Visit ebookball.

com to download the full version and

explore more ebook or textbook

(Ebook PDF) Quantitative MRI of the Brain 1st

edition by Mara Cercignani 1315363550
9781315363554 full chapters

_____ Click the link below to download _____

https://ebookball.com/product/ebook-pdf-quantitative-mri-of-
the-brain-1st-edition-by-mara-
cercignani-1315363550-9781315363554-full-chapters-22492/

Explore and download more ebook or textbook at ebookball.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

(Ebook PDF) Biomechanics of the Brain 2nd edition by Karol

Miller 3030049965 9783030049966 full chapters

https://ebookball.com/product/ebook-pdf-biomechanics-of-the-brain-2nd-
edition-by-karol-miller-3030049965-9783030049966-full-chapters-22578/

(Ebook PDF) Language in the Brain 1st edition by Helmut

Schnelle 0511730853 9780511730856 full chapters

https://ebookball.com/product/ebook-pdf-language-in-the-brain-1st-
edition-by-helmut-schnelle-0511730853-9780511730856-full-
chapters-9416/

(Ebook PDF) The Overflowing Brain Information Overload and

the Limits of Working Memory 1st edition by Torkel
Klingberg 0199888256 9780199888252 full chapters
https://ebookball.com/product/ebook-pdf-the-overflowing-brain-
information-overload-and-the-limits-of-working-memory-1st-edition-by-
torkel-klingberg-0199888256-9780199888252-full-chapters-22966/

(Ebook PDF) A Brief Survey of Quantitative EEG 1st edition

by Kaushik Majumdar 1351643517 9781351643511 full chapters

https://ebookball.com/product/ebook-pdf-a-brief-survey-of-
quantitative-eeg-1st-edition-by-kaushik-
majumdar-1351643517-9781351643511-full-chapters-22486/
(Ebook PDF) Concise Encyclopedia of Brain and Language 1st
edition by Harry Whitaker 0080964990 9780080964980
9780080964997 full chapters
https://ebookball.com/product/ebook-pdf-concise-encyclopedia-of-brain-
and-language-1st-edition-by-harry-
whitaker-0080964990-9780080964980-9780080964997-full-chapters-9510/

(Ebook PDF) Handbook of Quantitative Methods for Detecting

Cheating on Tests 1st edition by Gregory Cizek 1317588096
9781317588092 full chapters
https://ebookball.com/product/ebook-pdf-handbook-of-quantitative-
methods-for-detecting-cheating-on-tests-1st-edition-by-gregory-
cizek-1317588096-9781317588092-full-chapters-25420/

(EBook PDF) Hormonal Treatments in Gynaecology FP REV 1st

edition by Frank Columbus Mara Teresa Gmez Garca, Jose
Miguel Garca Cebrian 1685077846 978-1685077846 full
chapters
https://ebookball.com/product/ebook-pdf-hormonal-treatments-in-
gynaecology-fp-rev-1st-edition-by-frank-columbus-mara-teresa-gmez-
garca-jose-miguel-garca-cebrian-1685077846-978-1685077846-full-
chapters-22272/

LOCUS: LOcal Cooperative Unified Segmentation of MRI Brain

Scans 1st Edition by B Scherrer, M Dojat, F Forbes, C
Garbay ISBN 9783540757573
https://ebookball.com/product/locus-local-cooperative-unified-
segmentation-of-mri-brain-scans-1st-edition-by-b-scherrer-m-dojat-f-
forbes-c-garbay-isbn-9783540757573-13380/

(Ebook PDF) Research Design Qualitative Quantitative and

Mixed Methods Approaches 1st edition by John Creswell
1452226105 978-1452226101 full chapters
https://ebookball.com/product/ebook-pdf-research-design-qualitative-
quantitative-and-mixed-methods-approaches-1st-edition-by-john-
creswell-1452226105-978-1452226101-full-chapters-23032/
Q uantitative MRI
of the
Brain
PRINCIPLES OF PHYSICAL MEASUREMENT
Series in Medical Physics and Biomedical Engineering
Series Editors: John G. Webster, E. Russell Ritenour, Slavik Tabakov
and Kwan-Hoong Ng

Recent books in the series:

Quantitative MRI of the Brain: Principles of Physical Measurement, Second edition

Mara Cercignani, Nicholas G. Dowell and Paul S. Tofts (Eds)

A Brief Survey of Quantitative EEG

Kaushik Majumdar

Handbook of X-ray Imaging: Physics and Technology

Paolo Russo (Ed)

Xun Jia and Steve B. Jiang (Eds)

Todd A. Kuiken, Aimee E. Schultz Feuser and Ann K. Barlow (Eds)

Emerging Technologies in Brachytherapy

William Y. Song, Kari Tanderup, and Bradley Pieters (Eds)

Environmental Radioactivity and Emergency Preparedness

Mats Isaksson and Christopher L. Rääf

Michael G. Stabin

Radiation Protection in Medical Imaging and Radiation Oncology

Richard J. Vetter and Magdalena S. Stoeva (Eds)

Statistical Computing in Nuclear Imaging

Arkadiusz Sitek

John G. Webster (Ed)

Radiosensitizers and Radiochemotherapy in the Treatment of Cancer

Shirley Lehnert

Diagnostic Endoscopy
Haishan Zeng (Ed)

Medical Equipment Management

Keith Willson, Keith Ison, and Slavik Tabakov

Quantifying Morphology and Physiology of the Human Body Using MRI

L. Tugan Muftuler (Ed)
Q uantitative MRI
of the
Brain
PRINCIPLES OF PHYSICAL MEASUREMENT

Edited by
Mara Cercignani
Nicholas G. Dowell
Paul S. Tofts
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2018 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

The first edition of this book, entitled Quantitative MRI of the Brain: Measuring Changes Caused by Disease, was published by Wiley in 2003

Printed on acid-free paper

International Standard Book Number-13: 978-1-138-03285-9 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and
information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors
and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if per-
mission 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 elec-
tronic, 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, please access www.copyright.com (http://www.copyright.com/) or con-
tact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that
provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system
of payment has been arranged.

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 of Congress Cataloging‑in‑Publication Data

Names: Cercignani, Mara, editor. | Dowell, Nicholas G., editor. | Tofts, Paul S., editor.
Title: Quantitative MRI of the brain : principles of physical measurement /
editors, Mara Cercignani, Nicholas G. Dowell, Paul S. Tofts.
Other titles: Series in medical physics and biomedical engineering.
Description: Second edition. | Boca Raton, FL : CRC Press, Taylor & Francis
Group, [2017] | Series: Series in medical physics and biomedical
engineering | Includes bibliographical references and index.
Identifiers: LCCN 2017034655| ISBN 9781138032859 (hardback ; alk. paper) |
ISBN 1138032859 (hardback ; alk. paper) | ISBN 9781315363578 (e-book) |
ISBN 1315363577 (e-book) | ISBN 9781315363554 (e-book) | ISBN 1315363550
(e-book) | ISBN 9781315363561 (e-book) | ISBN 1315363569 (e-book) | ISBN
9781315363547 (e-book) | ISBN 1315363542 (e-book)
Subjects: LCSH: Brain--Magnetic resonance imaging.
Classification: LCC RC386.6.M34 Q365 2017 | DDC 616.8/04754--dc23
LC record available at https://lccn.loc.gov/2017034655

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
 Contents

Foreword....................................................................................................................................................... vii
Foreword to the First Edition........................................................................................................................ix
Editors.. ...........................................................................................................................................................xi
Contributors................................................................................................................................................ xiii
Introduction.................................................................................................................................................. xv
Introduction to the First Edition................................................................................................................xvii

1 Concepts: Measurement in MRI............................................................................................................. 1

Paul S. Tofts
2 Measurement Process: MR Data Collection and Image Analysis.. ...................................................... 13
Paul S. Tofts
3 Quality Assurance: Accuracy, Precision, Controls and Phantoms.. .................................................... 33
Paul S. Tofts
4 PD: Proton Density of Tissue Water. . ................................................................................................... 55
Shir Filo and Aviv A Mezer
5 T1 : Longitudinal Relaxation Time. . ....................................................................................................... 73
Ralf Deichmann and René-Maxime Gracien
6 T2 : Transverse Relaxation Time............................................................................................................ 83
Nicholas G. Dowell and Tobias C. Wood
7 T 2* : Susceptibility Weighted Imaging and Quantitative Susceptibility Mapping.. ............................. 97
Sagar Buch, Saifeng Liu, Yongsheng Chen, Kiarash Ghassaban and E. Mark Haacke
8 D: The Diffusion of Water (DTI). . ....................................................................................................... 111
Francesco Grussu and Claudia A.M. Gandini Wheeler-Kingshott
9 Advanced Diffusion Models............................................................................................................... 139
Aurobrata Ghosh, Andrada Ianus and Daniel C. Alexander
10 MT: Magnetisation Transfer................................................................................................................ 161
Marco Battiston and Mara Cercignani
11 CEST: Chemical Exchange Saturation Transfer. . ................................................................................185
Mina Kim, Moritz Zaiss, Stefanie Thust and Xavier Golay
12 MRS: 1 H Spectroscopy. . ....................................................................................................................... 203
Yan Li and Sarah J. Nelson

v
vi Contents

13 Multinuclear MR Imaging and Spectroscopy.................................................................................... 235

Wieland A. Worthoff, Aliaksandra Shymanskaya, Chang-Hoon Choi, Jörg Felder, Ana-Maria Oros-Peusquens
and N. Jon Shah
14 T1-Weighted DCE MRI........................................................................................................................ 251
Leonidas Georgiou and David L. Buckley
15 Functional and Metabolic MRI.......................................................................................................... 269
Claudine J. Gauthier and Audrey P. Fan
16 ASL: Blood Perfusion Measurement Using Arterial Spin Labelling................................................. 283
Lisa A. van der Kleij and Esben Thade Petersen
17 Image Analysis.................................................................................................................................... 303
Siawoosh Mohammadi and Martina F. Callaghan
18 Future of Quantitative MRI................................................................................................................ 325
Mara Cercignani

Appendix A: Greek Alphabet for Scientific Use............................................................................. 329

Appendix B: MRI Abbreviations and Acronyms ............................................................................ 331
Index. . ............................................................................................................................................ 333
 Foreword

It is remarkable how much progress has occurred since the pub­ the prospect of transforming communications, monitoring,
lication of Paul S. Tofts’ first edition in 2003. Remember that and reporting. It is the path where MR and medicine are head­
radiology began as a highly qualitative field where shapes and ing. Quantitation provides objective measures of activity and
patterns were correlated with diseases. Initially, magnetic res­ ­disease – vital to assessing the extent of pathology and deter­
onance imaging thrived because it could produce undistorted mining the success of particular treatment protocols. The power
views in three planes. It has evolved considerably from beautiful of quantitative MR is underscored today by its incorporation in
images to being able to truly predict in vivo pathology. The pre­ clinical trials to demonstrate efficacy.
vious edition of this book was ahead of its time in articulating Quantitative MRI of the Brain begins with basic discussion of
the vision of quantitative MRI. How far the field will go in the data collection and image generation. Chapter 3 discusses qual­
future depends on the creativity of investigators and their ability ity assurance, precision, and accuracy. These concepts are criti­
to understand and solve technical problems. cal to consistency, an essential element of quantitative imaging.
To reliably predict biology on combinations of techniques Once grounded in the essentials of reproducibility, the text leads
could change approaches to disease. There are indeed many the reader through relaxation times and biophysical principles
possibilities for the future including accuracy in differentiating of susceptibility, diffusion, and magnetisation transfer. The last
aspects of disease such as tumor versus edema, necrosis ver­ chapters focus on techniques including functional MRI, spec­
sus tumor, benign versus malignant, and the grail – ­biological troscopy, and perfusion.
aggressiveness of lesions. The opportunities are abundant. Paul S. Tofts and his co-editors, Mara Cercignani and
I will cite two examples. Think about being able to predict Nicholas G. Dowell, have done a magnificent job assembling
aggressiveness and tumor volume of prostate cancer – totally ­contributors with appropriate expertise and have been success­
changing how surveillance is carried out. MR may also be able ful in compiling the requisites necessary to achieve a high level
to accurately separate therapeutic tissue changes such as the of understanding in quantitative MR. This is not a small feat and
effects of radiation or immunotherapy from recurrent tumor. is necessary for advancing the field. Congratulations on a sig­
This is essential in detecting infiltrating neoplasms such as nificant accomplishment!
glioma.
For those engaged in imaging, understanding how physical Robert I. Grossman
principles can be applied to produce quantitative results has NYU Langone Health

vii
 Foreword to the First Edition

Paul S. Tofts has succeeded brilliantly in capturing the essence of methodology. Any physician/scientist participating in a scientific
what needs to become the future of radiology in particular, and study or clinical trial must be familiar with the concepts elucidated
medicine in general – quantitative measurements of disease. This in this book. Although the text is focused on the brain, the con-
is a critical notion. The discipline of radiology started with the abil- cepts pertain to any imaging study. How to ensure that your results
ity to discern the shadows that were abnormal. On chest x-rays, will stand the test of the critical review is the underlying theme of
one could see the ‘white in the right’ and that was correlated to the first section on the concept of measurement in MR. Thorough
the ­clinical diagnosis of pneumonia. It is truly amazing how long knowledge of the principles of accuracy, precision, and quality
such descriptions were adequate and indeed, the state of the art. assurance are essential to the writing of any imaging proposal and
This transcended the modern era of cross-sectional imaging. CT the subsequent performance of the study.
and then MR heralded the ability to not only observe pathological The second section is focused on the metrics themselves. Here,
states but to specifically define and locate such conditions. Based there are lucid discussions of MR parameters that are the windows
upon absorption of biophysical parameters combined with posi- on the pathological processes we wish to study. This is complete and
tion, one could suggest that a particular abnormality was a stroke covers the intrinsic MR parameters (T1, T2, PD), diffusion, magne-
rather than a tumour or an infection. Make no mistake about it, this tisation, transfer, spectroscopy, dynamic contrast, perfusion, and
was an incredible scientific leap and has totally changed the calculus fMRI. To appreciate the strengths and limitations of these measures
of medicine. In the twenty-first century radiologists have become enables the reader to identify optimal parameters for particular
both the diagnostician and the arbiter of therapeutic efficacy. In studies. It also assists in the interpretation of the current literature.
clinical neuroscience the function of the neurologic exam has been The section offers a complete survey of all of the metrics used in
diminished by the unbiased, reliable nature of imaging. This has clinical MR today.
been mirrored throughout the body. The preeminent role of imag- The chapter on the biological significance of the MR parameters
ing now requires a new level of metric-quantitative measurements. in multiple sclerosis translates the imaging parameters to their
Why is quantitative methodology so vital? First and foremost, biological correlates. This is important, for if the measures are just
it is relatively unbiased compared with qualitative descriptions. abstract it is hard to argue for their implementation.
Second, it lends itself easily to statistical modelling. Lastly, if per- The last major section of the book deals with the topics of
formed correctly, the data can be pooled over multiple ­centres to ­registration of images and other measures, including atrophy,
provide power regarding a clinical trial or longitudinal study. texture, and volumetric analysis. Image registration is funda-
Thus, the natural history of a disease such as multiple sclero- mental when performing any longitudinal analysis. Just think
sis may be ascertained by a time-dependent study. This was first about it. When a radiologist is asked if a lesion has changed
made apparent when the FDA in the United States approved on two different studies, one must be careful that the apparent
the use of interferon beta-1b in 1993 based upon MRI data that change is not the result of technical differences (slice alignment,
revealed a decrease in disease activity and lesion burden. The slice thickness, etc.).
effect of the drug could not be ascertained from the clinical I was honoured to be asked by Professor Tofts to write this
measure of disability, the acknowledged gold standard for mul- foreword. In my opinion this text is a beautifully executed work,
tiple sclerosis. Approval of interferon beta changed the course capturing what is essential for radiologists and scientists to under-
of clinical treatment trials. What has ensued is a discussion of stand about quantitative MR measures. There is no more quali-
surrogate markers in imaging, sensitivity, specificity, reproduc- fied individual up to this task than Dr. Tofts. He is a lucid and
ibility, etc. The bottom line is the emergence of the mandatory most thoughtful scientist. I wish to extend my congratulations to
need to incorporate quantitative imaging techniques into treat- him and the other authors on this effort. This book will become
ment trails. a ­classic and the first of many on this significant topic.
This book addresses the measurement process, the mea- Robert I. Grossman
sures, what the measures mean biologically, and image analysis New York University School of Medicine

ix
 Editors

Mara Cercignani is professor of Medical Physics at the Brighton

and Sussex Medical School (BSMS). She has worked in the field of
MRI since 1998, and received her Doctorate from University College
London in 2007. Before moving to BSMS in 2011, she worked at San
Raffaele Hospital in Milan (1998–2002), the Institute of Neurology
in London (2002–2007) and Santa Lucia Foundation in Rome (2007–
2011). Her main research interests lie with the field of quantitative
MRI, spanning from diffusion MRI to quantitative magnetisation
transfer imaging.

