGE?

The electronic pictures expect a fundamental part reliably. The clinical

imaging taking care of suggests managing pictures by using the PC. This
dealing with fuses numerous sorts of strategies and undertakings, for
302 Computational Imaging and Analytics in Biomedical Engineering

instance, picture getting, limit, show, and correspondence.9 The image is

a limit that implies an extent of properties like edification or concealing a
saw sight. The mechanized pictures enjoy a couple of benefits, for instance,
speedier and unassuming taking care of cost, basic taking care of and corre­
spondence, brief quality examination, different imitating with saving the
quality, fast and unobtrusive age, and flexible control. The bothers of cutting
edge pictures are misuse copyright, weakness to resize with saving the
quality, the need of huge breaking point memory, and the need of speedier
processor for control.10–15
An image taking care of technique is the utilization of PC to control the
modernized picture. This method enjoys many benefits, for instance, adapt­
ability, adaptability, data taking care of, and correspondence.16 With the
advancement of different picture resizing strategies, the photos can be kept
beneficially.17 This method has numerous plans of rules to perform into the
photos at the same time. The 2D and 3D pictures can be dealt with in various
viewpoints. The image dealing with techniques were laid out during the
1960s.18,19 Those systems were used for different fields like space, clinical
purposes, articulations, and TV picture improvement. During the 1970s with
the improvement of PC system, the cost of picture taking care of ended up
being less and speedier. During the 2000s, the image taking care of ended up
being speedier, humble, and less troublesome.20–23

20.5 WHAT IS DIGITAL IMAGE PROCESSING?

Progressed photography care of is that the operation of associate elec­

tronic laptop to manage progressive footage through associate algorithmic
program.24 As a subcategory or field of machine-driven signal taking care of,
innovative image coping with partakes in colorful high grounds over intro­
ductory photography care of. It permits considerably more broad extent of
calculations to be applied to the information and might avoid problems, for
case, the advancement of bouleversement and wringing throughout coping
with. Since footage are delineate north of 2 angles (perhaps more developed)
image coping with is also shown as advanced systems. The age and advance
of electronic photography care of chiefly plagued by 3 factors initial, the
improvement of PCs; second, the progression of calculation (particularly
the creation and improvement of separate numerical proposition); third, the
interest for a large extentof functions in tract, cultivating, military, diligence,
and clinical knowledge has extended.25
Efficient Machine Learning Techniques for Medical Images 303

20.6 WHAT ARE MEDICAL IMAGES?

Medical imaging, in any case called radiology, is the field of medicine

wherein clinical specialists recreate various pictures of parts of the body for
characteristic or treatment purposes. Medical imaging technique consoli­
dates innocuous tests that license experts to dissect wounds and diseases
without being nosy.27
Medical imaging is a central piece of the better consequences of
present- day drug.
Different sorts of clinical imaging strategies include:
• X-beams
• Attractive reverberation imaging (MRI)
• Ultrasounds
• Endoscopy
• Material imaging
• Electronic tomography (CT examine)
Other profitable clinical imaging systems consolidate nuclear prescrip­
tion helpful imaging strategies, similar to positron surge tomography (PET)
analyses.28–30 Various reasonsfor clinical imaging fuse compasses to see how
well your body is noting a treatment for a breakor infection (Fig. 20.2).

FIGURE 20.2 Medical image.

20.7 WHAT IS MEDICAL IMAGE PROCESSING?

Medical image handling includes the applying and essay of 3D image

datasets of the mortalbody, got most typically from a reckoned Tomography
304 Computational Imaging and Analytics in Biomedical Engineering

(CT) or glamourous Resonance Imaging (MRI) scanner to dissect patholo­

gies or companion clinical negotiations like careful medication, or for
exploration purposes (Fig. 20.3).31 Clinical image running is completed
by radiologists, specialists, and clinicians to promptly comprehend the life
structures of either individual cases or crowd gatherings.32

FIGURE 20.3 Medical image processing model.

Source: Reprinted with permission from Ref. [48]. Copyright © 2012, Society for Imaging
Informatics in Medicine

20.8 MEDICAL IMAGING TECHNOLOGIES

Progresses used in clinical imaging join those having a spot with the area
of radiography. X-bar and CT checks are fundamental resources, however
because of ionizing radiation, they ought to be used sparingly.33,34 Ionizing
radiation conveys with it possibility of growths, cascades, cell change,
and odd progression in incipient organisms. X-beams, including nuclear
appealing resonation (NMR), offer diminished bets and no ionizing radiation.
Ultrasound, using ultrasonic vibrations to make pictures, likely addresses the
most reliable sort of clinical imaging.35
The use of surface-mounted sensors to evaluate electrical development
is another safeguarded kind of clinical imaging and is used in electroen­
cephalography (EEG) and electrocardiography (ECG), but these advances
produce a change for a really long time chart rather than a graphical picture.36
Efficient Machine Learning Techniques for Medical Images 305

In different clinical imaging progresses modernized thinking (AI) is

working on the ability to interpret and analyze results. PC vision is being
used to ostensibly break down conditions not yet recognizable to the regular
eye.37

20.9 WHO USES MEDICAL IMAGING?

A radiographer in any case called a clinical imaging technologist or radi­

ology technologist is obligated for controlling clinical imaging frameworks.38
Radiographers are school ready with thorough data on the body’s plan
and how it is affected by different contaminations and wounds. They can
have pragmatic involvement with the systems referred toabove— including
MRIs and CT checks—as well as in locales, for instance,39
• Angiography—which incorporates imaging a patient’s veins and
heart.
• Adaptable radiography—which is the use of extraordinary machines
to perform imaging frameworks on patients who are too crippled to
even think about considering branching out to a crisis center.
• Fluoroscopy—which is a x-pillar that investigates the patient’s
internal body and shows moving pictures on a screen, like a film.
• Injury radiography—which frequently remembers work for emer­
gency workplaces.
Radiographers perform clinical imaging techniques in accordance with a
radiologist. Radiologists are experts arranged to examine and treat diseases
and wounds using clinical imaging propels. Radiologists are moreover
responsible for treating disorders—like harmful development and coronary
ailment—using radiation or insignificantly meddlesome, picture drove an
operation.40
At the point when the methods are done, the radiographer presents the
photos to the radiologist. The radiologist then separates the results, presents
an investigation of the disorder orinjury, and chooses the best treatment deci­
sions for the patient.

20.10 MEANING OF MEDICAL IMAGING

Clinical imaging helps experts to all the more promptly review patients’
bones, organs, tissue, and veins through effortless means. The methodologies
306 Computational Imaging and Analytics in Biomedical Engineering

help with choosing if operation would be a strong treatment decision; track

down tumors for treatment and removal; notice blood clusters or various
blockages; direct experts overseeing joint replacements or the treatment of
breaks; and help various frameworks including the game plan of contrap­
tions—like stents or catheters—inside the body.41
For the most part, clinical imaging has additionally evolved discoveries
and prescriptions by phenomenally reducing how much secret done by
subject matter experts, allowing them to even more truly deal with patients’
injuries and afflictions.