Nicholas G. Dowell is a lecturer in Imaging Physics at the Brighton

and Sussex Medical School (BSMS). He received his Doctorate in
solid-state nuclear magnetic resonance from the University of Exeter
in 2004 before moving to the Institute of Neurology, University
College London, to work on quantitative magnetic resonance imag-
ing. He moved to the newly opened Clinical Imaging Sciences Centre
at BSMS in 2007. His research interests lie principally with quantita-
tive magnetisation transfer, diffusion-weighted imaging, data mod-
elling and analysis techniques in the brain.

Paul S. Tofts is emeritus professor at the Brighton and Sussex

Medical School (BSMS). After obtaining a BA in Physics from
Oxford University in 1970, the new University of Sussex ­provided an
ideal contrasting environment where his D Phil was in experimental
NMR studies of helium at low temperature. When biomedical NMR
hardly existed, from 1975 he researched radioisotope and CT imag-
ing at the Royal Postgraduate Medical School, London. At University
College London, in 1978, he developed a prototype 31P NMR machine
for newborn babies and started a career in quantitative MR with the
first measurement of absolute metabolite concentrations in vivo.
An early MRI machine in 1985 devoted to multiple sclerosis
studies at the Institute of Neurology, Queen Square (now part
of University College London), enabled Paul to develop a whole
range of quantitative imaging techniques and to edit the first book on quantitative MRI. Analysis
of dynamic Gd-enhanced image data enabled the quantifying of blood–brain barrier leakage, and
his mathematical model is now used extensively.
The new imaging centre at BSMS attracted Paul to return in 2006 to Sussex as the foundation
chair of Imaging Physics until 2009. He has 215 publications, 15,000 citations and an h-factor of 62.

xi
 Contributors

Daniel C. Alexander Ralf Deichmann Claudine J. Gauthier

Centre for Medical Image Computing Brain Imaging Center Department of Physics/PERFORM
(CMIC) Goethe-University Centre
Department of Computer Science Frankfurt/Main, Germany Concordia University
University College London (UCL) Montreal, QC, Canada
London, UK Nicholas G. Dowell
Department of Neuroscience
Marco Battiston Leonidas Georgiou
Brighton and Sussex Medical School
Department of Neuroinflammation Division of Biomedical Imaging
Brighton, UK
UCL Institute of Neurology University of Leeds
University College London (UCL) Audrey P. Fan Leeds, UK
London, UK Richard M. Lucas Center for Imaging
Stanford University Kiarash Ghassaban
Sagar Buch Stanford, CA, USA Magnetic Resonance Innovations Inc.
The MRI Institute for Biomedical Research Detroit, MI, USA
Waterloo, ON, Canada Jörg Felder
Institute of Neuroscience and Medicine
David L. Buckley Aurobrata Ghosh
(INM-4)
Division of Biomedical Imaging Centre for Medical Image Computing
Forschungszentrum Jülich
University of Leeds (CMIC)
Jülich, Germany
Leeds, UK Department of Computer Science
Shir Filo University College London (UCL)
Martina F. Callaghan Edmond and Lily Safra Center for Brain London, UK
Wellcome Trust Centre for Neuroimaging Sciences (ELSC)
UCL Institute of Neurology The Hebrew University of Jerusalem Xavier Golay
University College London (UCL) Jerusalem, Israel UCL Institute of Neurology
London, UK
University College London (UCL)
Claudia A.M. Gandini London, UK
Mara Cercignani
Wheeler-Kingshott
Department of Neuroscience
Department of Neuroinflammation
Brighton and Sussex Medical School René-Maxime Gracien
UCL Institute of Neurology
Brighton, UK Department of Neurology
University College London (UCL)
London, UK University Hospital
Yongsheng Chen Goethe-University
Department of Radiology and
Brain MRI 3T Mondino Research Centre Frankfurt/Main, Germany
Wayne State University
Detroit, MI, USA C. Mondino National Neurological Institute
Pavia, Italy Francesco Grussu
Chang-Hoon Choi and Department of Neuroinflammation
Institute of Neuroscience and Medicine Department of Brain and Behavioural UCL Institute of Neurology
(INM-4) Sciences University College London (UCL)
Forschungszentrum Jülich University of Pavia London, UK
Jülich, Germany Pavia, Italy

xiii
xiv Contributors

Ewart Mark Haacke Siawoosh Mohammadi Stefanie Thust

The MRI Institute for Biomedical Research Medical Center Hamburg-Eppendorf UCL Institute of Neurology
Waterloo, ON, Canada Department of Systems Neuroscience University College London (UCL)
and Hamburg, Germany London, UK
Department of Radiology
Wayne State University Sarah J. Nelson Paul S. Tofts
Detroit, MI, USA Department of Radiology and Brighton and Sussex Medical School
and Biomedical Imaging Brighton, UK
Magnetic Resonance Innovations Inc. University of California San
Detroit, MI, USA Francisco Lisa A. van der Kleij
San Francisco, CA, USA University Medical Center Utrecht
Andrada Ianus
Utrecht, The Netherlands
Centre for Medical Image Computing
Ana-Maria Oros-Peusquens
(CMIC)
Institute of Neuroscience and Medicine Tobias C. Wood
Department of Computer Science
(INM-4) Department of Neuroimaging
University College London (UCL)
Forschungszentrum Jülich Institute of Psychiatry, Psychology and
London, UK
Jülich, Germany Neuroscience
Mina Kim King’s College London (KCL)
UCL Institute of Neurology Esben Thade Petersen London, UK
University College London (UCL) Danish Research Centre for Magnetic
London, UK Resonance Wieland A. Worthoff
Copenhagen University Hospital Institute of Neuroscience and Medicine
Yan Li Hvidovre, (INM-4)
Department of Radiology and Denmark Forschungszentrum Jülich
Biomedical Imaging Jülich, Germany
University of California San Francisco N. Jon Shah
San Francisco, CA, USA Institute of Neuroscience and Medicine Moritz Zaiss
(INM-4) Department of Medical Physics in
Saifeng Liu Forschungszentrum Jülich Radiology
The MRI Institute for Biomedical Research Jülich, Germany Deutsches Krebsforschungszentrum
Waterloo, ON, Canada (DKFZ)
Aliaksandra Shymanskaya Heidelberg, Germany
Aviv A. Mezer Institute of Neuroscience and Medicine
Edmond and Lily Safra Center for Brain (INM-4)
Sciences (ELSC) Forschungszentrum Jülich
The Hebrew University of Jerusalem Jülich, Germany
Jerusalem, Israel
 Introduction

Quantitative MRI has continued to evolve since the first edi- developments that have taken place since the start of this mil-
tion in 2003, Quantitative MRI of the Brain: Measuring Changes lennium, in both hardware and software. All chapters have
Caused by Disease. The technological revolution is nowhere been substantially rewritten, mostly with new authors. We have
near to complete. The need for a new edition was apparent for included two new subjects (CEST and multinuclear spectros-
many years, yet there was no way to manifest it. With retirement copy). The new book is slimmer and, we believe, easier to use.
came other interests. Then a chance approach from Francesca We appreciate the input from the authors and reviewers,
McGowan at Taylor & Francis, and the realisation that my col- often in difficult circumstances. Bruce Pike, Geoff Parker and
leagues at the Brighton and Sussex Medical School could be Martina Callaghan gave invaluable advice. Robert Grossman
co-editors, enabled the new edition to be born. The first edi- again generously agreed to write an introduction. I am grate-
tion was born ‘one balmy March evening on the banks of the ful to Nicholas G. Dowell and Mara Cercignani for agreeing to
river Brisbane’; this edition was conceived on the former Physics undertake this project. Marica Dowell and Claudia Tofts gave
Bridge tea room at the University of Sussex and born in the key support.
nearby Swan Inn at Falmer.
In this new edition we have placed more attention on the Paul S. Tofts
‘how to’ aspects and incorporated the profound technological Brighton and Sussex Medical School

xv
 Introduction to the First Edition

This book was conceived one balmy March evening on the banks enormity of the work he had just created. The effort to produce
of the river Brisbane, in Queensland, Australia, where I had just this book is equivalent to about three person-years.
arrived for a sabbatical, and it became clear that the traditions Describing the intersection of measurement science and MR
of measurement science and MRI should meet. The notion of imaging has been an international effort by the members of the
a guide, a cookbook, for quantitative MRI (qMR) techniques MRI research community, with much communication by email
took seed and attained its own life, insistently telling me, in a and rapid access to journal articles online, in a way that would
variety of auspicious places and times, what had to be included. not have been possible a few years ago; some co-authors have not
With the help of a network of enthusiastic colleagues from the even (yet) met each other. The global village has truly arrived.
International Society for Magnetic Resonance in Medicine, a The overview boxes, and many of the footnotes, are generally
description of the state of the art has been assembled, which my responsibility. The conventions regarding units and abbre-
would be impossible for a single author to achieve. viations follow those in the style guide of the journal Magnetic
The Muse of qMR visited me in many places: Brighton, Glasgow, Resonance in Medicine, as much as possible.
Hawaii (Waikiki Beach and Molaka’i), Lewes, London, Oxford In the past, physical science has been concerned with our
and Paris (Jeu de Paume). Others have written in Bordeaux, the view of the cosmos and of atomic particles. Now we have the
Bronx, Chalfont St Peter, Guilford, Leiden, London, Manchester, chance to see and measure inside our own living brain – to me
Nijmegen, Northwood, Nottingham, Oxford, Philadelphia, this is equally profound. Which will history judge as being more
Utrecht and probably many other places. Jacob Bronowski also important? A decade ago qMR techniques were almost non-
inspired me.1 During the creative part of the process, I have been existent; in a decade’s time they will be routine.
aware of ideas coming to me in a variety of inspiring places and In spite of what science has achieved, I am aware that many
times, and I am aware of the remark by the composer Stravinsky, people are apparently able to resist disease, and heal themselves
whilst writing his Rite of Spring: ‘I was the vessel through which and others, in ways that are still mysterious to Western science,
the Rite passed’. John Cleese’s view2 is that ‘The chief [condition using treatments such as acupuncture, body work,4 homeopathy,
in which creativity can thrive] is to give people time and space reiki and shiatsu. The placebo effect is a phenomenon consid-
without pressure, simply to dream. Intelligence increases when ered very powerful in medicine, and yet its mechanism of action
you think less. We have too much noise in our heads. We need is not fully understood. With qMR we may be in a position to
quiet spaces; it’s about allowing something to happen to you’. objectively record responses to such treatments.
With the knowledge came the responsibility to make it widely Key people have been an inspiration to me at various times
available. Poets speak of ‘channelled poetry’, where the poet in my scientific education: A. Thompson, Eddie Palmer, Donald
is just told what to write. Gibran3 (speaking of children) says, Edmonds and Michael Richards. Later on, John Clifton, Richard
‘they come through you, but not from you’ and also ‘work is love Edwards, Osmund Reynolds and Ian McDonald provided sup-
made visible’. Sometimes I seemed to be witnessing the creation port at crucial early times during my entry into medical physics.
of perfection. I am grateful to my colleagues at the Institute of Neurology for
At times, it has been a lonely activity; my son Alex regularly their patience whilst I absented myself, working on this book.
brought me back to happiness after bleak days of writing. I was The conceptual design of the book had critical input from Clive
reminded of the composer (possibly Rachmaninov) who worked Baldock, Mark van Buchem and Peter Jezzard. Kate Brunskill,
for 5 days, from 5 o’clock in the morning to 8 o’clock in the eve- Jackie Cheshire and Jackie Powell gave invaluable help in obtain-
ning, then collapsed with tiredness at the end, overcome by the ing references and illustrations.
The contributors and reviewers have put in an enormous
1 Jacob Bronowski’s The Ascent of Man (reprinted 2002) is a particularly amount of work, often working long antisocial hours, and
thrilling history of science, giving its cultural and historical contexts. I am most grateful to them. Together they form a body of
2 From the London Times, 24 October 2002, based on Hare Brain, Tortoise
Mind by Guy Claxton.
3 The Prophet by Kahlil Gibran. 4 For example, Biodanza (see www.biodanza.co.uk).

xvii
xviii Introduction to the First Edition

experts who hold the expertise in the field of qMR. Robert Without the support of the Society, this book would not have
Grossman was extremely generous in his foreword. At Wiley been possible.
in Chichester, Martin Rothlisberger, Karen Weller, Wendy I hope the subject of qMR will become an established subtopic
Pillar and Lynette James steered this project to completion of MRI; the website www.qMRI.org can serve to coordinate
in an enjoyable and professional way. The Multiple Sclerosis activities (and to record errors found in this book). This book
Society of Great Britain and Northern Ireland has supported can be enjoyed as a view of what is possible in research centres
the physics development in the Research Unit at the Institute now and what will become increasingly routine in the future.
of Neurology, Queen Square in a very generous way for many
years, enabling a broad range of qMR techniques to be built Paul S. Tofts
up. Many of the contributors are associated with this unit. Brighton and Sussex Medical School
 1
Concepts: Measurement in MRI1

Contents
1.1 Introduction.................................................................................................................................. 1
Measurement science and MRI come together • Limits to progress • Using this book
1.2 History of measurement..............................................................................................................4
Early measurement • The longitude problem: John Harrison • Scientific societies • Units of
measurement • Mathematical physics • Scientific medicine • Early qMRI
1.3 Measurement in medical imaging.............................................................................................8
Images, partial volume and maps • Study design • Usefulness of an MR parameter
Paul S. Tofts 1.4 The future of quantitative MRI................................................................................................ 10
Brighton and Sussex Technology and methodology • International standardisation and certification
Medical School References.................................................................................................................................................11

1.1 Introduction chemistry, electrical engineering and the manufacturing indus-

try, there is a strong tradition of measurement, international
1.1.1 Measurement science and agreements on standards and training courses for laboratory
MRI come together practitioners. International standards of length, mass and time
have been in existence for many years. Secondary standards are
Measurement science has been around a long time; MRI2 has
produced, which can be traced back to the primary standards.
been around for about 35 years. This book is about the blending
National and international bodies provide coordination.
of the two paradigms.
As individual scientists we may have a passionate desire to use
We have come to expect to be able to measure certain quanti-
our talents for the benefit of mankind, preferring to devote our
ties with great accuracy, precision and convenience. Instruments
energy to finding better ways of helping our fellow humans to be
for mass, length and time are all conveniently available, and we
healthy than to improving weapons for their destruction. In this
expect the results to be reproducible when measured again and
context, developing measurement techniques in MRI constitutes
also to be comparable with measurements made by others in
a perfect application of traditional scientific skills to a modern
other locations. In the human body we expect to measure some
problem.
parameters (height, weight, blood pressure) ourselves, recognis-
MRI is now widespread and accepted as the imaging method
ing that some of these parameters may have genuine biological
of choice for the brain (and for many body studies). It is gener-
variation with time. More invasive measurements (e.g. blood
ally used in a qualitative way, with the images being reported
alcohol level or blood sugar level) are also expected to have a
non-numerically by radiologists. Many MRI machines now
well-defined normal range and to be reproducible. In physics,
have independent workstations, connected to the scanner
and the database of MR images, which enable and encourage
1 Reviewed by Mara Cercignani. simple quantitative analysis of the images in their numerical
2 The term Magnetic resonance imaging was invented by U.S. radiologists to
(i.e. digital) form. However, the data collection procedure often
describe nuclear magnetic resonance (NMR) imaging. The ‘nuclear’ part
was removed from the name NMR to prevent the public being alarmed. prevents proper quantification being carried out; variation in
NMR spectroscopy (Chapter 12) was originally concerned with identify- machine parameters such as transmitter output gain, flip angle
ing chemical compounds, and there was no spatial information contained value (and its spatial variation), receiver gain, and image scal-
in the data. It developed separately from imaging, on different machines, ing may all be acceptable for qualitative analysis but cause
and is often referred to as magnetic resonance spectroscopy (MRS). Modern irreversible confusion in images to be quantified. Researchers
MRS is carried out largely on MRI machines and uses the imaging gra-
dients to localise the spectra to particular parts of the body. For these
may be unaware of good practice in quantification and collect
reasons, MRI is now considered to include spectroscopy. MR, or magnetic or analyse data in an unsuitable way, even though the MRI
resonance, is a more correct term and refers to MRI and MRS together. machine is capable of more.

1
2 Quantitative MRI of the Brain

The process of quantifying or measuring parameters in the Thus, MRI has been undergoing a paradigm shift4 in how it
brain necessarily takes more time and effort than a straight- is viewed and used. In the past it was used for forming qualita-
forward qualitative study. More MRI scanner time is needed, tive images (the ‘happy-snappy MRI camera’, taking pictures); in
and considerable physics development effort and ­computing the future it may be increasingly used as a scientific i­ nstrument
resources may be needed to set up the procedure. In addi- to make measurements of clinically relevant quantities. The
tion, analysis can be very time-consuming, and support of the dichotomy can be seen in the MRI literature; radiological
­procedure is required to measure and maintain its reliability descriptions often speak of signal hyperintensity in a sequence
over time. Processes have to be found that are insensitive to with a particular weighting, whilst studies using physical mea-
­operator procedure (whether in the data collection or image surements often report localised concentration values, normal
analysis) and to scanner imperfections (such as radiofrequency ranges, age and gender effects, and reproducibility. As measure-
non-uniformity from a particular head coil), that provide good ment becomes more precise and analysis enables clinically rel-
coverage of the brain in a reasonable time and that are stable evant information to be extracted from a myriad of information,
over study times, which may extend to decades. it will become possible, in principle, to make measurements on
The benefits of quantification are that fundamental research an individual patient to characterise the state of their tissue,
into biological changes in disease, and their response to poten- guiding the choice of treatment and measuring its effect. The
tial treatments, can proceed in a more satisfactory way. Problems issues involved in bringing qMRI into the radiological clinic
of bias, reproducibility and interpretation are substantially were well summarised in an editorial in the American Journal of
reduced. MRI can move from a process of picture-taking, where Neuroradiology (McGowan, 2001; Box 1.1).
reports are made on the basis of unusually bright, dark, small As part of this ongoing paradigm shift, our view of what
or large objects, to a process of measurement, in the tradition of MRI can tell us is changing. When it started, information was
scientific instrumentation, where a whole range of quantities can largely anatomical (anatomical MRI), in the sense that relatively
be tested to see whether they lie in a normal range and whether large structures would be observed. Changes in their geomet-
they have changed from the time of a previous examination. ric characteristics (usually size), compared to normal subjects,
In this book, the intention is to demonstrate the merging of or to a scan carried out in previous weeks or months, would be
these two traditions or paradigms, that is measurement and noted. Quantitative examples would be volume and atrophy.
MRI, to form the field of quantitative MRI (qMRI).3 The MRI Functional MRI (fMRI) claimed the complementary ground,
measurement process is analysed, often in great detail. Limits studying short-term changes in tissue arising from carrying out
to accuracy and precision are identified, as far as possible, with particular (neural) functions. Microstructural MRI occupies a
the intention of identifying methods that are reliable and yet third role, as shown in this book. Many MR parameters (such as
practical in a clinical MRI scanning environment. The biologi- diffusion, magnetisation transfer and spectroscopy) show struc-
cal meaning of the many MR parameters that are available is tural changes in tissue arising from damage caused by disease
explored, and clinical examples are given where MR parameters at a microscopic level. To observe these changes directly would
are altered in disease. Often these changes have been observed require imaging resolution of the order of 1–100 µm,5 since they
qualitatively, and they serve to encourage us to improve the mea- generally involve a variety of biological changes at the cellular
surement techniques, in order that more subtle effects of disease level. These can be observed by pathologists in post-mortem
can be seen, earlier than currently possible, and in tissue that is tissue, using optical or electron microscopy and special stain-
currently thought to be normal as judged by conventional MRI. ing techniques (histopathology). This resolution is well below
The ideal is to obtain push-button (turnkey) techniques for each the spatial resolution of MRI (which is about 1 mm on clinical
of the many MRI parameters in this book, such that an MRI scanners). However, changes at the microscopic level (e.g. in
radiographer (technologist) can measure each of these param- cellular structure) give changes in the MR parameters (e.g. in
eters reliably and reproducibly, with a minimum of human water diffusion) that can be observed at coarser spatial resolu-
training or intervention, in the same way that we can currently tion (of about 1 mm). Thus, structural changes of sizes well below
step onto a weighing machine and obtain a digital readout of those that would be called anatomical can be detected. In addi-
our mass. In the case of qMRI, the output would be consider- tion, the concentrations of chemical compounds (metabolites)
ably richer, perhaps showing images of abnormal areas (com- in cells, and their changes, can be measured with spectroscopy.
puted from large databases of normal image data sets), changes The physiological permeability of the endothelial membrane
from a previous MRI exam, possible interpretations (diagnoses) around blood vessels can be measured using dynamic imaging
and an indication of certainty for each piece of information. The of Gd-contrast agent.
advances in the pre-scan and the spectroscopy MR procedures,
which used to be very time-consuming and operator-dependent
and are now available as fully automated options, show how this
4 Thomas Kuhn, in The Structure of Scientific Revolutions, first introduced
the idea of paradigm shifts. An example would be the move from a classi-
might be possible.
cal physics to a quantum physics view of the world. A paradigm is a pat-
tern or model, a way of viewing the world or part of it, a point of view, or a
mindset.
3 The website www.qmri.org can be used for updates. 5 1 micron (µm) is 10−3 mm or 10−6 m.
Concepts: Measurement in MRI 3

Box 1.1 On the use of quantitative MR imaging

There are a growing number of quantitative MR applica- agents. Coupled with results of other studies, including
tions that represent evolutionary change in the use of MR investigations in animal models in which correlation may be
imaging. These applications also include magnetisation observed between the results of an invasive or destructive test
transfer techniques, absolute T1 and T2 measurements, func- and the results of noninvasive MR imaging, human studies
tional imaging, and a number of spectroscopic techniques. serve to connect clinical observation with imaging findings.
A significant challenge in the clinical employment of quan- It is statistically advantageous to follow up preliminary
titative methods is that underlying physical mechanisms may studies that use ‘‘many’’ measures with targeted studies that
not yet be fully understood in the context of what can be have the power to accept or reject the hypothesis that certain
measured with the MR imaging experiment. For example, measures are significantly correlated.
one can associate the presence of abnormalities in quantita- Investigators differing from the authors of the original
tive measures with the presence of disease, but causality may study may do this only when precise and comprehensive data
not be established. regarding the study methods are provided. However, even
Thus, results are sometimes limited to empirical findings when the authors make a good-faith effort to disclose every
of correlation with some other measure or observable pro- nuance of the experimental method, it still may be difficult
cess. Still such results are potentially of great value by provid- to control for differences in MR hardware and software. This
ing means of noninvasive disease characterization and, thus, is in part because modern MR system design objectives are
insight into the natural history of disease. focused on obtaining excellent-quality clinical images for
Another substantial benefit is derived from the use of conventional, subjective interpretation.
validated methods to study the efficacy of novel therapeutic

Source: McGowan, J. C., American Journal of Neuroradiology, 22(8), 1451–1452, 2001.

These changes may occur both in a so-called lesion, which is process is driven by demand from clinical purchasers, by whether
tissue seen at post-mortem and in conventional MRI to be vis- competing manufacturers offer such facilities, and by whether
ibly different from the surrounding tissue, and in the ‘normal- public medical funding bodies such as the US Food and Drugs
appearing’ tissue, which appears normal in conventional MRI. Administration (FDA) are likely to approve reimbursement of
Lesions are usually described as focal, meaning that the change the cost of such procedures from medical insurance policies. The
is localised to a relatively small area (a few millimetres or centi- existence of a large and growing installed base of high-quality,
metres), with a distinct boundary; thus, its differing brightness reliable and ever-improving MRI machines primarily designed
in an image distinguishes it from the surrounding tissue (con- for routine clinical use, largely in environments where they can
sidered normal). In contrast, a diffuse change may extend over be run as parts of profitable businesses, has enabled and encour-
more area, has no distinct boundary and is harder to detect by aged the development on these machines of qMRI techniques,
simple visual observation of the image. Diffuse changes are often although they are still of interest to only a (growing) minority
well characterised by quantification, since it is the absolute value of users.7 As MRI machines evolve, the qMRI techniques usually
of quantities within the area that is measured, without reference have to be re-implemented.
to surrounding tissue or the need for a distinct boundary. Research institutions have particular structural strengths and
weaknesses. qMRI needs input from chemists, computer scien-
1.1.2 Limits to progress tists, neurologists, physicists, radiologists and statisticians. There
It may appear that qMRI research proceeds under its own impe- may be good career support for those applying methods to study
tus. However, the current state and rate of progress in developing clinical problems but none for those basic scientists inventing
reliable qMRI methodology are determined by several factors: and developing the methods. There may be a clash of paradigms
MRI manufacturers, research institutions, pharmaceutical or traditions, between those who have been educated in a hier-
companies, computer and electronics technology and publicly archical environment where asking questions is considered to be
funded research councils. irrelevant or subversive and those who consider asking questions
MRI machine manufacturers (vendors) will take on some to be an absolute basic necessity of undertaking modern high-
of the measurement procedures over time, incorporating them quality scientific research. The availability of talented research-
into their research and development programs and then offer- ers in turn depends on how much value is placed on science in
ing them as turnkey (push-button) products.6 The speed of this
7 The ‘killer app’ can sometimes galvanise action around making a qMRI
parameter available. This is when an application is found that has a clear
6 These are often sold as extras. clinical importance (e.g. MD in stroke).
4 Quantitative MRI of the Brain