20.11 ADVANTAGES OF MEDICAL IMAGE PROCESSING

The guideline benefit of clinical picture dealing with is that it considers each
around, but inoffensive examination of internal life fabrics. 3D models of the
actuality designs of interestcan be made and audited to also foster treatment
results for the case, encourage better clinical contrivances and medicine
movement structures, or achieve further tutored anatomize. It has come one
of the crucial instruments used for clinical progress of late.42
The constantly dicing down at nature of imaging joined with state-of­
the-art programming instruments works with exact progressed duplication
of factual plans at colorful scales, too comparatively likewise with for the
utmost part changing parcels including bone and fragile apkins. Assessment
and development of re-sanctioning models which intertwine pukka factual
calculations allow the important occasion to more complete the process of
understanding, for illustration of relationship between determined life fabrics
and clinical contrivances.43

20.12 HOW DOES MEDICAL IMAGE PROCESSING WORK?

The course of medical image includes taking care of thresholds by getting

rough data from CT or MRI pictures and reproducing them into a plan proper
for use in applicable programming.44 A 3D bitmap of grey scale powers
containing a voxel (3D pixels) network makes the normal commitment for
picture taking care of. CT check grey scale power depends upon X-shaft
ingestion, while in MRI still hanging out there by the strength of signs from
proton patches during loosening up and after use of particularly strong
charming fields.
Efficient Machine Learning Techniques for Medical Images 307

For clinical guests, the duplicated picture volume is constantly dealt

with to part out and modify different regions of factual interest, analogous
to towel and bone.45 In Synopsys Simpleware programming, for illustration,
guests can do different picture taking care of undertakings at the 2D and 3D
positions, including:
• Lessening and wiping out unfortunate clatter or antiquated aberra­
tions with picturechannels.
• Managing and retesting input data to make it more direct and briskly
to manage pictures.
• Using division instruments to perceive different actual regions,
including motorizedtechniques using AI-based AI computations.
• Applying assessment and estimations gadgets to assess different bits
of the image data,for example, centerlines.
• Acquiring CAD models, similar to embeds or clinical devices, to
focus on how they communicate with individual life structures.
• Trading managed models for real science based diversion, further
game plan work, or for 3D printing certifiable pantomimes of the
presence structures being suggested to an incredible continuous
representation of how clinical picture dealing with included
patient-unequivocal hemodynamic entertainments of awesome
aortic investigations, a piece of work finished at University
College London into better getting hazardous vascular conditions.
• Experts used Simpleware programming to deal with CT yields and
create models sensible for CFD assessment, with the going with
propels taken.
• CT looks are gained from patient-unequivocal occurrences of aortic
investigations.
• Information is imported to Simple wareScanIP to imitate patient
estimation, including the treatment of noise, and division of areas
of interest like the dismantled aorta and branches Scripting is
used to normally finish smoothing computations to kill pixelation
trinkets. Surface models are made from the dismantled aorta and
imported to ANSYS® programming to set up CFD reenactments,
including intraluminal strain and divider shear-stress-based
records.
• Simulation results create hemodynamic pieces of information that can
be used to help future clinical understanding.
308 Computational Imaging and Analytics in Biomedical Engineering

20.13 HOW IS ARTIFICIAL INTELLIGENCE USED IN MEDICINE?

Man-made consciousness in medication is the utilization of AI models to

look through clinical information and uncover bits of knowledge to assist
with further developing wellbeing results and patient encounters. Because
of ongoing advances in software engineering and informatics, man-made
reasoning (AI) is rapidly turning into a vital piece of present-day medical
care.45 Artificial intelligence calculations and different applications fueled
by AI are being utilized to help clinical experts in clinical settings and in
continuous exploration.
At present, the most well-known jobs for AI in clinical settings are clinical
choice help and imaging examination. Clinical choice help devices assist
suppliers with settling on choices about medicines, drugs, psychological
wellness, and other patient necessities by furnishing them with speedy
admittance to data or examination that is pertinent to their patient. In
clinical imaging, AI instruments are being utilized to investigate CT filters,
x-beams,46 MRIs, and different pictures for sores or different discoveries that
a human radiologist could miss.
The difficulties that the COVID-19 pandemic made for some wellbeing
frameworks likewise drove numerous medical services associations all over
the planet to begin field-testing new AI-upheld advancements, for example,
calculations intended to assist with checking patients and AI-controlled
devices to screen COVID-19 patients.
The exploration and aftereffects of these tests are as yet being accumu­
lated, and the general guidelines for the utilization AI in medication are as yet
being characterized. However open doors for AI to help clinicians, scientists,
and the patients they serve are consistently expanding. Now, there is little
uncertainty that AI will turn into a center piece of the advanced wellbeing
frameworks that shape and back current medication.47

20.14 MAN-MADE INTELLIGENCE APPLICATIONS IN MEDICAL

FIELD

There are different ways AI can strongly influence the demonstration of

medicine, whether it is through speeding up the speed of investigation or
helping clinicians with making better decisions. The following are a couple
of examples of how AI could be used:
Efficient Machine Learning Techniques for Medical Images 309

20.14.1 WHAT IS MACHINE LEARNING?

Man-created intelligence may be an operation of man-created mindfulness

(AI) that empowers systems to therefore acquire and ameliorate for a reality
while not being expressly changed. Simulated intelligence revolves the
sweetening of computer programs that may get to knowledge and use it to
seek out every alone. Computer grounded intelligence is that the risk that
a computer program will learn and accommodates new knowledge without
mortal intervention. Simulated intelligence may be a field of man-created
attentiveness (AI) that keeps a PC’s empirical estimations current paying
very little brain to changes within the generalprudence.

20.14.2 USES OF MACHINE LEARNING

• Artificial insight is employed in varied locales for various reasons.

Exchanging constructions may be conversant in understanding new
pursuit open doorways. Progressing and online business stages may be
tuned to offer precise and re-tried plans to their shoppers considering
the clients’ net search history or past exchanges. Attributing affilia­
tions will be part of AI to expect dreadful advances and assemble a
credit risk model. Server ranch centers will utilize AI to hide tremen­
dous extents of reports from all edges of the globe. Banks will create
compulsion clear confirmation contraptions from AI techniques some
machine learning ways. PC grounded know-how assessments are
often mentioned as composed or freelance. Administered AI calcula­
tions will apply what has been conceded within the history to new
knowledge victimization named counsels for anticipated unhatched
events, ranging from the examination of a proverbial coming up with
dataset, the skill estimation conveys a reasoned capability to create
hypotheticals relating to the outgrowth values. The basic structure will
provide centers to any new commitment once decent coming up with.
The skill calculation will in like manner discrepancy its outgrowth
and therefore the right, organized outgrowth and notice botches to
vary the model meetly.

On the other hand, freelance AI calculations are used once the data
won’t to make preparations is neither gathered nor named. Freelance skill
focuses on how structures will accumulate associate capability to depict a
310 Computational Imaging and Analytics in Biomedical Engineering

hid development from unlabeled knowledge. The circumstance does not

understand the proper outgrowth; however, it researches and might attract
derivatives from datasets to depict coated structures from unlabeled data.
Semimanaged AI estimations fall several spots in coordinated and freelance
skill, since they use each named and unlabeled knowledge for obtaining
ready-naturally a confined quantum of named knowledge and a lot of unla­
beled knowledge. The structures that use this technique will astoundingly
additionally foster skill perfection picked once the noninheritable named
knowledge needs blessed and applicable coffers for set it up/gain from it.
Else, obtaining unlabeled knowledge overall does not bear recent coffers.
Support AI calculations may be a skill system that helps out its gift state
of affairs by transferring exercises and tracks down bumbles or remuner­
ates. Trial and error hunt and conceded value are the foremost applicable
characteristics of facilitate skill. The procedure subventions machines and
programming specialists to usually select the foremost ideal approach of
acting within a particular setting to grow its show. Essential award analysis
is reckoned upon for the knowledgeable to admit that movement is great; this
can be called the assistance signal (Fig. 20.4).

FIGURE 20.4 Machine learning types.