society, schools and universities and whether appropriate post- post-mortem and possibly animal studies). For each MR param-
graduate training opportunities exist. The International Society eter, the following aspects are important: (1) the biological signifi-
for Magnetic Resonance in Medicine (ISMRM)8 is a powerful cance of the MR parameter, (2) how it can be measured accurately
force bringing together researchers from different institutions and slowly, (3) how it can be measured practically and quickly,
who are working on similar methodologies, through both its (4) examples of clinical applications, (5) what can go wrong in
journals and its scientific meetings. the measurement procedures, (6) QA approaches (controls and
The demand from pharmaceutical companies and neurolo- phantoms), (7) normal values for tissue, (8) reproducibility perfor-
gists for qMRI measurements to be used in drug trials is large mance that can be achieved, (9) multicentre studies and (10) future
and likely to increase (Filippi and Grossman 2002; Filippi developments. The editors did their best to get authors to consider
et al., 2002; McFarland et al., 2002; Miller 2002; Sormani et al., all 10 of these aspects in their chapters.
2011; Mallik et al., 2014). The traditional double-blind placebo- This book is intended to be a repository of qMRI method-
controlled Phase III trial involves many patients (typically 100– ology, of particular use to PhD students; hopefully, the meth-
1000) being studied for several years in order to obtain enough odology will not need to be reinvented by each generation of
statistical power to determine whether a drug is effective. The researchers. The first edition of this book (Tofts 2003) contains
large sample size is needed to deal with the variability of dis- some information not present in this edition that may be worth
ease in the absence of treatment, and the imperfect treatment consulting; the chapter authors in this edition have changed and
effect (which may vary according to patient subgroup). Such necessarily give a different perspective.
trials typically cost several hundred million US $. qMRI can In Chapters 2 and 3, the issues in measurement that occur
potentially shorten the procedure, by identifying treatment repeatedly throughout the book, as each MR parameter is consid-
failures early on in the testing process, on a smaller sample. ered, are examined in more detail. These are grouped into the pro-
If there is no observed biological effect from the treatment, it cesses of data collection, data analysis and quality assurance, all of
may be considered unlikely that the drug is working (this will which crucially affect how well MR quantities can be measured.
depend on the particular way the drug has been postulated to Units are usually given in SI (System International), and conven-
act). For example, if a potential treatment for multiple sclerosis tions used in this book for physical units and symbols (e.g. TR, TE,
(MS) showed no effect on all the MR measures that are known T1, T2) are those recommended in the style guide for the journal
to be abnormal in MS, it would probably be dropped in favour Magnetic Resonance in Medicine, published for the International
of other drugs. With new biotechnology and gene-based treat- Society for Magnetic Resonance in Medicine.9 Most of the focus is
ments being developed, the number of candidate drugs for on techniques that can be implemented on standard clinical MRI
evaluation will increase by a large factor, and traditional trials scanners; some techniques (e.g. 31P spectroscopy or 23Na imaging)
will become too expensive and slow to evaluate all of them. need non-standard hardware as an add-on.
Thus, direct in vivo qMRI observation of treatment effect could This book is intended for researchers who already have a basic
become increasingly valued. knowledge of how MRI works and some knowledge of the brain,
The rapid increase in power and availability of computing including the major diseases (cancer, epilepsy, stroke, multiple
technology has also been key in enabling data acquisition and sclerosis and dementia). Newcomers can find many appropri-
image analysis techniques to be realised. Numerically designed ate books (Table 1.1) and also helpful websites such as ISMRM.
magnets, coils and radiofrequency pulses, digital receivers and A table of common MRI abbreviations is given in Appendix 2.
rapid image registration and analysis have all changed the way
that MRI is carried out.
1.2 History of measurement
The resources available from pharmaceutical companies to
drive the process of developing and supporting reliable qMRI 1.2.1 Early measurement
measures may exceed those available from traditional publicly
Early quantitative techniques focused around the desire to mea-
funded research sources. Traditional research council sources
sure distance, mass, monetary value and time. An awareness of
have been willing to support the application of qMRI methods
these can give us perspective in our own endeavours to quantify!
to study particular diseases but often unwilling to support the
Developed in about 3000 BC in ancient Egypt, the cubit was
development of new quantitative methods, sometimes claiming
a ubiquitous standard of linear measurement, equal to 524 mm.
that MRI manufacturers should be doing this.
It was based on the length of the arm from the elbow to the
extended fingertips and was standardised by a royal master cubit
1.1.3 Using this book of black granite, against which all cubit sticks used in Egypt were
to be measured at regular intervals.10 The precision of the thou-
The field of quantitative MRI should include the following four
sands of cubit sticks used in building the great Pyramid of Giza
key areas: basic concepts of measurement, how to measure each
MR parameter (to include both acquisition and analysis) and
the biological significance of each parameter (with input from 9 See http://www.ismrm.org/journals.htm
10 Much of the historical material in this chapter comes from the
8 www.ismrm.org Encyclopaedia Britannica.
Concepts: Measurement in MRI 5

TABLE 1.1 Recommended books for background reading in MRI and neuroanatomy
Title Authors Date published Number of pages Description
MRI: Physical Principles and RW Brown, YC Norman Cheng, 2014 976 Thorough exposition of MRI
Sequence Design EM Haacke, MR Thompson, principles; 2nd edition
R Venkatesan
MRI from Picture to Proton DW McRobbie, EA Moore, MJ Graves 2017 400 Written by experienced physicists;
new edition
Quantitative MRI in Cancer TE Yankeelov, DR Pickens, RR Price 2011 338 Multi-author book by radiologists and
physicists; a ‘sister book’ to this one
Quantitative MRI of the Spinal Cord J Cohen-Adad, C Wheeler-Kingshott 2014 330 Multi-author, ‘sister’ book to this one
Diffusion MRI: Theory, Methods, DK Jones 2011 784 Covers much of quantitative brain
and Applications MRI from a physics point of view
Handbook of MRI Pulse Sequences M Bernstein, K King, X Zhou 2004 1040 A pulse programmer’s friend
MRI in Practice C Westbrook, CK Roth, J Talbot 2011 456 Established book giving
radiographer’s viewpoint
Barr’s The Human Nervous System: JA Kiernan, R Rajakumar 2013 448 Includes complete description of the
An Anatomical Viewpoint brain
Note: Statistics books are in Chapter 2 (Table 2.2).

is thought to have been very high, given that the sides of the 1.2.2 The longitude problem: John Harrison
pyramid are identical to within 0.05%.
Early astronomers developed remarkably precise measure- In the 18th century, the problem of navigation around the globe
ment methods (as demonstrated at Stonehenge); their ability was severe. Although latitude (distance from the equator) could
to guide navigation and predict eclipses brought them fame. be measured accurately, using the elevation of the sun above the
In the 16th century, precise calculations of planetary orbits by horizon at noon (the time of maximum altitude), longitude (the
Copernicus, Kepler and Galileo challenged the intellectual dom- easterly or westerly distance around the globe, now measured
inance of the Catholic church, bringing an end to the idea that from Greenwich, London, UK) could not be (Sobel, 2005). Samuel
all heavenly bodies rotate around the Earth. Pepys, commenting on the pathetic state of navigation, had written
In 1581, the word quantitative 11 was first used, meaning of ‘the confusion all these people are in, how to make good their
‘involving the measurement of quantity or amount’. Quantity reckonings, even each man’s with itself’, recognising the distinc-
means ‘size, magnitude or dimension’ in Middle English. In tion between intra- and inter-observer variation. Newton wrote of
1847, quantitative analysis was used, meaning ‘chemical analysis the sources of error involved in trying to measure time at sea: ‘One
designed to determine the amounts or proportions of the com- [method for determining longitude] is by a Watch to keep time
ponents of a substance’. In 1878, quantify was used to mean ‘to exactly. But, by reason of motion of the Ship, the Variation of Heat
determine the quantity of, to measure’, and hence quantification and Cold, Wet and Dry, and the Difference of Gravity in different
is ‘the operation of quantifying’. In 1927, quantitate was used to Latitudes, such a watch hath not yet been made’.
mean ‘to measure or estimate the quantity of, especially to mea- As a result many lives were lost at sea, through shipwreck and
sure or determine precisely’. However, Webster’s dictionary calls failure of supplies, and navigation was such a sensitive issue that
this term a ‘back-formation’,12 which is probably as derogatory sailors were forbidden to carry out their own calculations, for
as a dictionary compiler can be, and this term is not used in this fear that they would show up errors in those of their superior
book, nor is it in the Oxford English Dictionary. officers. The growth of vastly profitable world trade was held
Francis Bacon (1561–1626) had a great influence on generations back. In this context, the Longitude Act of 1714 was passed in the
of British scientists who followed him (Gribbin, 2003). He stressed British Parliament, offering a reward of £10,00013 to anyone who
collecting as much data as possible, then setting out to explain the could devise a method of measuring longitude accurately.
observations, instead of dreaming up an idea and then looking for The challenge of solving the ‘longitude problem’, as it came to
facts to support it. Science must be built on the foundation pro- be known, was taken up by an English clockmaker, John Harrison,
vided by the facts. What would he say about the modern ‘hypoth- who lived near the port of Hull and had heard the stories of souls
esis-driven’ research? In 1662, the Royal Society of London for the going to their death and the reward offered. He built four clocks
Promotion of Natural Knowledge received its charter from King altogether. The first kept good time on land (better than 1 second
James II, as one of the first (and best known) scientific societies. per month) and in small trips out to sea. The Longitude Board
could give incentive awards to help impoverished inventors bring
promising ideas to fruition. He succeeded in getting a full trial at
11 See Webster’s Dictionary and the Oxford English Dictionary. sea with the navy, on a voyage to Lisbon in 1736; his clock showed
12 A back-formation is a word formed by subtraction of a real or supposed
affix from an already existing longer word. Thus, quantitate was created
from quantitation. 13 The sum was graded according to the accuracy that could be achieved.
6 Quantitative MRI of the Brain

unexpected error at sea, being susceptible to an artefact caused by Weights and Standards became the National Bureau of Standards,
accelerations in the motion at sea. His own perfectionism, and obsti- then the National Institute of Standards and Technology (NIST).
nacy all round, delayed matters, and the next trial, taking his fourth In 1960, the 11th General Conference of Weights and Measures,
clock to the West Indies, did not take place for another 25 years. meeting in Paris, established the International System of Units,
The Longitude Board was dominated by eminent astronomers and based on the metre, kilogram, second, ampere, degree Kelvin
others from the naval establishment and repeatedly refused to give and candela. These units are often called the SI units, after the
Harrison his payment, requiring that the chronometer should first French expression Système Internationale, and are preferred
be taken from prototype into mass production. The Board realised in the scientific community.14 The kilogram is represented by
that replicate voyages and clocks were needed to establish the repro- a cylinder of platinum–iridium alloy kept at the International
ducibility, without which the accuracy could not be guaranteed. Bureau of Weights and Measures in France,15 with a duplicate in
A single measurement could not establish the maximum error. His the USA; the other units are defined with respect to natural stan-
son William took up his case, and the Royal Society offered him dards (e.g. the metre is defined by the wavelength of a particu-
a Fellowship. It was only intervention by King George III, and the lar visible atomic spectral line). National centres such as the US
passing of a second act by Parliament, that gave him his recognition, NIST and the UK National Physical Laboratory are now centres
at the age of 80, 46 years after he had built his first sea clock. of expertise in measurement science.
This story, of finding a scientific solution to a human problem,
has all the elements of the struggles that modern scientists may 1.2.5 Mathematical physics
have to develop a technique that they believe will save lives, and
many parallels can be seen. Harrison’s clocks are preserved in In parallel with the development of physical instruments
the old Royal Observatory at Greenwich. had been the discovery of mathematical techniques. Ancient
Babylonians, Egyptians, Greeks, Indians (Harappans) and
Chinese all had mathematics, originally used for computing
1.2.3 Scientific societies
areas and volumes of regular objects and also used for han-
The Lunar Society of Birmingham (England) was a group of dling monetary currency. In the 6th century BC, Pythagoras
­forward-thinking scientists who met between 1766 and 1791. They established the link between the musical note of a string and
met on the day of the full moon (so that travel would be easier) its length. This bridge between the world of physical experience
and flourished independently of the Royal Society (in London). and that of numerical relationships has been called the birth
Birmingham was the location of much inventive scientific activ- of mathematical physics, where numbers explain the origin of
ity stimulated by the industrial revolution. Both of Charles physical forms and qualities. Newton’s differential calculus and
Darwin’s grandfathers (Josiah Wedgewood, the pottery manu- Fourier’s transform are essential tools used by our current MRI
facturer, and Erasmus Darwin, the naturalist) were members, as scanners. Early digital computers, most famously used to deci-
were Matthew Boulton (the manufacturer), Joseph Priestly (who pher the Enigma code used by submarines during the Second
discovered oxygen) and James Watt (who invented the steam World War, developed to the stage we take for granted today.
engine). The Industrial Revolution in Britain and the rest of
Europe gave commercial impetus to the invention of a variety of
1.2.6 Scientific medicine
measuring instruments to be used in the manufacturing process.
Lord Kelvin, delivering a lecture on electrical units of measure- In medicine, the concepts of the new scientific methods, includ-
ment in 1883, expressed the desire of his time to quantify: ing quantification, were applied. William Harvey (1578–1657)
was a physician and scientist who studied the blood circulation
When you can measure what you are speaking about, extensively and was the first to measure the cardiac volume and
and express it in numbers, you know something about estimate the total blood volume in the human body. In 1833,
it; but when you cannot measure it, when you cannot William Beaumont, a US army surgeon, published a series of
express it in numbers, your knowledge is of a meagre and studies16 on a soldier who had been wounded in the stomach
unsatisfactory kind: it may be the beginning of knowl-
and then developed a flap that could be opened. Beaumont could
edge, but you have scarcely, in your thoughts, advanced
to the stage of science, whatever the matter may be. watch food in the stomach and extract gastric juice. Nowadays
we have more convenient ways of making in vivo studies.
although he might have added a caveat about the danger of num-
bers giving a pseudo-scientific respectability to some studies.
14 The engineering community in the USA still uses units based on the
British imperial system (although these are not used in the UK any more).
1.2.4 Units of measurement Incompatibility between imperial and metric units was blamed for a space
In the newly formed United States of America, it was found vehicle failure in the late 1990s.
15 The BIPM, Bureau International des Poids et Mesures; http://www.bipm.
impossible to reform the archaic system of weights and measures org/en/about-us/
inherited from the British, in spite of the Napoleonic metric 16 From “The Man with a Lid on his Stomach,” in the Faber Book of Science,

system that had recently been adopted in France. The Office of edited by John Carey.
Concepts: Measurement in MRI 7

Box 1.2 A plea for ‘good quality data collection’ in MRI1

The history of image processing in nuclear medicine shows the newer technique of GD-DTPA scanning, and therefore
that collection of good quality image data is at least as we have developed this data collection technique in prefer-
important as access to image processing techniques. Even ence. A second example is the use of expensive classification
now one could argue that real improvements in the useful- techniques on image data clearly showing gross nonunifor-
ness of image data come from instrumental improvements mity which can be removed relatively simply.
rather than from more sophisticated ways of image process- Having taken care of the instrumental aspects and
ing. However in the case of large datasets that are already of obtained good quality data, the processing requirements
good quality, the problem is then one of data presentation may become less expensive, and mostly consist of PACS, 3D
and reduction, rather than correcting images to compensate display, calculation of functional images, and segmentation
for errors in data collection. algorithms. Where sophisticated forms of information pro-
With this philosophy we have initially concentrated on cessing are required, to make full use of them they must be
collecting good quality data, that are sensitive to the clini- integrated into a programme that includes aspects of data
cal question being studied. For example T2 weighted images collection such as sequence design, quality control of instru-
of the brain can show Multiple Sclerosis lesions, and one mental parameters, validation of the quantitative results, and
could develop sophisticated algorithms for measuring lesion good experimental design. In summary, we believe that data
volume to assess disease and therapies; however the images must be appropriate, and of good quality, before undertaking
show oedema and scar tissue, which are secondary to the dis- any processing.
ease process. Primary visualisation of the disease is shown by

Source: Tofts, P.S., et al., Prog. Clin. Biol. Res., 363, 1991a; Tofts, P.S., et al., Information Processing in Medical Imaging, Wiley-
Liss Inc, 1991b. With permission.
1 PACS is ‘picture archival and computing system’ and refers to computer-based systems to store, display and interrogate large quantities of medical
images. By ‘functional images’, it was meant parametric maps of any kind (e.g. permeability).

In the late 1970s, scientists started connecting medical imag- completed a PhD thesis, ‘Some Exercises in Quantitative NMR
ing hardware to computers that look extremely basic by modern Imaging’. The aim was to ‘assess the potential of NMR imaging and
standards, motivated by the desire to manipulate and interro- spectroscopy with respect to tissue characterisation and evaluation
gate the images. Sophisticated medical imaging instruments of tissue response to radiotherapy and hypothermia’ (Bakker et al.,
were produced in nuclear medicine, ultrasound, X-ray com- 1984). Quantification was recognised by some radiologists as hav-
puted tomography and nuclear magnetic resonance. ing a potential role in studying disease (Tofts and du Boulay, 1990):
In about 1978 the annual meetings on Information Processing
in Medical Imaging started taking place. In 1989, it was argued Serial measurements in patients and correlation with
that attention to good data collection was at least as important as similar studies in animal models, biopsy results and
autopsy material taken together have provided new
sophisticated image processing (Box 1.2). The notion that good
knowledge about cerebral oedema, water compartmen-
quantification required attention to both data collection and tation, alcoholism and the natural history of multiple
image analysis techniques was born, and this complementarity sclerosis. There are prospects of using measurement to
can be seen in the structure of this book. Experience has shown monitor treatment in other diseases with diffuse brain
that advances are often made by groups who have access to both abnormalities invisible on the usual images.
data collection (so that the acquisition technique can be opti- When making quantitative measurements, the physi-
mised for the job in hand) and to advanced analysis techniques cist can adopt the paradigm of the scientific instrument
(to obtain the most from the data). Computing groups working designer, who is presented with a sample (the patient)
isolated from the clinical questions and acquisition hardware about which he or she wishes to make the most care-
ful, detailed measurements possible, in a non-destruc-
may produce solutions to non-existent problems, or use data that
tive way, using the infinitely adjustable instrument (the
are degraded by poor acquisition techniques. imager). The biological question to be answered, and
thus the biophysical feature to be measured, needs very
careful choice.
1.2.7 Early qMRI
Premature babies were studied with 31P MRS in 1983 (Cady et al., Quantitative magnetic resonance was the subject of a small
1983), prompting the measurement of absolute concentration of meeting organised by the UK Institute of Physics and Engineering
metabolites in the brain (Wray and Tofts, 1986). In 1985, Bakker in Medicine in 1997 at Dundee, Scotland, and it is here that the
8 Quantitative MRI of the Brain

expression qMR was first used. qMRI has now come to denote that are those of the image pixel, and the third dimension is the slice
part of MR which is concerned with quantitative measurements, in thickness. Often the image data set is 3D, although we can only
the same way that fMRI, MR angiography (MRA), MRS and quan- see a 2D slice through it at any one time.
titative magnetisation transfer (qMT) denote subspecialties of MR. The interplay between pixels and voxels is subtle. At times
when we are thinking of images, pixels are more natural, and
1.3 Measurement in medical imaging in fact the term originates from the science of interpreting
images of 2D surfaces (e.g. in robot vision or remote sensing of
Physical quantities can be intensive or extensive, and when we the Earth by satellite). Yet when we are thinking of the cuboids
are considering various properties and manipulations to quanti- of 3D tissue from which the pixel intensities originate, voxels
ties it can be helpful to be aware of these differences. An inten- are more natural and serve to remind us to think about the tis-
sive quantity 17 can describe a piece of tissue of any size, and it sue, not the image. Slices of voxels are inside the object, whilst
does not alter as the tissue is subdivided (assuming it is uni- surfaces of pixels are outside the object. Some imaging proce-
form). Examples are density, temperature, colour, concentration, dures will use very small pixels (‘in-plane’ resolution) yet set a
magnetisation, membrane permeability, capillary blood volume large slice thickness (in order to retain signal-to-noise ratio).
and perfusion per unit volume of tissue, texture and the MR An extreme example would be a voxel of size 0.7 × 0.7 × 5 mm,
parameters proton density, T1, T2, the diffusion coefficient of a which appears to have the ability to resolve small structures,
liquid and magnetisation transfer. An extensive quantity refers yet any structures that are not oriented nearly perpendicular
to a piece of tissue as a whole, and subdivision reduces (or at least to the slice plane would be blurred by the large slice thickness.
changes) the value of the quantity. Examples are mass, volume, In this case the voxel would be shaped like a matchstick (i.e.
shape and total blood supply to an organ. have a large aspect ratio); a more appropriate voxel size might
Some intensive quantities, such as metabolite concentration, be 1.5 × 1.5 × 2.2 mm, which has the same volume (and hence
local blood flow or local permeability, can be expressed either per signal-to-noise ratio, for a given imaging time) but is more
unit mass of tissue or per unit volume of tissue. Traditionally, phys- likely to resolve small structures. Three-dimensional imaging
iologists have used the former system, since the mass of a piece of sequences can give us voxels which are isotropic (i.e. have the
excised tissue is more easily determined that its volume. In qMRI, same dimensions in all three directions).
where the volume of each voxel is well defined, the latter system Structures in the brain have very fine detail and very often there
is more natural. Conversion from per mass to per volume can be are two (or more) types of tissue inside the voxel. The resulting
achieved by multiplying by the density of brain18 (1.04 g ml−1 or NMR signal from this voxel is simply a combination, or weighted
1040 kg m−3 for both white and grey matter) (Whittall et al., 1997). average, of what each individual tissue would give if it filled the
whole voxel. Thus, if we are trying to measure the T1 of grey matter,
near to cerebrospinal fluid (CSF) the value measured will be some-
1.3.1 Images, partial volume and maps
where between that of pure grey matter and pure CSF, depending
Images and maps are terms used to mean different things. on the relative proportions of brain tissue and CSF in the voxel.
An image is produced by the MRI scanner and has an intensity19 This is called the partial volume effect and is a major source of error
that depends on a variety of parameters, including some that when making measurements in brain tissue at locations near to
describe the tissue (e.g. PD, T1, T2, and combinations of these) boundaries with other tissue types. The value measured in the tis-
and some that are characteristic of the scanner (e.g. the scanner sue is altered by its proximity to another tissue, and the determi-
transmit flip angle and receiver gain). The image consists of a nation of boundaries and of volumes is brought into error. Partial
two-dimensional (2D) matrix of numbers stored in a computer volume errors can be reduced by using smaller voxels, although
(often part of a three-dimensional (3D) image data set). Each the price paid is that of a worsening of the signal-to-noise ratio. An
location in the matrix is called a pixel (‘picture element’), which inversion pulse before data collection can remove signal from a tis-
is typically square and 1–2 mm wide. The image data come from sue with a particular T1 value (as in the FLAIR and STIR sequences,
a slice of brain tissue that has been interrogated, or imaged. This which null the signal from CSF and fat, respectively).
slice has a specified thickness (usually 1–5 mm), and each pixel in A parametric map can be calculated from two (or more)
the image in fact derives from a cuboidal box-shaped piece of tis- images of the same piece of tissue. A simple example would be
sue, called a voxel. The first and second dimensions of the voxel to collect two images with differing amounts of T2 weighting.
The ratio of these two images then only depends on the tissue
17 Intensive (dictionary definition): of or relating to a physical property, mea- parameter T2 and is independent of scanner parameters (such
surement, etc., that is independent of mass; extensive: a property that is as transmitter or receiver settings). By calculating this ratio for
dependent on mass. each pixel, a third matrix, or map, can be formed, which has the
18 For example, the normal concentration of water in white matter is about
appearance of an image (brain structures can be identified) but
0.690 g water per g tissue (0.690 kg water per kg tissue), equivalent to 0.718 g
is conceptually different from an image, in that individual pixel
water per ml tissue (718 kg water per m3 tissue) (see first edition, p 91).
19 Often called the signal intensity, since it is proportional to the signal volt- values now have a numerical meaning (such as value of T2, in
age induced in the radiofrequency (RF) coil by precessing magnetisation milliseconds, at each location in the brain), rather than repre-
in that piece of tissue seen in that voxel of the image. senting signal intensity on an arbitrary scale.
Concepts: Measurement in MRI 9

1.3.2 Study design TABLE 1.2 Good practice in study design and statistics
1. Optimise the instrumental precision before starting the study.
Many studies set out to compare groups of subjects using
2. Talk to a statistician before and after collecting the data.
the classic double-blind randomised controlled trial design.
3. Collect interleaved control and patient data.
Typically, a new MR parameter will be evaluated in a par-
4. Control for age and gender during subject recruitment.
ticular disease by measuring it in a group of patients and 5. Inspect the data in scatter plots.
in a group of controls. The controls could be on placebo or 6. Model the data, including random and systematic error.
another (established) treatment. Other differences between 7. Adjust for age and gender during analysis.
the groups (‘confounding variables’) should be removed as 8. Avoid if possible doing t-tests with many comparisons.
much as possible, hence the need for age and gender matching. 9. Be aware that correlations are hard to interpret.
The scanning should be carried out at the same time, using 10. Give confidence limits on group means and differences.
interleaved controls, rather than leaving the controls until
the end of patient scanning (when a step change in the mea- results with a different group of patients. Methodological pit-
surement procedure could produce an artificial group differ- falls, as illustrated by existing published work, should be iden-
ence). Some patients may be on treatment which alters the tified before the study begins. Some errors (e.g. the presence of
MR results. Matching can be improved by dynamic matching, poor reproducibility, which would be detected with repeated
carried out as part of subject recruitment as the study pro- scanning, or scanning controls after an upgrade, not inter-
ceeds. Thus, if controls are in short supply, but patients plen- leaved with the patients) will irreversibly destroy the value of
tiful, then each time a control is recruited, a matched patient the data.
is selected from the available patients. In placebo-controlled Selection of MR parameters requires thought. To acquire
trials, allocation of a patient to the placebo or treated group all the parameters discussed in this book would require more
can be decided at the time of recruitment, to keep the groups time than can be fitted into one examination (although as
matched at all times. Double-blinding 20 is a powerful way of scanners get faster and techniques are optimised, acquisi-
reducing bias in treatment trials. The person giving the treat- tion times have come down). Parameters should be selected
ment, the person making the measurement, 21 and the patient according to the biological changes that are expected in the
are all blinded to whether they are receiving a genuine treat- particular disease being studied. Measuring several relevant
ment or a placebo (Table 1.2). parameters can be powerful (see Chapter 2, Section 2.2.2.5).
Inexperienced researchers should beware of ‘stamp collect- Mixed-parameters acquisition can address specific questions
ing’ when ‘interesting patients’ are studied, almost at random, (e.g. diffusion weighted spectroscopy, or MT prepared multi-
with no hypothesis or controls.22 To design high-quality inves- echo measurements. Multiparametric studies are addressed
tigations that will be accepted for publication by the best inter- further in Chapter 18. The excellent consensus review by
national journals, the investigator should be aware of what O’Connor et al., on imaging biomarkers elegantly summarises
work has already been published or presented at international many of the issues involved in quantification, and is worth
scientific conferences.23 studying in detail (O’Connor et al., 2017).
A literature search24 should be carried out. Studies should
not be replicated unless there is a case for confirming the
1.3.3 Usefulness of an MR parameter
From a clinical point of view, a potential new quantity to charac-
20 The double-blind design is not suitable for treatments where the practitio-
terise brain tissue can be evaluated by considering three factors:25
ner plays an essential part in the treatment. This is particularly relevant
in so-called alternative therapies (e.g. acupuncture, homeopathy, osteopa- Sensitivity: does the quantity alter with disease? Is the
thy, psychotherapy and reiki). Although a placebo cannot be given, dif-
false negative rate low?
ferent treatments can be compared. Even in conventional clinical trials,
the patient often guesses whether they have a placebo or not from the side Validity: is it relevant to the biological changes that are
effects, and also those with greater side effects may be more likely to drop taking place?
out. More research on methodology may be needed to find suitable study
designs to overcome these problems. Reliability: is it reproducible? Is the false positive rate
21 Ideally this includes both the radiographer making the scan and the low?
observer analysing the MR data.
22 The term hobby researcher describes this phenomenon well.
Thus, the concept of validity (which is absent from a judge-
23 See also Chapter 2, Section 2.2.2.1, on two kinds of studies: ‘fishing expedi-
ment based merely on accuracy and precision) enables the
tion’ and hypothesis-driven.
24 For example using PubMed, from the US National Library of Medicine, relevance of a metric to be considered. For example, intra-
available free of charge online http://www.ncbi.nlm.nih.gov. From here cranial volume could be measured very accurately and pre-
you can download PDF files of papers (provided you are logged on to an cisely but would be completely irrelevant in most situations.
academic website, e.g. a university). Usually you can ‘search forwards’,
that is see which papers have cited the paper you are looking at (the ‘cited
by’ list). Thus, a complete picture of publications on a particular topic can 25 See first edition, Chapter 12, discussion by the psychologist N Ramsey in
be built up quite quickly and conveniently. the context of fMRI.
10 Quantitative MRI of the Brain

160 data;26 and (5) the biological significance27 of MR parameters

140 influences what use can be made of qMRI (in particular, can
access to the current biology predict the future clinical status?).
120
Where might these improvements lead? How would we know
100
Sample size

when no more improvements are worthwhile? We should take

80 time out from the detailed improvements to consider the big-
60 ger picture.28 The concept of the ‘perfect qMRI machine’ (see
40 Section 1.4.2) might give a clue. A major improvement in qMRI
would come about if there were an international certification
20
scheme for qMRI measurements which have reached the level of
0 2 4 6 8 10 12 the perfect machine; a proposal is made below.
Instrumental SD