20.15 ARTIFICIAL INTELLIGENCE SURGERY

This is possibly the main locale for Machine Learning, and it will end up
being impressively more notable in the near future. You can isolate robotized
operation into the going with classes:
• Programmed stitching.
• Careful work process displaying.
• Improvement of automated careful materials.
• Careful expertise assessment.
Efficient Machine Learning Techniques for Medical Images 311

Sewing basically suggests shutting everything down excruciating injury.

Making this cycle robotized makes the whole methodology more restricted
while eliminating pressure on the subject matter expert.
The best hardships for AI in healthcare:

Area Challenges
Data governance Clinical information is as yet private and taboo for access. In any
case, as per a Wellcome Foundation study in the UK, just 17%
of public respondents are against offering their clinical data to
outsiders.
Transparent algorithms The need for straightforward calculations is not simply expected
to meet severe medication improvement guidelines, yet addition­
ally as a general rule, individuals need to see how precisely
calculations create ends.
Optimizing electronic There is still a great deal of split information between various
records informational indexes that need genuinely sorting out. At the
point when the current situation improves, it will provoke
propels in private treatment courses of action.
Embracing the power The medical care industry should change its view on the worth
of data silos of information and the manner in which it could bring esteem
from the drawn out viewpoint. Drug organizations, for instance,
are ordinarily hesitant to change their item methodologies and
exploration without prompt monetary advantages.
Data science experts Attracting more Machine Learning trained professionals and
data science specialists is truly huge for both the clinical
consideration and medication adventures.

20.15.1 TOP TEN OPERATIONS OF MACHINE LEARNING IN

DRUG COMPANY AND MEDICINE

The persistent making range of functions of AI in clinical advantages

grants United States to require a goose at a future wherever informa­
tion, assessment, andadvance work indivisibly to assist bottomless cases
while not them genuinely feting it. Once a brief time, it is going to be
exceptionally standard to search out ML-based operations introduced
with steady patient information open from completely different clinical
advantages systems in colorful countries, consequently growing the ample­
ness of recent treatment opinions that were distant at one time. Then area
unit the trendy ten functions of AI in clinical benefits.
312 Computational Imaging and Analytics in Biomedical Engineering

1. Characteristic conditions and opinion

One in every of the focal cubic centimeter operations in clinical advantages
is that the distinctive proof and assurance of diseases and affections that
area unit for the utmost half flashed back to be arduous to anatomize.
This could fuse something from cancers that area unit exhausting to trace
down throughout the retired stages, to alternative ingrain afflictions. IBM
Watson genetic science is a perfect depiction of however designing internal
reckoning with genome-grounded nasty growth sequencing will facilitate
in creating a speedy finding. Berg, the biopharma beast is victimization
AI to cultivate important specifics in areas like medical specialty. P1vital’s
PReDicT (Predicting Response to Depression Treatment) hopes to cultivate
AN economically possible strategy for designation and provides treatment in
routine clinical conditions.

2. Drug discovery and producing

One in every of the abecedarian clinical uses of AI lies in planning stage
drugs speech act method. This conjointly fuses R&D propels, for case, frontal
line sequencing and perfection drug which may facilitate in chancing ex
gratia ways for treatment of complex affections. As of now, the AI methods
incorporate freelance admitting which may fete plans in information while
not giving any hypotheticals. Project Hanover created by Microsoft is
together with ML-grounded developments for quite whereas together with
creating AI-grounded advancement for dangerous development treatment
and tweaking drugs mix for AML (Acute Myeloid Leukemia).

3. Medical imaging designation

Man-created intelligence and vital accomplishment area unit each to blame
for the high position advancement referred to as pc Vision. This has detected
protestation within the InnerEye drive created by Microsoft that manages
image demonstrative instruments for image assessment. As AI finally ends
up being a lot of accessible and as they fill in their educational limit, want
to check more information sources from varied clinical imagination come
back a bit of this AI-driven scientific cycle.

4. Personalized drugs
The conventions will still the method that a lot of be possible by coordinative
individual substance with even handed examination is additionally ready
Efficient Machine Learning Techniques for Medical Images 313

area unit for redundant essay and higher grievance assessment. This moment,
specialists area unit restricted to poring a selected course of action of ends or
check the bet to the case considering his intriguing history and open heritable
info. Anyhow, AI in drug is creating implausible strides, and IBM Watson
medical specialty is at the frontal line of this advancement by victimization
patient clinical history to assist with making completely different treatment
opinions. Sooner instead of late, we will see more contrivances and biosen­
sors with current substance assessment capacities hit the request, permitting
more information to open up for similar fashionable in school ML-grounded
clinical advantages marches.

5. AI-grounded behavioral revision

Social revision could be a vital piece of preventative tradition, and since the
time the addition of AI in clinical advantages, bottomless new associations
area unit jumping up within the fields of infection balance and clear proof,
patient treatment, etc. Somatix could be a B2B2C-grounded information
examination association that has sent a ML- grounded operation to check
movements that we tend to create in our everyday schedules, permitting
United States to induce our careless approach to acting and do polar upgrades.

6. Sensible health records

Study the apprehensive of current substance records is an fierce cycle, and
flashing back that advancement has had its impact in operating with the
info member method, very so as of currently, an even bigger piece of the
cycles place coffers into an opportunity to bring to a close. The introductory
occupation of AI in clinical advantages is to ease cycles to avoid wasting
time, trouble, and rich person. Record request systems victimization vector
machines and ML-grounded OCR protestation ways that area unit sluggishly
grouping brume, for case, Google’s Cloud Vision API and MATLAB’s AI
grounded handwriting protestation development. MIT is moment at the
terribly front of encouraging the exceptional amount of clever, sensible
substance records, which is able to be part of ML-grounded sacrifices starting
from the foremost stage to assist with finish, clinical treatment studies, etc.

7. Clinical test and exploration

PC-grounded intelligence includes a number of implicit operations within
the field of clinical starters and assessment. As anybody within the drug
company business would tell you, clinical fundamentals bring an enormous
314 Computational Imaging and Analytics in Biomedical Engineering

loading of your time and rich person and may invest in some occasion to
complete, once in mistrustfulness. Applying ML- grounded perceptive
assessment to fete implicit clinical starter contenders will facilitate consul­
tants with drawing a pool from a good assortment of instructional effects, for
case, past skilled visits, on-line recreation, etc. AI has conjointly detected
use in icing nonstop checking and information access of the abecedarian
individualities, seeing the trendy model size as tried, and victimization the
ability of electronic records to lessen information grounded botches.

8. Crowdsourced information collection

Overtly supporting is veritably disreputable within the clinical field pres­
ently, permitting investigators and consultants to induce to a colossal
proportion of knowledge enraptured by individuals considering their own
concurrence. This live substance information has uncommon ramifications
within the method drugs are going to be seen down the road. Apple’s analysis
Kit licenses guests to induce natural operations that apply ML-grounded
facial protestation to endeavor to treat Asperger’s and degenerative disorder
impurity. IBM really helped out Medtronic to decompress, total, and create
open polygenic disorder and hormone information persistently established
on the freely upheld info. The movements being created in IoT, the clinical
thought assiduousness is at now following down new habits by that to use
this information and attack serious to- examine cases and facilitate within the
overall improvement of finish and drug.

9. Higher radiotherapy
One in every of the foremost pursued uses of AI in clinical thought is within
the area of Radiology. Clinical image assessment has colorful separate rudi­
ments which may arise at a selected definition of your time. There are unit
colorful injuries, illness foci, etc., which may not be primarily displayed
victimization advanced circumstances. Since ML-grounded estimations gain
from the massive range of various models open exhausting, it becomes less
advanced to anatomize and notice the rudiments. One in every of the fore­
most notable functions of AI in clinical image assessment is that the depic­
tion of papers, for case, injuries into orders like traditional or uncommon,
injury or non-sore, etc. Google’s DeepMind Health is de facto serving to
experimenters in UCLH with creating calculations which may fete the isola­
tion among sound and dangerous towel and farther foster radiation treatment
for identical.
Efficient Machine Learning Techniques for Medical Images 315

10. Irruption prediction

System should know how the grounded advancements and AI area unit
moment conjointly being place to use in checking and prognosticating
pestilences every round the earth. moment, scientists approach a good deal
of information assembled from satellites, harmonious net primarily based
diversion revives, web site info, etc., faux mind networks facilitate to appear
at this info and anticipate everything from nature fever occurrences to outra­
geous nonstop overwhelming affections. Awaiting these eruptions is very
precious in immature countries as they have introductory clinical structure
and academic systems. A abecedarian illustration of this can be the ProMED-
correspondence, AN Internet- grounded specifying stage that screens propel­
lant impurities and arising bones and handovers flare reports unceasingly.

KEYWORDS

• machine learning
• image processing
• artificial intelligence
• electronic tomography
• convolutional neural networks
• MRI Images
• CT Images

REFERENCES

1. Saravanakumar, C.; Gururama Senthilvel, P.; Thirupurasundari, D.; Periyasamy,

P.; Vijayakumar, K. In Plant Syndrome Recognition by Gigapixel Image Using
Convolutional Neural Network, Proceedings of the Fist International Conference on
Advanced Scientific Innovation in Science, Engineering and Technology, Chennai,
India, ICASISET 2020, May 16–17, 2020; EAI, 2021. https://doi.org/10.4108/
eai.16-5- 2020.2304207
2. Nagarajan, G.; Sampath Kumar, K. In Security Threats and Challenges in Public
Cloud Storage, 2021 International Conference on Advance Computing and Innovative
Technologies in Engineering (ICACITE); IEEE, 2021. https://doi.org/10.1109/
icacite51222.2021.9404641
3. Kirubakaran, J.; Venkatesan, G. K. D. P.; Sampath Kumar, K.; et al. Echo State Learned
Compositional Pattern Neural Networks for the Early Diagnosis of Cancer on the
316 Computational Imaging and Analytics in Biomedical Engineering

Internet of Medical Things Platform. J. Ambient Intell. Human Comput. 2021, 12,
3303–3316. https://doi.org/10.1007/s12652-020-02218-1
4. Vijayaraj, A.; Vasanth Raj, P. T.; Jebakumar, R.; Gururama Senthilvel, P.; Kumar, N.;
Suresh Kumar, R.; Dhanagopal, R. Deep Learning Image Classification for Fashion
Design. InWireless Communications and Mobile Computing; Hashmi, M. F., Ed.;
Hindawi Limited, 2022; vol. 2022, pp 1–13. https://doi.org/10.1155/2022/7549397
5. Saravanakumar, P.; Sundararajan, T. V. P.; Dhanaraj, R. K.; Nisar, K. Memon, F. H.; et
al. Lamport Certificateless Signcryption Deep Neural Networks for Data Aggregation
Security in WSN. Intell. Autom. Soft Comput. 2022, 33 (3), 1835–1847.
6. Jeyaselvi, M.; Dhanaraj, R. K.; Sathya, M.; et al. A Highly Secured Intrusion Detection
System for IoT Using EXPSO-STFA Feature Selection for LAANN to Detect Attacks.
Cluster Comput. 2022. https://doi.org/10.1007/s10586-022-03607-1
7. Das, B.; Mushtaque, A.; Memon, F.; Dhanaraj, R. K.; Thirumalaisamy, M.; Shaikh, M.
Z.; Nighat, A.; Gismalla, M. S. M. Real-Time Design and Implementation of Soft Error
Mitigation Using Embedded System. J. Circuits Syst. Comput. World Scientific
Pub Co Pte Ltd., 2022. https://doi.org/10.1142/s0218126622502802
8. Pichumani, S.; Sundararajan, T. V. P.; Dhanaraj, R. K.; Nam, Y.; Kadry, S. Ruzicka
Indexed Regressive Homomorphic Ephemeral Key Benaloh Cryptography for Secure
Data Aggregation in WSN. J. Int. Technol. 2021, 22 (6), 1287–1297.
9. Dhanaraj, R. K.; Krishnasamy, L.; et al. Black-Hole Attack Mitigation in Medical
Sensor Networks using the Enhanced Gravitational Search Algorithm. Int. J. Uncertain.
Fuzziness Knowledge-Based Syst. 2021, 29, 397–315. https://doi.org/10.1142/
S021848852140016X
10. Dhanaraj, R. K.; Ramakrishnan, V.; Poongodi, M.; Krishnasamy, L.; Hamdi, M.;
Kotecha, K.; Vijayakumar, V. Random Forest Bagging and X-Means Clustered
Antipattern Detection from SQL Query Log for Accessing Secure Mobile Data. In
Wireless Communications and Mobile Computing; Jain, D. K., Ed.; Hindawi Limited.,
2021; vol 2021, pp 1–9. https://doi.org/10.1155/2021/2730246
11. Dhanaraj, R. K.; Lalitha, K.; Anitha, S.; Khaitan, S.; Gupta, P.; Goyal, M. K. Hybrid
and Dynamic Clustering Based Data Aggregation and Routing for Wireless Sensor
Networks. J. Intell. Fuzzy Syst. 2021, 40 (6), 10751–10765). https://doi.org/10.3233/
jifs-201756
12. Krishnamoorthi, S.; Jayapaul, P.; Dhanaraj, R. K.; et al. Design of Pseudo-random
Number Generator from Turbulence Padded Chaotic Map. Nonlinear Dyn. 2021. https://
doi.org/10.1007/s11071-021-06346-x
13. Dhanaraj, R. K.; Krishnasamy, L.; Geman, O.; Izdrui, D. R. Black Hole and Sink Hole
Attack Detection in Wireless Body Area Networks. Comput. Mater. Continua 2021, 68
(2), 1949–1965. DOI: 10.32604/cmc.2021.015363
14. Ramasamy, M. D.; Periasamy, K.; Krishnasamy, L.; Dhanaraj, R. K.; Kadry, S.; Nam, Y.
Multi-Disease Classification Model Using Strassen’s Half of Threshold (SHoT) Training
Algorithm in Healthcare Sector. IEEE Access DOI: 10.1109/ACCESS.2021.3103746.
15. Ramakrishnan, V.; Chenniappan, P.; Dhanaraj, R. K.; Hsu, C. H.; Xiao, Y.; Al- Turjman,
F. Bootstrap Aggregative Mean Shift Clustering for Big Data Anti-pattern Detection
Analytics in 5G/6G Communication Networks. Comput. Electr. Eng. 2021,
95, 107380. https://doi.org/10.1016/j.compeleceng.2021.107380
16. Krishnasamy, L.; Ramasamy, T.; Dhanaraj, R.; Chinnasamy, P. A Geodesic Deployment
and Radial Shaped Clustering (RSC) Algorithm with Statistical Aggregation in Sensor
Networks. Turkish J. Electr. Eng. Comput. Sci. 2021, 29 (3). DOI: 10.3906/elk-2006-124
Efficient Machine Learning Techniques for Medical Images 317