1.4.2 International standardisation
FIGURE 1.1 The effect of instrumental precision (ISD) on the power and certification
of a study and the required sample size. By reducing the ISD, the sample
size required is dramatically reduced, with a consequent saving in the Standards already exist for measurements in many physical
cost and duration of the study. This is a simulation based on group com- quantities. Readily available machines to measure voltage, body
parison between controls (parameter value mean = 100, SD = 3) and the or food weight, and temperature often come with a certificate
same number of patients (effect size = 5, SD = 4.25). Power P = 80%, conforming to the International Standards Organisation,29
significance level α = 0.05, using G*power3 which is an established soft- guaranteeing a particular performance in terms of total error.
ware that can be downloaded free of charge. The concept of the ‘perfect machine’ originated in the build-
ing of the 200-inch Palomar telescope in the USA in 1933; at the
An alternative viewpoint (closely related) is the set of four time it was the most perfect telescope that could be built.30 The
­psychometric properties often used to assess scores: acceptabil- concept can usefully be applied to an MRI machine used for
ity, reliability, validity and responsiveness (Hobart et al., 2000). quantification. Here it proposed that
The impact of poor reproducibility on the power of a study can
be dramatic (Figure 1.1). Appropriate methods for analysing A Perfect Quantitative MRI machine is one that, in
MR data are still under discussion. The clinical metrics are also making a measurement, contributes no significant extra
being scrutinised and redesigned (Fischer et al., 1999; Hobart variation to that which already exists from biological
et al., 2000). Developments in psychology may be ahead of variation.
those used in this field (Krummenauer and Doll, 2000).
Various grades of performance can be envisaged, depending on
the purpose the measurement. Comparison with normal variation
1.4 The Future of quantitative MRI will be the most demanding; comparison with variation within a
disease might also be appropriate, depending on the context, and
1.4.1 Technology and methodology would be less demanding. Here a proposal is made for three levels,
each with an appropriate medal31 (see Table 1.3).
Since the first edition of this book, MRI technology has advanced.
The standard field strength has moved from 1.5T to 3T, with 4.7T
and 7T machines becoming more common. Major manufacturers
offer little below 1.5T for brain imaging. Gradients have improved 26 The first edition contained four chapters on analysis, recognising that it
in both strength and speed, enabling fast 3D acquisition to become is at least as important as acquisition. Spatial registration, shape, texture,
volume, atrophy and histograms were considered.
standard. RF transmit coils have responded to the higher fre- 27 The first edition had a chapter on the biological significance of MR param-
quencies by including designs to increase uniformity and reduce eters in multiple sclerosis. Post-mortem studies of tissue can establish a rela-
SAR. RF receive arrays use multiple coils for improved SNR. The tionship between a qMR parameter (e.g. NAA) and a biological parameter
only downsides for qMRI are the need to measure the transmit (e.g. neuronal density). Often the relationships are complex and depend on
field B1+ and the loss of the reciprocity principle (see Chapter 2). which biological parameters can vary (i.e. the disease context). Spatial reg-
istration of the biological specimen and the MR image is crucial.
Methodologies have continued to advance; this edition has three 28 In Thomas Mann’s Death in Venice, the writer is on the Venice beach. He
new topics: advanced diffusion, multinuclear MRS and CEST. sees the detail, in the foreground: children constructing a sand castle. He
Pointers to the future are in Chapter 18. turns his gaze to the horizon, empty and infinite. What would it be to be a
qMRI has five principal aspects: (1) concepts have hardly measurement hero?
29 For example, ISO 17025 is the main ISO standard used by testing and cali-
altered since the first edition of this book, just become more
bration laboratories.
clear; (2) MRI physics above 1.5T is more complex, with the loss 30 Building the 200-inch telescope at Palomar, California, is described in the
of reciprocity; (3) technical advances continue to alter the envi- book The Perfect Machine: Building the Palomar Telescope.
ronments in which qMRI must be re-implemented; (4) analy- 31 Medals are proposed, inspired by the ISMRM use of medals to acknowl-

sis techniques make a crucial difference to the value of the MRI edge sponsorship at its annual scientific meeting.
Concepts: Measurement in MRI 11

TABLE 1.3 qMRI Medals for perfect machines: a proposal large (10%–20%) (Parkes et al., 2004) and it might not be dif-
Medal Target study Criterion ficult to build a perfect gold-medal ASL machine (i.e. one with
ISD < 3%). A second example might be in the context of a serial
Bronze Group comparison ISD < 0.3 GSDa
study in relapsing–remitting MS. The within-subject varia-
Silver Multicentre study BCSD < GSDb
tion in lesion load is highly variable, and perfect gold-medal
Gold Serial study ISD < 0.3 WSSDc
machines for lesion volume already exist.
Note: SD = standard deviation; BSD = biological SD; GSD =
Who might administer such a scheme? Award of medals
group SD; ISD = instrumental SD; ICSD = inter-centre SD;
BCSD = between-centre SD; WSSD = within-subject SD. might be determined by the reviewers of a paper submitted
a In a group comparison, within-group variation GSD2 should to a journal claiming the status or by an international com-
dominate (i.e. machine variation ISD makes an insignificant contri­ mittee (perhaps sponsored by the ISMRM). Prizes could be
bution to total within-group variation). awarded (a kind of modern-day John Harrison longitude
b The effect of between-centre variation (BCSD) should be less
prize).32
than within-group variation.
c In a serial study, total within-subject variation WSSD2 should The closing words from the first edition are still true:
dominate (i.e. machine variation ISD makes an insignificant contri­
bution to total within-subject variation). Progress towards such automation [of measurement tech-
niques] will take time, and the persistence of John Harrison
the clockmaker may enable us to put our work into its his-
torical perspective. We are present at a true technological
Bronze medal: In a group comparison, the total variance in
revolution which is exposing our inner biological workings
each group determines the power and sample size needed (see
in ever increasing detail. A few decades ago this was incon-
Figure 1.1). This is the sum of the variance from genuine biological ceivable; in a few decades’ time the techniques will be as
spread (characterised by an SD equal to biological SD, BSD) and routine as measuring the mass of the body.
that given by the imperfect machine (characterised by an SD equal
to instrumental SD, ISD), that is total group variance GSD2 =
BSD2 + ISD2. Thus, if ISD = 0.3 GSD, the contribution of machine
References
variance to the total variance is 9% and may be considered neg- Bakker CJ, de Graaf CN, van Dijk P. Derivation of quantitative
ligible. This concept allows a perfect machine for group compari- information in NMR imaging: a phantom study. Phys Med
sons to be specified (Table 1.3). The criteria for each MR parameter Biol 1984; 29(12): 1511–25.
would vary; some might be easy to achieve, others might need a Cady EB, Costello AM, Dawson MJ, Delpy DT, Hope PL,
long, sustained effort. The value of BSD would depend on the kind Reynolds EO, et al. Non-invasive investigation of cerebral
of subjects considered; in pooling normal values, a correction for metabolism in newborn infants by phosphorus nuclear mag-
age and gender dependence should be applied (treating them as a netic resonance spectroscopy. Lancet 1983; 1(8333): 1059–62.
confounding variable and standardising all values to a fixed age Filippi M, Dousset V, McFarland HF, Miller DH, Grossman RI.
and gender). The estimates of SD have associated uncertainties, Role of magnetic resonance imaging in the diagnosis and
which are significant if the sample size is small (see Chapter 3, monitoring of multiple sclerosis: consensus report of the
Equation 3.2 and Figure 3.6), and these would need to be taken White Matter Study Group. J Magn Reson Imaging 2002;
into account when considering if a criterion had been reached. 15(5): 499–504.
An example might be the MTR results reported in Chapter 3 Filippi M, Grossman RI. MRI techniques to monitor MS evo-
(Figure 3.8). The stable scanner gave a normal group SD GSD = lution: the present and the future. Neurology 2002; 58(8):
0.4 pu and a measured instrumental SD ISD = 0.15 pu. From 1147–53.
these, ISD = 0.375 GSD, and the criterion in Table 1.3 (ISD < 0.3 Fischer JS, Rudick RA, Cutter GR, Reingold SC. The multiple
GSD) is not quite satisfied. sclerosis functional composite measure (MSFC): an inte-
Silver medal: In multicentre studies, inter-centre variation has grated approach to MS clinical outcome assessment.
to be controlled, although some differences can be absorbed by National MS Society Clinical Outcomes Assessment Task
the statistical analysis (provided each subject is always imaged Force. Mult Scler 1999; 5(4): 244–50.
at the same centre). MTR histogram matching using body-coil Gribbin J. Science: a history 1543–2001. London, UK: Penguin;
transmission (Tofts et al., 2006) is probably a perfect silver- 2003.
medal MTR machine. Hobart J, Freeman J, Thompson A. Kurtzke scales revisited: the
Gold medal: In a serial study, instrumental variation can application of psychometric methods to clinical intuition.
hide subtle within-subject biological changes. The power of a Brain 2000; 123(Pt 5): 1027–40.
serial study can be limited by such biological variation; often Krummenauer F, Doll G. Statistical methods for the comparison
it is small and unknown and may be extremely hard to mea- of measurements derived from orthodontic imaging. Eur J
sure. Gold medals will be the hardest to obtain; for some MR Orthod 2000; 22(3): 257–69.
parameters the gold medal may be impossible. An exception
is cerebral blood perfusion, measured by arterial spin label-
ling (ASL; Chapter 16). The natural within-subject variation is 32 Section 1.2.2.
12 Quantitative MRI of the Brain

Mallik S, Samson RS, Wheeler-Kingshott CA, Miller DH. Tofts PS, du Boulay EP. Towards quantitative measurements
Imaging outcomes for trials of remyelination in multiple of relaxation times and other parameters in the brain.
sclerosis. J Neurol Neurosurg Psychiatry 2014; 85(12): Neuroradiology 1990; 32(5): 407–15.
1396–404. Tofts PS, Steens SC, Cercignani M, Admiraal-Behloul F, Hofman PA,
McFarland HF, Barkhof F, Antel J, Miller DH. The role of MRI van Osch MJ, et al. Sources of variation in multi-­centre brain
as a surrogate outcome measure in multiple sclerosis. Mult MTR histogram studies: body-coil transmission eliminates
Scler 2002; 8(1): 40–51. inter-centre differences. Magma 2006; 19(4): 209–22.
McGowan JC. On the use of quantitative MR imaging. AJNR Tofts PS, Wicks DA, Barker GJ. The MRI measurement of NMR
Am J Neuroradiol 2001; 22(8): 1451–2. and physiological parameters in tissue to study disease
Miller DH. MRI monitoring of MS in clinical trials. Clin Neurol process. Prog Clin Biol Res 1991a; 363: 313–25.
Neurosurg 2002; 104(3): 236–43. Tofts PS, Wicks DAG, Barker GJ. The MRI measurement of NMR
O’Connor JP, Aboagye EO, Adams JE, Aerts HJ, Barrington SF, and physiological parameters in tissue to study disease
Beer AJ, et al. Imaging biomarker roadmap for cancer process. In: Ortendahl DA, Llacer J, editors. Information
studies. Nat Rev Clin Oncol 2017; 14(3): 169–86. Processing in Medical Imaging. New Jersey, Wiley-Liss Inc,
Parkes LM, Rashid W, Chard DT, Tofts PS. Normal cerebral per- 1991b; p. 313–25.
fusion measurements using arterial spin labeling: repro- Whittall KP, MacKay AL, Graeb DA, Nugent RA, Li DK,
ducibility, stability, and age and gender effects. Magn Paty DW. In vivo measurement of T2 distributions and
Reson Med 2004; 51(4): 736–43. water contents in normal human brain. Magn Reson Med
Sobel D. Longitude. New York: Harper Perennial; 2005. 1997; 37(1): 34–43.
Sormani MP, Bonzano L, Roccatagliata L, De Stefano N. Wray S, Tofts PS. Direct in vivo measurement of absolute metab-
Magnetic resonance imaging as surrogate for clinical end- olite concentrations using 31P nuclear magnetic resonance
points in multiple sclerosis: data on novel oral drugs. Mult spectroscopy. Biochim Biophys Acta 1986; 886(3): 399–405.
Scler 2011; 17(5): 630–3.
Tofts PS. Quantitative MRI of the brain: measuring changes
caused by disease. New York: Wiley; 2003.
 2
Measurement Process:
MR Data Collection and
Image Analysis

Contents
2.1 Magnetic resonance data collection........................................................................................ 13
Subject positioning and the prescan procedure • The NMR signal • The static magnetic
field B0 • Static field gradients • Radiofrequency transmit field B1+ • Slice and slab
profile • B1+ transmit field mapping • B1− receive sensitivity field • Image noise • The reciprocity
principle and its failure • Non-Uniformity correction • Scanner stability
Paul S. Tofts 2.2 Image analysis, statistics and classification............................................................................22
Brighton and Sussex Types of image analysis • Types of statistical analysis
Medical School References................................................................................................................................................28

2.1 Magnetic resonance then the amount of movement that took place is available as
output from the program. Research on both why some subjects
data collection move and on what limits how long a subject can stay in the scan-
The process of collecting magnetic resonance (MR) data from ner would improve the quality of MR data that can be obtained.
a subject, in the form of images, spectra or maps, is analysed in It may become possible to use fast MR (or optical) imaging to
some detail. dynamically alter the slice positions, tracking the movement of
a subject in real time (although movement to a location of dif-
ferent static or RF field value would require some sophisticated
2.1.1 Subject positioning and the
correction). If Gd contrast agent is to be injected, a line is placed
prescan procedure
into the subject’s arm, so that injection is carried out without
The subject is positioned on the scanner couch by the radiogra- disturbing the positioning of the subject. A power injector is
pher (technologist). The subject should be comfortable, to reduce usually used to provide a consistent injection procedure, with
movement during the scan as much as possible. The radiogra- synchronisation to the scanner.
pher should use any insight into the subject’s emotional state After the subject has been placed in the magnet bore, the
to reduce anxiety if necessary; preparation on a separate couch automatic prescan procedure generally includes the following
may be helpful. A cushion under the knees can reduce cramp. steps, which take account of differences between subjects and
Occasionally it is desirable to place the subject prone. Prone are crucial to quantification. The receiver gain is adjusted to use
positioning of the head may be more comfortable if support is the available dynamic range of the receiver channel, without
provided for the forehead and cheekbones, leaving a gap for overloading it. The gain must be fixed for subsequent scans, if
the nose, as used in a massage table. Movement of the body can image intensity values are to be combined in some way (e.g. for
cause a head movement; a nasal positioning device (Tofts et al., a dynamic Gd scan series, where images are collected at a range
1990) can help cooperative subjects to keep still. Some patients of time points after injection of contrast agent – see Chapter 14).
will find it hard to keep still because of their disease; researchers The transmitter output is adjusted to give the desired flip angle
involved in the study are usually motivated to keep very still. (FA) in the subject. This can be carried out in a number of ways;
Some kinds of movement are very common, especially rotation ideally, only the signal from the relevant piece of tissue (e.g. a
in the sagittal plane (‘nodding’). Movement can be monitored by slice or a spectroscopic voxel) is optimised. A multislice or vol-
repeated localiser images throughout the study. If spatial regis- ume acquisition cannot have the correct FA at all locations,
tration between different image datasets is used (see Chapter 17), because of transmit field non-uniformity.