17. Kumar, D. R.; Krishna, T. A.; Wahi, A. Health Monitoring Framework for in Time
Recognition of Pulmonary Embolism Using Internet of Things. J. Computat. Theor.
Nanosci. 2018, 15 (5), 1598–1602. https://doi.org/10.1166/jctn.2018.7347
18. Krishnasamy, L.; Dhanaraj, R. K.; Ganesh Gopal, D.; Reddy Gadekallu, T.; Aboudaif,
M. K.; Abouel Nasr, E. A Heuristic Angular Clustering Framework for Secured
Statistical Data Aggregation in Sensor Networks. Sensors 2020, 20 (17), 4937. https://
doi.org/10.3390/s20174937
19. Sathyamoorthy, M.; Kuppusamy, S.; Dhanaraj, R. K.; et al. Improved K-Means Based Q
Learning Algorithm for Optimal Clustering and Node Balancing in WSN. Wirel. Pers.
Commun. 2022, 122, 2745–2766. https://doi.org/10.1007/s11277-021-09028-4
20. Dhiviya, S.; Malathy, S.; Kumar, D. R. Internet of Things (IoT) Elements, Trends
and Applications. J. Computat. Theor. Nanosci. 2018, 15 (5), 1639–1643. https://doi.
org/10.1166/jctn.2018.7354
21. Jena, S. R.; Shanmugam, R.; Dhanaraj, R. K.; Saini, K. Recent Advances and Future
Research Directions in Edge Cloud Framework. Int. J. Eng. Adv. Technol. 2019, 9 (2),
439–444. https://doi.org/10.35940/ijeat.b3090.129219
22. Kumar, R. N.; Chandran, V.; Valarmathi, R. S.; Kumar, D. R. Bitstream Compression
for High Speed Embedded Systems Using Separated Split Look Up Tables (LUTs).
J. Computat. Theor. Nanosci. 2018, 15 (5), 1719–1727. https://doi.org/10.1166/
jctn.2018.7367
23. Irfan, S.; Dhanaraj, R. K. BeeRank: A Heuristic Ranking Model to Optimize the
Retrieval Process. Int. J. Swarm Intell. Res. 2021, 12 (2), 39–56. http://doi.org/10.4018/
IJSIR.2021040103
24. Rajesh Kumar, D.; Shanmugam, A. A Hyper Heuristic Localization Based Cloned Node
Detection Technique Using GSA Based Simulated Annealing in Sensor Networks. In
Cognitive Computing for Big Data Systems Over IoT; Springer International Publishing,
2017; pp. 307–335. https://doi.org/10.1007/978-3-319-70688-7_13
25. Prasanth, T.; Gunasekaran, M.; Kumar, D. R. In Big data Applications on Health Care,
2018 4th International Conference on Computing Communication and Automation
(ICCCA), Dec 2018. https://doi.org/10.1109/ccaa.2018.8777586
26. Sathish, R.; Kumar, D. R. In Dynamic Detection of Clone Attack in Wireless Sensor
Networks, 2013 International Conference on Communication Systems and Network
Technologies, Apr 2013 . https://doi.org/10.1109/csnt.2013.110
27. Rajesh Kumar, D.; ManjupPriya, S. In Cloud based M-Healthcare Emergency Using
SPOC, 2013 Fifth International Conference on Advanced Computing (ICoAC), Dec
2013. https://doi.org/10.1109/icoac.2013.6921965
28. Sathish, R.; Kumar, D. R. In Proficient Algorithms for Replication Attack Detection in
Wireless Sensor Networks &#x2014: A Survey, 2013 IEEE International Conference ON
Emerging Trends in Computing, Communication and Nanotechnology (ICECCN), Mar
2013. https://doi.org/10.1109/ice- ccn.2013.6528465
29. Lalitha, K.; Kumar, D. R.; Poongodi, C.; Arumugam, J. Healthcare Internet of Things –
The Role of Communication Tools and Technologies. In Blockchain, Internet of Things,
and Artificial Intelligence; Chapman and Hall/CRC, 2021; pp 331–348. https://doi.
org/10.1201/9780429352898-17
30. Rajesh Kumar, D.; Rajkumar, K.; Lalitha, K.; Dhanakoti, V. Bigdata in the Management
of Diabetes Mellitus Treatment. In Studies in Big Data; Springer: Singapore, 2020; pp
293–324. https://doi.org/10.1007/978-981-15-4112-4_14
318 Computational Imaging and Analytics in Biomedical Engineering

31. Chandraprabha, M.; Dhanaraj, R. K. In Machine Learning Based Pedantic Analysis of

Predictive Algorithms in Crop Yield Management, 2020 4th International Conference on
Electronics, Communication and Aerospace Technology (ICECA), Nov 5, 2020. https://
doi.org/10.1109/iceca49313.2020.9297544
32. Dhanaraj, R. K.; Rajkumar, K.; Hariharan, U. Enterprise IoT Modeling: Supervised,
Unsupervised, and Reinforcement Learning. In Business Intelligence for Enterprise
Internet of Things; Springer International Publishing, 2020; pp 55–79. https://doi.
org/10.1007/978-3-030-44407-5_3
33. Cynthia, J.; Sankari, M.; Suguna, M.; Kumar, D. R. In Survey on Disaster Management
Using VANET, 2018 4th International Conference on Computing Communication and
Automation (ICCCA), Dec 2018. https://doi.org/10.1109/ccaa.2018.8777331
34. Dhanaraj, R. K.; Shanmugam, A.; Palanisamy, C.; Natarajan, A. Optimal Clone
Attack Detection Model using an Energy-Efficient GSA based Simulated Annealing in
Wireless Sensor Networks. Asian J. Res. Soc. Sci. Humanit. 2016, 6 (11), 201. https://
doi.org/10.5958/2249-7315.2016.01186.2
35. Sathya, K.; Kumar, D. R. In Energy Efficient Clustering in Sensor Networks Using
Cluster Manager, 2012 International Conference on Computing, Communication and
Applications, Feb 2012. https://doi.org/10.1109/iccca.2012.6179177
36. Lalitha, K.; Varadhaganapathy, S.; Santhoshi, S.; Kumar, D. R. In A Review on
Possibilities of Hearing Loss and Implantable Hearing Devices for Teenagers, 2018
4th International Conference on Computing Communication and Automation (ICCCA),
Dec 2018. https://doi.org/10.1109/ccaa.2018.8777336
37. Kumar, R. N.; Karthick, S.; Valarmathi, R. S.; Kumar, D. R. Design and Analysis of
Multiply and Accumulation Units Using Low Power Adders. J. Computat. Theor.
Nanosci. 2018, 15 (5), 1712–1718.
38. Poongodi, C.; Lalitha, K.; Dhanaraj, R. K. The Role of Blockchains for Medical
Electronics Security. In Essential Enterprise Blockchain Concepts and Applications;
Auerbach Publications, 2021; pp 231–262.
39. Lalitha, K.; Kumar, D. R.; Poongodi, C.; Arumugam, J. Healthcare Internet of Things–
The Role of Communication Tools and Technologies. In Blockchain, Internet of Things,
and Artificial Intelligence; Chapman and Hall/CRC, 2021; pp 331–348.
40. Sathish, R.; Kumar, D. R. In Proficient Algorithms for Replication Attack Detection
in Wireless Sensor Networks—A Survey, 2013 IEEE International Conference ON
Emerging Trends in Computing, Communication and Nanotechnology (ICECCN);
IEEE, Mar 2013; pp 1–7.
41. Kumar, D. R.; Sathish, R. Mitigation of Replication Attack Detection in Clusters
Through a Mobile Agent in Wireless Sensor Networks. Int. J. Eng. Res. Appl. 2013,
ISSN: 2248–9622.
42. Arvindhan, M.; Rajeshkumar, D.; Pal, A. L. A Review of Challenges and Opportunities
in Machine Learning for Healthcare. In Exploratory Data Analytics for Healthcare,
2021; pp 67–84.
43. Arvindhan, M.; Rajesh Kumar, D. Analysis of Load Balancing Detection Methods Using
Hidden Markov Model for Secured Cloud Computing Environment. In Applications
of Computational Methods in Manufacturing and Product Design. Lecture Notes in
Mechanical Engineering; Deepak, B. B. V. L., Parhi, D., Biswal, B., Jena, P. C., Eds.;
Springer: Singapore, 2022. https://doi.org/10.1007/978-981-19-0296-3_53
Efficient Machine Learning Techniques for Medical Images 319