13
14 Quantitative MRI of the Brain

The pulse sequences, containing long lists of radiofrequency terms, it says that if we have trouble getting the applied B1 field
(RF) and magnetic field gradient pulses, are then run. Signals into a particular location in the sample, using a particular coil,
are recorded; localisation of the origin of signal is achieved we will have as much trouble getting the signal out of that loca-
using a combination of slice or slab selection, frequency encod- tion, using the same coil. This is discussed in more detail in
ing and phase encoding gradients. Images can be weighted by Section 2.1.10 below.
various parameters (e.g. T1, T2 or D – see the chapters on each The dependence of magnetisation on absolute temperature
MR parameter). Images are reconstructed using Fourier trans- is relevant when room temperature concentration standards
formation; the magnitude of the complex data is usually calcu- are used (as in measurements of proton density and metabo-
lated (this is not vulnerable to unpredictable phase shifts). Full lite concentrations). As a particular concentration of protons
descriptions of the MR imaging process are available elsewhere is cooled (e.g. from body temperature to room temperature),
(Brown et al., 2014, and also see Table 1.1 in Chapter 1). its magnetisation increases and it can produce more signal
(see Chapter 3).
2.1.2 The NMR signal
2.1.3 The static magnetic field B0
The signal δv from precessing nuclei in a small volume δVs in
the sample is given by the following (Hoult and Richards 1976; In a superconducting magnet, the value of the static field is set
Hoult 1978): at the time of installation by adjusting the amount of circulating
current stored in the windings. There may be a very small decay
δυ = ω 0 B1xy M xy δVs cos(ω 0t ) (2.1) over time, which is compensated for by adjusting the current
through room temperature windings or by adjusting the centre
where ω0 is the Larmor1 frequency (in radians s−1); the life of frequency of the transmitter.
Sir Joseph Larmor, the Irish physicist, is described by Tubridy When the subject is placed in the magnet, the magnetic sus-
and McKinstry (2000). B1xy is the component 2 of the RF field B1 ceptibility of the tissue alters the field inside the brain slightly.
produced in the transverse plane at the location of the sample The transmitter centre frequency is adjusted to bring the pro-
by unit current in the coil, during transmission.3 M xy is the tons back onto resonance. The shim coil currents are adjusted
transverse component of the magnetisation of the sample.4 For to obtain a spatially uniform B0 distribution, as far as possible.
protons the equilibrium magnetisation M0 is as follows (Brown Remaining static field gradients caused by spatially varying
et al., 2014): tissue susceptibility (particularly near tissue–air interfaces, such
as the temporal lobes) can be a problem, particularly for spec-
N γ  2 B0 troscopy and echo planar imaging, which are very sensitive to
M0 = (2.2) such gradients. In spectroscopy the line position will be altered
4 kT
and possibly broadened. In gradient echo and echo planar imag-
ing there may be signal dropout due to intravoxel dephasing.7 In
where N is the number of protons per unit volume, γ is the mag-
spin echoes, the dephasing effect of these gradients is corrected
netogyric ratio,5 ħ = h/2π, where h is Planck’s constant, B0 is the
provided the spins are stationary; however, in the presence of
magnitude of the main static magnetic field, k is Boltzmann’s
diffusion, spins moving through a gradient will not be rephased
constant and T is the absolute temperature6 of the sample.
and signal loss will once again be seen. Such signal loss will not
The proportionality of received signal with the magnitude
normally cause systematic error in quantification, although the
of the applied field per unit current, shown in Equation 2.1, is
lowered signal-to-noise ratio (SNR) will give increased random
called the principle of reciprocity (Hoult and Richards 1976),
errors, and in situations where the absolute signal level is impor-
and this has been a key concept in quantitative MR. In simple
tant (e.g. PD) there will also be a systematic error.
1 The Larmor frequency is the frequency at which protons precess around
A further source of degradation is that echo planar images
the main static field B0. (and to a lesser extent gradient echo images, which have a much
2 A linear coil produces two counter-rotating components; one is in the shorter echo time than echo planar sequences) will suffer geo-
right direction for NMR and is useful; the other is not used but contrib- metric distortion, such that the image is shifted or warped in the
utes to noise and power requirements. In a circularly polarised coil only locality of susceptibility gradients (Moerland et al., 1995; Hutton
the useful component is produced and detected.
et al., 2002; Jezzard 2002). This in turn prevents straightforward
3 This equation is valid for the simple case of a single transmit/receive coil at
low field. In this case the transmit field B1+ equals the receive field B1− (i.e. spatial registration of such images with those having negligible
B1+ = B1− = B1xy); see also Section 2.1.10. distortion (principally those that are spin-echo–based, although
4 After a single 90o RF pulse, M xy = M0.
5 γ = ω0/B0, where B0 is the static magnetic field strength, in Tesla. For pro-
tons, γ = 2.675 108 rad s−1 T−1 (equivalent to 42.57 MHz/T) (Brown et al., 7 In intravoxel dephasing, the different components of magnetisation in
2014). Greek letters are described in Appendix 1. a voxel, experiencing different static fields, become out of phase with
6 The absolute temperature is measured in degrees Kelvin (K) from –273oC, each other, and the total transverse magnetisation vector in the voxel is
which is called absolute zero. Thus the freezing point of water (0oC) is reduced. In a spin echo, this dephasing is corrected by the 180o refocusing
273K, and body temperature (37oC) is 310K. pulse; in a gradient echo the uncorrected dephasing leads to signal loss.
Measurement Process: MR Data Collection and Image Analysis 15

gradient echo sequences often also have negligible distortion) a constant RF field of that value.9 The current in the transmit
and thwarts any attempt at measuring volume. The image inten- coil required to achieve this value of RF field depends on the
sity is likely to be altered by distortion (since a given amount of Q10 of the coil. Q is determined mostly by power losses in the
signal will be placed into a voxel that is too large or too small). subject (caused mostly by its electrical conductivity) rather than
A third degradation is that off-resonance effects in such locali- by losses in the coil itself. As the subject is moved into the coil,
ties may reduce the apparent FA and distort a 2D slice selection its conductivity loads the coil, and a greater current, and hence
process (see Section 2.1.6). voltage, is required to produce a given B1 value. The amount of
The static field can be mapped straightforwardly using coil loading (i.e. the amount of power that is removed from the
the phase shift after a gradient echo (Sled and Pike 2000; coil and is deposited in the subject) varies from subject to sub-
Hetherington et al., 2006). ject. The prescan procedure sets out to obtain the same value
of B1 (and hence FA), regardless of the loading produced by the
particular subject. The transmitter output is adjusted, usually
2.1.4 Static field gradients
automatically, often by adjusting attenuators in the amplifier.
Having taken a lot of care to achieve a uniform static magnetic Non-linearity in the transmitter output stage may occur, lead-
field, switched field gradients8 are deliberately introduced as ing to incorrect B1+ values and distortion of selective pulses. Such
part of the imaging process. The slew rates are very fast, giv- gross non-linearity over the normal range of amplitudes of the
ing typical switching times of <100 μs. Eddy currents can be selective pulses would probably be picked up as artefacts in the
induced in surrounding conducting structures; these have the routine imaging (depending on the amplitudes of the selective
effect of producing small transient shifts in B0, distorting spectra pulses used). Calibration of the transmitter output stage is usu-
and images. Eddy currents are reduced to low levels by several ally carried out periodically as part of routine preventive main-
devices. Actively shielded gradient coils limit the magnetic flux tenance (Venkatesan et al., 1998). Transmitter linearity can be
outside the coil; current pre-emphasis circuits drive the coils in investigated as follows: An oscilloscope can be used to measure
such a way as to counteract the effects of the eddy currents; and the output voltage as a function of software hard pulse ampli-
conduction loops are eliminated from the scanner bore con- tude (Alecci et al., 2001); this will be accurate to within a few per-
struction materials. cent, depending on the oscilloscope. The output can be stepped
The remaining non-idealities are twofold. Firstly, the gradi- by the software for convenience. A more accurate method is
ent amplitudes may be incorrect, by up to about 1% (depending to use NMR to measure the B1+ amplitude, as follows. At each
on the calibration procedure). This gives rise to small errors in amplitude, observe the signal from a small sample as a function
the size of objects (since the gradient change corresponds to a of hard pulse duration. The null duration (i.e. a 180o pulse) gives
change in magnification, or of voxel size) and also to errors in an accurate and precise measurement of B1+. Ensure there is no
the estimates of diffusion coefficients and tensors. Very pre- pulse droop, using an oscilloscope. A plot of B1+ versus software
cise measurements of voxel dimensions, using image registra- pulse amplitude should be linear.
tion, enable the value of small gradient changes to be measured RF non-uniformity is the largest cause of error in qMR.
(Lemieux and Barker 1998). Secondly, gradient coils do not pro-
duce a completely linear variation of static field with distance Radiofrequency field inhomogeneities are the most
(i.e. the gradients are non-uniform); this in turn produces errors irksome sources of nonidealities (especially as a
in the gradient amplitude (according to the position) and gives result of their omnipresence). The spatial variation
rise to spatial distortion (Moerland et al., 1995; Jezzard 2002). of the RF field sensitivity of the transmit and receive
Non-linearity from gradient coils is minimal in the central (head) coils enter the signal expression for any sequence in
region when using body gradient coils. Manufacturers usually the form of altering the flip angle at a given spatial
measure and correct for the geometric distortion caused by non- location as well as altering the received signal from
linearity; this can be seen by turning the correction off. [the same] spatial location. (Haacke et al., 1999 p. 661)

At 1.5T the effect is noticeable (Barker et al., 1998); at higher

2.1.5 Radiofrequency transmit field B1+
fields the problem becomes worse (see Figure 2.1). An elliptical
Usually transmission is by the body coil, which has relatively object (such as the head) in a circularly polarised coil gives a
good uniformity over the head region. Typical pulse amplitudes diagonal non-uniformity pattern (Sled and Pike 1998). At 3T
are 10–20 μT. Sometimes amplitudes are expressed in hz or using a birdcage coil a 20% reduction in B1+ was measured at the
radians s−1, giving the rate of nutation of magnetisation around
9 The nutation rate is ω1 = γB1 rad s−1; thus a RF field of 10 μT corresponds to
2680 rad s−1 or 426 hz. The duration of a hard (i.e. non-selective) θ pulse is
τθ = θ/(γB1); thus a 90o 10 μT hard pulse lasts 587 us.
10 Q stands for quality factor, denoting how long a coil will ring after being

8 Imaging gradients can be up to about 20 mT m−1; diffusion gradients are excited. High-Q coils are less damped, lose less energy per cycle, ring for
often higher (to shorten the echo time), up to 80 mT m−1, and may use a longer, provide a greater B1 for a given current and provide a greater signal
dedicated head coil set. Switching (slew) rates are up to 200 T m−1 s−1. for a given amount of precessing magnetisation.
16 Quantitative MRI of the Brain

VB1+ SI phase, allows uniformity to be optimised (Ibrahim et al., 2001).

This has been termed RF shimming (Collins et al., 2005).
In the early days of MRI, it was feared that the reduced RF
penetration at high frequencies (the ‘skin effect’) (Bottomley
and Andrew 1978) caused by the electrical conductivity of the
64 MHz

tissue would prevent head imaging above 20 MHz. In fact it is

more than offset by the amplifying effect of dielectric resonance
(which increases B1+ in objects whose size is comparable with the
half-wavelength of electromagnetic waves at the frequency of
observation; see Figure 2.1).

2.1.6 Slice and slab profile

175 MHz

In 2D (slice selective) imaging, slice selection is a key prob-

lem. The observed transverse magnetisation is the sum of spins
within the voxel that have experienced a variety of histories,
according to their location in the slice selective gradient, the
local FA and the amount of relaxation (i.e. T1/TR). Thus the slice
profile can be distorted; there is not a single effective FA within
a voxel, and signal modelling for quantification is often com-
plex and inaccurate (Parker et al., 2001).11 Slice selection is often
used for an EPI readout and also is needed for some MR param-
eters (e.g. T2, T2* ); then such effects must be taken into account.
260 MHz

Magnetisation preparation by a hard (non-selective) pulse fol-

lowed by 2D readout can mitigate the distortion.
Fast and powerful gradients have driven the widespread use
of 3D imaging, which has no such problems. Slab selection (in
which signal from outside the desired field of view is suppressed)
is usually applied in each phase encoding direction, to prevent
wrap-around.12
345 MHz

2.1.7 B1+ transmit field mapping

The determination of the RF ‘active’ field B1+ at each location in
space is important for two reasons. First, the local FA can then
be found; this is needed for many parameter calculations (e.g. T1
from VFA Chapter 513). Second, its value is needed for accurate
FIGURE 2.1 RF signal non-uniformity in the head. An accurate measurement of some MR parameters (e.g. MT).
mathematical model of the head inside a birdcage coil, with flip angle = Early methods of determining FA, or setting it to a required
90o at the head centre. Frequencies correspond to fields of 1.5T, 4.1T, 6.1T value, were developed in spectroscopy. A sample inside a uni-
and 8.1T. VB1+ is the excitation field (normalised by the factor V), SI is form RF coil gives a maximum signal when the FA is 90o, pro-
the signal intensity from a gradient echo sequence. The doming effect vided it can relax fully between pulses (i.e. TR >> T1). The pulse
at the centre of the head becomes increasingly pronounced at higher amplitude,14 or duration, is increased from a low value; at first
fields, and an annular region of reduced signal is visible further from
the centre. (From Collins, C.M., and Smith, M.B., Magn. Reson. Med.,
45(4), 684–691, 2001.) 11 Slice distortion and correction is described and illustrated in more detail
in the first edition.
12 Slab selection is a more gentle process than slice selection (each voxel has
periphery of the brain, compared to its value at the centre (Alecci a single well-defined FA) and probably produces little distortion; however,
et al., 2001). Anatomically accurate models of B1+ distribution in there seems to be no literature on this.
the head, using detailed anatomical knowledge (Collins and 13 Chapter 5, Section 5.4.4 also contains a discussion of B + mapping.
1
14 A rectangular (‘hard’) pulse, of amplitude B and duration τ, produces a
Smith 2001; Ibrahim et al., 2001) show that at high fields, up to 1