44. Arvindhan, M.; Dhanaraj, R. K. The Firefly Technique with Courtship Training
Optimized for Load Balancing Independent Parallel Computer Task Scheduling in
Cloud Computing. Int. J. Health Sci. 2022, 6 (S1), 8740–8751. https://doi.org/10.53730/
ijhs.v6nS1.6999
45. Juyal, V.; Pandey, N.; Saggar, R. In Impact of Varying Buffer Space for Routing Protocols
in Delay Tolerant Networks, 2016 International Conference on Communication and
Signal Processing (ICCSP); IEEE, 2016. https://doi.org/10.1109/iccsp.2016.7754562
46. Juyal, V.; Singh, A. V.; Saggar, R. In Message Multicasting in Near-Real Time Routing
for Delay/Disruption Tolerant Network, 2015 IEEE International Conference on
Computational Intelligence & Communication Technology (CICT); IEEE, 2015. https://
doi.org/10.1109/cict.2015.79
47. Juyal, V.; Saggar, R.; Pandey, N. On Exploiting Dynamic Trusted Routing Scheme in
Delay Tolerant Networks. Wirel. Pers. Commun. 2020, 112, 1705–1718. https://doi.
org/10.1007/s11277-020-07123-6
48. Deserno T. Medical image processing. Optipedia, SPIE Press, Bellingham, WA. 2009.
INDEX

A digitalization, 120
DMAIC (define, measure, analyze,
Achilles tendon ruptures (ATR), 177
improve, and control), 116
Artificial general intelligence (AGI), 113
DSS and CDSS, 115
Artificial intelligence (AI), 2, 105
EMR, 116, 117
ACM SigCHI meeting, 110, 111
EMR quality improvement, 114
artificial general intelligence (AGI), 113
HCI professionals, 118
business intelligence (BI)
MOOCs, 119
commonplace measurements, 131
portable innovations, 119
exploration views (EV), 134 Subjective Mental Effort Question
frameworks, 131 (SMEQ), 116
human memory, restrictions, 135 UI augmentation, 115
human-focused plan, 133 HospiSign stage, 106
measure of information, 136 human-PC communication stage, 106
OLAP inquiries, 131 logical and designing disciplines, 108
primary objective, 133 man-made consciousness, 108
SAP Research, 132 market investigation
digital hearing hyper-combined foundation market, 138
comparative perceptions, 123 hyper-converged infrastructure market,
Dicta-Sign Wiki, 125 139
human–PC cooperation plan, 120, 121 Hyper-Converged Infrastructure (HCI)
intelligent sonifications, 121 Market, 136
k-Nearest Neighbors Algorithm hyper-intermingling framework, 137
(k-NN), 125 hyper-joined foundation market, 138
Sign Wiki model, 125 hyper-merged foundation market, 137
signal-based gadgets, 125 hyper-united foundation market, 138
sound items, 124 monetary firms, 137
unpredictable human activities, 122 Ordinary AI (GOFAI), 110, 111
digital humanities (DH) rationalistic reaction, 112
advanced humanities, 126 track plans, 107
advanced philology, 127 two-way course, 110
challenge, 130 Atrial fibrillation, 78
computational philology, 127 materials and methods
enormous tent, 127 ADAM solver, 81–82
human–PC connection, 129 classifier performance enhancement,
interfaces, 130, 131 82–83
rural American 1950s, 128 ECG data description, 78, 79, 81
Edified experimentation, 112 time-frequency features, 83–84
flexibility, 109 Attack target scale (ATS), 288
Hadoop with, 112 Attacker scale (AS), 288
healthcare systems Autism spectrum disorder (ASD), 180
computerized diagnostics, 115 Automatic segmentation, 41
322 Index

B local binary pattern (LBP), 154

directive pixel area, 155
Binary BAT Algorithm (BBA), 245
GLCM medium, 156–158
Biomedical image processing
performance evaluation parameter, 159
challenges in, 24
reckoning operation, 155–156
heterogeneity of images, 25
methodology
robustness of algorithms, 25
materials and methods, 152
unknown delineation of objects, 25
methods, 153
Blood oxygen level-dependent (BOLD), 183
Computer-aided design (CAD), 102
Blood–brain barrier (BBB), 181
Contrast Limited Adaptive Histogram
Brain surface extractor (BSE), 43
Equalization (CLAHE), 256
Brainsuite tool, 67
Cooperation attack frequency (CAF), 287
Business intelligence (BI)
commonplace measurements, 131
D
exploration views (EV), 134
frameworks, 131 Dental images
human memory, restrictions, 135 literature survey
human-focused plan, 133 computed tomography (CT), 203
measure of information, 136 deep learning model, 203
OLAP inquiries, 131 semiautomatic model, 203
primary objective, 133 methodology
SAP Research, 132 Haung thresholding, 204
mandible region in sample 1 and 2, 204
C Otsu thresholding, 204
Panoramic Images, 204
Cache management error, 295, 296
results and discussion, 204
Cancer cells, 256
Cephalometric, 202 edge detection technique, 211, 212
Cervical cancer, 148 grayscale segmentation, 205
detection of, 150 Haung threshold method, 211
feature extraction, 153 original grayscale image, 209
literature survey segmented mandible length, 206, 207
general diagnosis, 150 segmented mandible region, 205, 206,
techniques for, 150–152 210
local binary pattern (LBP), 154 threshold value, 207, 208, 209
directive pixel area, 155 Dermoscopy
GLCM medium, 156–158 methods and materials
performance evaluation parameter, 159 classifier explanation, 232
reckoning operation, 155–156 NSCT algorithm, 231–232
methodology result and discussion
materials and methods, 152 normal and abnormal images, 232
methods, 153 Design detecting
Cervix cancer, 148 materials and methods
Channeled hotelling observer (CHO), 102 classification accuracy, 250
Collusion attack strength (CAS), 289 CSO + K means, 251, 252
Computer-aided automatic detection, 147 Cuckoo search optimization, 247
feature extraction, 153 extraction formula table, 248–249
literature survey image description, 245
general diagnosis, 150 K-Means is clustering, 247–248
techniques for, 150–152 median filter, 246
Index 323

segmentation accuracy comparison, 251 F

SVM classifier, 249, 250
Fast Discrete Curvelet Transform, 266
Diabetes mellitus (DM), 215
Feedback
biomedical datasets, 216
accuracy of, 282
healthcare professionals, 217
collusion assaults, 282–283
linear discriminant analysis (LDA), 218
discrimination attacks, 283
machine learning (ML), 219
intoxication attacks, 283
Naive Bayes (NB), 217
newcomer attacks, 284
National Health and Nutrition Survey
storage and evaluation, 281–282
(NHNS), 218
TMS, 284–285
Pima dataset, 218
T-RBAC scheme, 284
principal component analysis (PCA), 218
Fluid attenuated inversion recovery
proposed method, 219
(FLAIR), 178
classification, 220–221
Full width half maximum (FWHM), 31
results and discussion
Functional magnetic resonance imaging
dataset, 221–225
(fMRI), 68
support vector machine (SVM), 217
blood oxygen level-dependent (BOLD),
training methods, 218
183
Dicta-Sign Wiki, 125
brain functional activities, 181–182
Diffusion kurtosis imaging (DKI), 183
brain seizure images
Diffusion tensor imaging (DTI), 179
brain parts, 71–73
Digital hearing
emotional changes, 183
comparative perceptions, 123
neuro-developmental disorder, 182–183
Dicta-Sign Wiki, 125
post-traumatic stress disorder (PTSD), 182
human–PC cooperation plan, 120, 121
Fused images, 266
intelligent sonifications, 121
k-Nearest Neighbors Algorithm (k-NN), Fuzzy Local Information C-means
125 Clustering, 262
Sign Wiki model, 125
signal-based gadgets, 125
G
sound items, 124 Gaussian filter, 202
unpredictable human activities, 122 Genetic meta analysis
Digital humanities (DH) GWA datasets, 5
advanced humanities, 126 HapMap loci, 4
advanced philology, 127 image processing, 5–6
challenge, 130 meta analysis risk, 5–6
computational philology, 127 Gradient echo imaging (GRE), 179
enormous tent, 127
human–PC connection, 129 H
interfaces, 130, 131 Healthcare systems
rural American 1950s, 128 computerized diagnostics, 115
digitalization, 120
E DMAIC (define, measure, analyze,
Edge detection technique, 211 improve, and control), 116
Edified experimentation, 112 DSS and CDSS, 115
Electroencephalography (EEG), 88 EMR, 116, 117
Electronic pictures, 301–302 EMR quality improvement, 114
Exploration views (EV), 134 HCI professionals, 118
324 Index

MOOCs, 119 digitalization, 120

portable innovations, 119 DMAIC (define, measure, analyze,
Subjective Mental Effort Question improve, and control), 116
(SMEQ), 116 DSS and CDSS, 115
UI augmentation, 115 EMR, 116, 117
High angular resolution diffusion imaging EMR quality improvement, 114
(HARDI), 183 HCI professionals, 118
Histogram equalization (HE), 101 MOOCs, 119
HospiSign stage, 106 portable innovations, 119
Human Papilloma Virus (HPV), 148 Subjective Mental Effort Question
Human-computer interaction (HCI), 105 (SMEQ), 116
ACM SigCHI meeting, 110, 111 UI augmentation, 115
artificial general intelligence (AGI), 113 human-PC communication stage, 106
business intelligence (BI) logical and designing disciplines, 108
commonplace measurements, 131 man-made consciousness, 108
exploration views (EV), 134 market investigation
frameworks, 131 hyper-combined foundation market,
human memory, restrictions, 135 138
human-focused plan, 133 hyper-converged infrastructure (HCI)
measure of information, 136 market, 136, 139
OLAP inquiries, 131 hyper-intermingling framework, 137
primary objective, 133 hyper-joined foundation market, 138
SAP Research, 132 hyper-merged foundation market, 137
digital hearing hyper-united foundation market, 138
comparative perceptions, 123 monetary firms, 137
Dicta-Sign Wiki, 125 Ordinary AI (GOFAI), 110, 111
human–PC cooperation plan, 120, 121 rationalistic reaction, 112
intelligent sonifications, 121 track plans, 107
k-Nearest Neighbors Algorithm two-way course, 110
(k-NN), 125 Hyper-converged infrastructure (HCI)
Sign Wiki model, 125 market, 136
signal-based gadgets, 125
sound items, 124 I
unpredictable human activities, 122 Image quality assessment (IQA), 268
digital humanities (DH)
advanced humanities, 126 K
advanced philology, 127
challenge, 130 K-Nearest Neighbors Algorithm (k-NN), 125
computational philology, 127
enormous tent, 127 L
human–PC connection, 129 Larmor frequency, 25, 26
interfaces, 130, 131 Linear discriminant analysis (LDA), 218
rural American 1950s, 128 Local binary pattern (LBP), 154
Edified experimentation, 112 directive pixel area, 155
flexibility, 109 GLCM medium, 156–158
Hadoop with, 112 performance evaluation parameter, 159
healthcare systems reckoning operation, 155–156
computerized diagnostics, 115 Long short-term memory (LSTM), 78
Index 325

Lung cancer, 236 artificial intelligence (AI), 310

convolutionary neural network (ConvNet/ clinical test and exploration, 313–314
CNN) conditions and opinion, 312
architecture, 238 crowdsourced information collection, 314
detection using CT images drug discovery and producing, 312
bar graph-comparison, 262 grounded behavioral revision, 313
data sets, 261 higher radiotherapy, 314
removal of bone region, 257–258 irruption prediction, 315
late diagnosis, 236, 237 medical imaging designation, 312
methods and materials personalized drugs, 312–313
pre-processing, 237 sensible health records, 313
PET images, detection use, 308
bar graph, 265 digital image processing
data sets, 265 defined, 302
FLICM detection, 262–264 key phases of, 301
Fuzzy Local Information C-means electronic pictures, 301–302
Clustering, 262 image process, 300–301
image fusion, 266 man-made intelligence applications,
random forest classifier, 239 308–310
result and discussion medical imaging techniques, 303
Fast Discrete Curvelet Transform, 266 advantages of, 306
fused images, 266 channeled hotelling observer (CHO), 102
image quality assessment (IQA), 268 computer-aided design (CAD), 102
KNN classifier, 239 histogram equalization (HE), 101
multi model imaging fusion, 268 history of, 98
Non Sub-sampling Contourlet KARDIO, 102
Transform (NSCT), 267 meaning, 305–306
peak signal-to-noise ratio (PSNR), 268, modified histogram-based contrast
269 enhancement with homomorphic
performance analysis for, 268 filtering (MHFIL), 101
random forest classifier, 239 OS-EM rebuilding method, 103
RMSE, 270–271 processing, 303–304
segmented images, 239 technologies, 304–305
SIM, 270 and types, 97
spatially weighted Fuzzy C means uses, 305
clustering (SWFCM) working, 306–307
algorithm, 259–260 Magnetic resonance images (MRI)
applications
M Achilles tendon ruptures (ATR), 177
Machine learning (ML), 219 autism spectrum disorder (ASD), 180
architecture, 98 blood–brain barrier (BBB), 181
active learning, 100 cardiovascular disease diagnosis, 181
deep learning, 100–101 colorectal cancer, 177
evolutionary learning, 100 diffusion kurtosis imaging (DKI), 183
reinforcement learning, 100 diffusion tensor imaging (DTI), 179
semisupervised learning, 99 fluid attenuated inversion recovery
supervised learning, 99 (FLAIR), 178
unsupervised learning, 99 gliomas, 177
326 Index

gradient echo imaging (GRE), 179 T1-weighted versus T2-weighted images,

hematologic marrow diseases, 177 29
high angular resolution diffusion types
imaging (HARDI), 183 T1-weighted images, 176
lung cancer, 178 T2-weighted images, 176
mapping cerebral connectivity, 179–180 Market investigation
out-of-field recurrence (OFR), 177–178 hyper-combined foundation market, 138
SPIN echo magnetic resonance hyper-converged infrastructure (HCI)
imaging, 179 market, 136, 139
stroke, automated quantitative hyper-intermingling framework, 137
segmentation, 178 hyper-joined foundation market, 138
susceptibility-weighted imaging (SWI), hyper-merged foundation market, 137
183–184 hyper-united foundation market, 138
concept of monetary firms, 137
Larmor frequency, 25, 26 Medical imaging techniques, 21, 22
longitudinal magnetization, 27 clustering methods
longitudinal relaxation, 27 classification of, 192
out-of-phase condition, 26 hard computing clustering, 193
transversal relaxation, 27, 28 hybridsegmentation, 196
functional magnetic resonance imaging clustering techniques, 197
(fMRI)
MATLAB software, 197–198
blood oxygen level-dependent
image segmentation strategies, 191–192
(BOLD), 183
MRI brain image, 188
brain functional activities, 181–182
automatic segmentation, 189–190
emotional changes, 183
manual segmentation, 189
neuro-developmental disorder, 182–183
MRI scan, 22–23
post-traumatic stress disorder (PTSD),
MRI technique, 190–191
182
segmentation, classification, 191
noise in
Rician noise, 24 soft computing clustering
SNR, 23–24 Fuzzy C-means algorithm, 194–195
zero-mean Gaussian-distributed noise, K-means clustering, 193–194
23 spatial Fuzzy C-means algorithm,
quality metrics 195–196
SSIM, 38 Melanoma skin cancer, 229
T1 and T2 images, 38 materials and methods
results and discussions, 31 classification accuracy, 250
deformation field, 34, 35 CSO + K means, 251, 252
DICOM format, 33 Cuckoo search optimization, 247
joint histogram, 34, 36 extraction formula table, 248–249
Nifti images, 31, 32, 33, 35 image description, 245
segmented brain parts, 36, 37 K-Means is clustering, 247–248
spatial preprocessing, 29 median filter, 246
coregister, 30 segmentation accuracy comparison, 251
full width half maximum (FWHM), 31 SVM classifier, 249, 250
normalization, 30–31 Microarray data, 2
realignment (REALIGN), 30 Modified histogram-based contrast
segmentation, 31 enhancement with homomorphic filtering
smoothing, 31 (MHFIL), 101
Index 327

N R
National Health and Nutrition Survey Radial Basis Function Network (RBFN), 245
(NHNS), 218 Realignment (REALIGN), 30
Non Sub-sampling Contourlet Transform Receiver-running feature (ROC), 6–7
(NSCT), 267 Registration process, 65–66
Regression, 66
O Risk ratio (RR), 6
Odds ratio (OR), 6
Open source tools S
Doccano, 171 Seizure
NLTK, 171 genetic meta analysis
Spark NLP, 171 GWA datasets, 5
TextBlob, 171
HapMap loci, 4
Ordinary AI (GOFAI), 110, 111
image processing, 5–6
Out-of-field recurrence (OFR), 177–178
meta analysis risk, 5–6
hemosiderin, 64
P image preprocessing, 64
Papanicolaou smear test, 149 intensity normalization, 65
Partial volume classifier (PVC), 44 motion, 65
Peak signal-to-noise ratio (PSNR), 268, 269 registration process, 65–66
Pediatric brain regression, 66
cortical surface extraction and slice time correction, 65
brain regions for different datasets, 48–59 smoothing spatial filter, 66
brain surface extractor (BSE), 43
materials and methods
cerebrum labeling, 44
Brainsuite tool, 67
gray matter (GM), 59
nonuniformity correction, 43–44 clustering, method, 3–4
outputs of, 45–49 MedCalc, 3
partial volume classifier (PVC), 44 meta-analysis, 4
skull stripping, 43 nonlinear regression, 7
surface and volume registration odds ratio (OR), 6
(SvReg), 44–45 receiver-running feature (ROC), 6–7
tissue classification, 44 risk ratio (RR), 6
topology correction, 44 results and discussion, 67
volume estimation, 45 analysis techniques, 73–74
white matter (WM), 59 cerebrum labeling, 69
WISP removal, 44 cluster analysis, 8
materials and methods continuous measure, 11–12
human brain’s important parts and correction of topology, 69
functions, 42–43 correlation, 13
real-time datasets, 42 cortical thickness estimation, 70
Pima dataset, 218 functional magnetic resonance imaging
Post-traumatic stress disorder (PTSD), 182 (fMRI), 68
Principal component analysis (PCA), 218 generic meta analysis, 10–11
hemosiderin, 68
Q identification of cortex, 69
Quality metrics MedCalc, 7
SSIM, 38 medical care system, 73
T1 and T2 images, 38 non-linear regression method, 15–16
328 Index

pial area, 70 Sign Wiki model, 125

risk ratio and difference, 9, 13–14 Social media, 164
scalp and skull processing, 69 Spatial preprocessing, 29
scrub mask, 69 coregister, 30
seizures, 74 full width half maximum (FWHM), 31
splitting of hemisphere, 70 normalization, 30–31
stripping of skull, 68 realignment (REALIGN), 30
surface thickness of cortical area, 70 segmentation, 31
surface volume registration, 70 smoothing, 31
tissue, 69 Spatially weighted Fuzzy C means
WISP and DEWISP process, 70 clustering (SWFCM), 255
Sentiment analysis, 164 algorithm, 259–260
approaches for Subjective Mental Effort Question (SMEQ),
hybrid-based approach, 167 116
Lexicon-based approach, 166 Support vector machine (SVM), 169, 217
machine learning techniques, 166–167 Surface and volume registration (SvReg),
classification 44–45
levels, 165 Surgery, 148
limitation of, 172 Susceptibility-weighted imaging (SWI),
machine learning techniques 183–184
ADA boost algorithms, 170
classification algorithms, 168 T
data preprocessing, 168
decision tree, 169 Total dermoscopic value (TDO), 244
dimensionality reduction algorithms, 169 T-RBAC scheme, 284
feature selection, 168 Trust evaluation
feature vector construction, 168 design methodologies
gradient boosting, 170 trust evaluation system (TES), 279–281
K-means clustering, 169 existing system
KNN, 169 proposed system, 278–279
linear regression, 168–169 Service-Level Agreements (SLAs), 277
logistic regression, 169 system architecture, 278
Naive-Bayes, 169 feedback
random forest algorithm, 169 accuracy of, 282
support vector machine (SVM), 169 collusion assaults, 282–283
open source tools discrimination attacks, 283
Doccano, 171 intoxication attacks, 283
NLTK, 171 newcomer attacks, 284
Spark NLP, 171 storage and evaluation, 281–282
TextBlob, 171 TMS, 284–285
social media analytics T-RBAC scheme, 284
architecture, 170–171 security and accuracy, comparison, 289
Sexually Transmitted Infections (STIs), 149 actual availability, 294
Short time Fourier availability model experiments, 295–296
materials and methods collusion attacks, 293
EOG artifacts, 90 credibility model, 291–292
implementation using MatLab, 90–93 experimental evaluation, 290–291
Matlab transformSTFT, 90 Peaks behavior model, 294
STFT implementation, 88–89 reallocation effectiveness, 296
Index 329

structural rigidity, 294 deep learning model, 203

tolerable caching error rate, 296 semiautomatic model, 203
simulation results methodology
attack target scale (ATS), 288 Haung thresholding, 204
attacker scale (AS), 288 mandible region in sample 1 and 2, 204
collusion attack frequency, 287 Otsu thresholding, 204
collusion attack strength (CAS), 289 Panoramic Images, 204
collusion set (CS), 288 results and discussion, 204
cooperation attack frequency (CAF), 287 edge detection technique, 211, 212
feedback frequency, 287 grayscale segmentation, 205
trust model, 285, 286 Haung threshold method, 211
original grayscale image, 209
U segmented mandible length, 206, 207
UI augmentation, 115 segmented mandible region, 205, 206,
210
V threshold value, 207, 208, 209

Visual interpretation, 2 Z
X Zero-mean Gaussian-distributed noise, 23

X-ray image, 202

literature survey
computed tomography (CT), 203