8T, non-uniformity increases, as dielectric resonance increases FA of γ B1 τ. Depending on the spectrometer hardware, either the pulse
amplitude or duration is varied to give the required FA. Early spectrom-
the sensitivity near the centre of the head; measurements at 7T
eters had fixed amplitude, and the duration was altered. Modern MRI
confirm this (Collins et al., 2002). Adjustment of the current machines often also allow variable amplitude (since a selective [‘soft’]
applied through each port on a multiple transmit coil, and its pulse must keep its duration fixed).
Measurement Process: MR Data Collection and Image Analysis 17

the signal increases almost linearly, then reaches a maximum, slice selection artefacts; (7) have acceptable SAR value; (8) work
then declines. Further increase in the pulse gives a null signal over a wide FA range if needed;17 and (9) ideally be capable of
(corresponding to an FA of 180o), and this condition can often be implementation using a standard pulse sequence.
found more precisely than the maximum at 90o. Early work often used the double-angle method (Stollberger
Modern methods (summarised in Table 2.1) should meet the and Wach 1996). Two 2D acquisitions are made, with nominal
following nine criteria: the method should (1) provide an accu- FA values typically 30o and 60o; the ratio of signals gives the
rate value, to within about 1% (since some measurements are actual FA values but only under the condition of TR >> T1 (com-
very sensitive to FA errors, e.g. T115); (2) have good precision;16 plete relaxation). The needs for speed and 3D acquisition may
(3) have reasonable imaging time (ideally less than 1 minute; have pushed this method aside. However, an EPI variant pro-
this favours 3D methods); (4) be independent of T1 effects; (5) be vides a 2-minute solution using a standard pulse sequence, and a
independent of B0 (off-resonance) effects; (6) be unaffected by 2D performance comparable to the AFI and BS methods (Boudreau
et al., 2017) (Figure 2.2).
The two principle methods currently in use are actual flip
TABLE 2.1 B1+ mapping methods (a selection)
angle imaging (AFI) and Bloch–Siegert (BS) (Sacolick et al.,
Reference Summary 2010 2011; Whisenant et al., 2016). AFI (Yarnykh 2007) uses
Magnitude Methods Yarnykh (2007) Two pulses, same FA, two two pulses with the same FA, at different (short) TRs, in a 3D
TRs (3D) sequence. There are residual T1 effects, which could be measured
Actual flip angle Hurley et al. (2012) Combined with T1 using VFA (variable flip angle – see Chapter 5). In VAFI (Hurley
imaging (AFI) measurement
et al., 2012), VFA and AFI are combined to take proper account
Dual angle method Stollberger and Wach Two pulses, two FAs, long
(DAM) (1996) TR (2D)
of T1. AFI can be seen as an ingenious variant on the dual angle
Insko and Bolinger Original paper
method. The ratio of signals from two acquisitions is again
(1993) used; however, it is the TR not the FA that is altered. In the BS
Cunningham et al. Saturate for short TR approach, an off-resonant pulse is used to produce a phase that is
(2006) proportional to the square of the local B1+ field; this is then read
Boudreau et al. (2017) EPI-DA; fast uses EPI out. It is fast and independent of B 0 error (Sacolick et al., 2010
180o null Dowell and Tofts Can use short TR 2011; Duan et al., 2013; Whisenant et al., 2016).
(2007) The 180o null method uses a standard 3D sequence to pro-
Phase Methods Sacolick et al. (2010) Phase α B12; 3D vide a map in 4 minutes – see Figure 2.3 (Dowell and Tofts
Bloch–Siegert (BS) Sacolick et al. (2011) 3 sec acquisition for 2007). The move to higher fields and the consequent increase
prescan
in B1+ ­non-uniformity has prompted MRI manufacturers to
Phase sensitive Morrell (2008) Wider range than DAM
invest in mapping techniques (Sacolick et al., 2010; Nehrke and
Note: FA, flip angle; EPI-DA, EPI-Double Angle. Börnert 2012). Summaries with discussion, comparisons and

T1W

AFI

EPI-DA

Slice No. 8 10 12 14 16 18 20

FIGURE 2.2 B1+ maps at 3T using the actual flip angle imaging and EPI-DA methods (see Table 2.1) (From Boudreau, M., et al., J. Magn. Reson.
Imaging, 2017).

15 A 1% error in FA gives a 2% error in T1; see Chapter 5, Equation 5.10.

16 The B1 function will vary slowly with space, so some spatial smoothing is 17 Although some methods presume a small range of FA values, they can
permissible. usually be adapted to a wider range by adding more measurement pulses.
18 Quantitative MRI of the Brain

TR = 10T1 Thus parallel imaging should be used with caution, with low
TR = T1 acceleration factors.
1.0 B1− cannot be determined directly and with the loss of the
TR = 0.1T1
reciprocity principle22 has become one of MRI’s ‘unsolved problems’.
0.5 For most MR parameters it is a ratio of signals that is measured,
and absolute sensitivity has no influence. However when measuring
absolute concentrations of protons (i.e. PD and magnetic resonance
Signal

0.0 spectroscopy (MRS) absolute metabolite concentrations), a map of

B1−(r) is required, and some attempts have been made to estimate it.
–0.5 The bias field approach uses information that B1−(r) varies
slowly with position (Volz et al., 2012; see also Watanabe et al.,
2011; Jin et al., 2012; Sabati and Maudsley 2013). A PD-weighted
–1.0 image has intensity proportional to B1−(r)PD(r); smoothing this
0 100 200 300 400 500 600
(fwhm = 60 mm) gives a map proportional to B1−(r) (this assumes
FA (degrees)
PD(r) has little low spatial frequency content). A single reference
value for PD (e.g. from cerebrospinal fluid or a mean value for
FIGURE 2.3 Simulation showing how the 180o signal null can be used
the whole brain) then enables B1−(r) to be determined and its
to map B1+, regardless of T1. (From Dowell, N.G., and Tofts, P.S., Magn.
effect removed in a PD(r) estimation. The method relies on PD(r)
Reson. Med., 58(3), 622–630, 2007.)
being smooth and there being no large abnormalities. A varia-
tion on this is to constrain the possible behaviour of B1−(r) by
often optimisation of sequences are given by Lutti et al. (2010),
using the Maxwell equations; the method has been successful in
Morrell and Schabel (2010), Sacolick et al. (2010), Volz et al.
the ordered environment of a single transmit/receive coil and a
(2010), Hurley et al. (2012), Park et al. (2013) and Pohmann and
phantom (Sbrizzi et al., 2014; see also Chapter 4, Section 4.3.2).
Scheffler (2013).18
The receiver gain can be altered during the prescan pro-
Inflow effects: In the special case of bulk blood flow, the effec-
cedure to account for the magnitude of the signal (the gain is
tive B1+ value may be smaller than that in the voxel of interest.
much reduced in spectroscopy). Ideally it will be fixed during
ASL19 uses this effect; blood arrives in the voxel with a different
the acquisition of the image series,23 at a suitable value that will
magnetisation from local water that has a recent history in the
not overload the receive chain nor introduce extra noise. If it
voxel. In DCE imaging,20 blood is often imaged with a 2D slice to
is altered, a correction might be possible, depending on what
measure its T1; however, blood from outside the slice may have a
information on receiver gain is available and whether analogue
lower FA and hence higher magnetisation than that predicted by
values of attenuation or gain are accurate.
a naive use of the FA value in the voxel.21

2.1.8 B1− receive sensitivity field 2.1.9 Image noise

Electrical noise comes from random thermal agitation
Reception is usually by a close-fitting multiple array head coil, for
(Brownian motion) in the subject, the RF coil and possibly
good sensitivity. The receive sensitivity field B1−(r) is d
­ etermined
the preamplifier. With good design, contributions from the
by both the receive coil array geometry and the properties of the
hardware are made insignificant and the dominant source is
object being imaged (whether a head or a phantom). Dielectric
the subject.
resonance tends to give a high B1− in the central parts of the
Artefacts (often from movement during the phase encod-
object (similar to the B1+ distribution), whilst multiple surface
ing process) constitute an additional source of unpredictable
coils have greater response from the periphery; these effects can
error; these can be assessed by viewing the air surrounding the
be made to partially cancel (Figure 2.4).
head. The centre of the grey level display window is set to zero.
In a multi-array receive system, parallel imaging options
Any visible artefacts are of interest, since they almost certainly
(Grappa, Sense, Smash, etc.) can produce image artefacts, aris-
extend to the high SNR parts of the image, in the brain, and may
ing from the reconstruction of low spatial frequencies, which
exceed the random noise.
might be unstable and degrade quantification performance.
2.1.9.1 Optimised sequence parameters
18 Teaching sessions at the annual International Society for Magnetic The effect of image noise can be reduced by careful choice of
Resonance in Medicine meeting by PF van de Moortele (2015) and sequence parameters. Increasing the voxel size reduces noise
Lawrence L Wald (2016) were also insightful. (at the expense of spatial resolution). Increasing the number of
19 See Chapter 16.

20 See Chapter 14, Section 14.4.3.

21 Blood flowing at 1 m s−1 into a 5 mm 2D slice will spend only 5 ms in the 22 See Section 2.1.10.
slice; thus it may experience only one pulse (if TR = 5 ms), insufficient to 23 For example, during acquisition of several TE values to determine T2
reach a new equilibrium magnetisation. (Chapter 6).
Measurement Process: MR Data Collection and Image Analysis 19

(a) (b) (c)

FIGURE 2.4 Simulated magnetic resonance images showing the results of different coil configurations at 300 MHz (7T). Use of a single volume
coil in both transmission and reception (a) results in relatively strong B1+ and B1− fields near the centre (due to constructive interference) sur-
rounded by weaker fields (due to destructive interference) resulting in a centre-bright appearance of the signal intensity distribution. Use of a single
volume coil in transmission but an array of decoupled coils in reception with sum-of-magnitude reconstruction (b) results in relatively strong B1+
near the centre but relatively strong B1− near the periphery, resulting in a more homogeneous signal intensity distribution. Use of a transmit array
with RF shimming to lessen the pattern of constructive and destructive interference in transmission and a receive array with sum-of-magnitude
reconstruction (c) produces a very homogeneous image, even at this high frequency with only eight elements in transmission and reception. (From
Collins, C.M., and Wang, Z., Magn. Reson. Med., 65(5), 1470–1482, 2011.)

averages (NEX – the number of excitations) reduces the noise Wheeler-Kingshott et al., 2002), and ASL perfusion, where the
(SNR α NEX 1/2). Increasing the TR often also increases the SNR. signal difference is comparable with the noise (Karlsen et al.,
Changing NEX or TR usually increases the acquisition time, and 1999). If image averaging is used to improve SNR, this should
for a fixed acquisition time an optimal combination of param- be carried out on the complex images, before formation of the
eters can be found, which maximises SNR. The acquisition of an magnitude.
image dataset from which parameters such as blood–brain per- Correcting single-coil image data is possible; however in
meability are to be estimated can also be optimised (Tofts 1996) multicoil receive systems correction is not straightforward and
(see Figure 2.5 and Chapter 3 section 3.2.3 on modelling error). additionally the noise may vary with position in the image (see
Chapter 17).
2.1.9.2 Rician noise distribution in magnitude
image gives systematic error 2.1.9.3 Noise estimation
Most images are constructed from the magnitude of the com- An estimate of the noise value in high SNR regions is often
plex image data, and phase information is discarded. Magnitude desirable (e.g. for modelling error propagation). For a single
data does not have a normal distribution (it cannot be negative). coil system, if the standard deviation in high SNR regions of
At low values of SNR, the distribution from a single receive coil the image is σ, then sampling the Rician noise in an air region
follows a Rician probability distribution with a non-zero mean of interest (ROI) will give a mean value of 1.25σ and a standard
value (Henkelman 1985, 1986; Brown et al., 2014). At high values deviation of 0.66σ (Edelstein et al., 1984; Gudbjartsson and Patz
of SNR this approximates to a Gaussian (normal) distribution; at 1995; Andersen 1996), enabling σ to be estimated (see Figure 2.6).
zero SNR it becomes a Rayleigh distribution (Figure 2.6). Noise has also been estimated by subtracting the squares of
This effect constitutes a systematic error, or bias, in addi- images (Sijbers et al., 1998)24.
tion to the random error that noise always contributes. It In a multicoil receive system, two approaches are possible.
can be seen most clearly by looking at the mean value in air (1) Placing an ROI in a uniform region of tissue will give an indi-
regions of the image, where a non-zero value will be found. cation of image noise; however the SD value will also contain
The techniques most affected are those that use low SNR image a contribution from any tissue non-uniformity in the ROI, and
data, principally T 2 and ADC, where the decay of signal may
be followed down into the noise, (Miller and Joseph, 1993; 24 See also the IPEM QA report (McRobbie 2017), page 21 onwards.
20 Quantitative MRI of the Brain

0.05

Minimum detectable [Gd] (mM) 0.04

Gradient echo
0.03
Spin echo

0.02

0.01

0 200 400 600 800 1000

TR (ms)

FIGURE 2.5 Sequence optimisation by noise modelling. Mathematical modelling of image noise propagation predicts the minimum amount
of Gd contrast agent that can be detected using a T1-weighted sequence. By optimising the repetition time, TR, in a spin echo or gradient echo
sequence, its performance in white matter can be optimised. Theory indicates that, for a spin echo, the optimum TR = T1 /2 (here it was assumed
T1 = 600 ms). The gradient echo (flip angle, FA = 50°) can achieve the same sensitivity, provided the correct TR is used. The examination time was
fixed at 10 min. (From Tofts, P.S., Magn. Reson. Imaging, 14(4), 373–380, 1996.)

the multiple coil system will in any case cause the noise value If histograms are to be produced from parameter maps, image
to vary with position. (2) Subtraction of two repeated images is despiking26 should be carried out, to prevent the discrete image
likely to give a better result. The difference image is expected to probability distribution from producing artefacts in the histo-
have a normal distribution with an SD of √2 σ; the distribution grams (Tozer and Tofts 2003).
of values, and their mean, should be checked for any unexpected
effects. This method has been used in phantoms (Murphy et al.,
2.1.10 The reciprocity principle and its failure
1993; Goerner and Clarke 2011). The noise in brain images is likely
to be different (because of different coil loading and B1− distribu- In the early days of MRI, the reciprocity theorem was key in
tion), and an explicit measurement in brain is preferred. Provided enabling B1− (r) to be found. (Hoult and Richards 1976; Hoult
movement is controlled, and measurements are restricted to uni- 1978, 2000). In this case B1− = B1+ = B1, and in this paradigm many
form areas of tissue, then this ought to be possible. papers described ‘B1 mapping’ (when in fact they are mapping B1+).
Reciprocity was demonstrated at fields up to 1.5T (Tofts and Wray
2.1.9.4 Image quantisation errors
1988; Michaelis et al., 1993; Barker et al., 1998; Fernandez-Seara
Image intensity values are usually stored as integers, typi- et al., 2001). Under varying values of coil loading, the product of
cally to 12-bit precision (i.e. 1 in 4096). The floating point pulse length27 or transmitter output voltage and signal is constant.
values from the image reconstruction process are rounded to Given B1−, absolute measurements of PD and MRS metabolite con-
the nearest integer.25 Typical signal values may be 500–1000; centration were possible (Provencher 1993, 2001; Fernandez-Seara
thus the electrical noise will be a few units (for SNR = 100), et al., 2001).28 Reciprocity served us well for 40 years.
and quantisation noise (maximum value 0.5) should be insig-
nificant compared to electrical noise. If the floating point
numbers are truncated (not rounded) a small amount of bias
26 The integers are converted to floating point numbers, and random noise
with maximum magnitude 0.5 is added, forcing the image intensity values
(0.5 image units) will be introduced. to have a continuous distribution, before calculation of the maps.
27 Data from Tofts and Wray 1988, reanalysed in the first edition, p 304.

28 The first edition of this book contains much on reciprocity (p 37) and its

25 Rounding introduces a maximum error of 0.5, with rms value 1 12  0.3 application in PD (p 95) and MRS (p 304–5).
 Measurement Process: MR Data Collection and Image Analysis 21

B1+ B1–

A
(in units of σ)

3
MAVE
SIcalc SIexp
2

MSD
1

0 5
1 2 3 4
A (in units of σ)

FIGURE 2.6 Rician noise in single-coil magnitude data. For a uni- FIGURE 2.7 Reciprocity failure at 7T. Theoretical B1 distributions in
form region of an image with real amplitude A (measured in units of a 16 cm spherical NaCl solution phantom at 7T, imaged with a 10 cm
σ, the standard deviation of the noise in each dimension of the normal surface coil. B1+ is the circularly polarised transmit field, B1− is the
distribution), the average value MAVE and the standard deviation MSD in receive field, and differences between the two can be seen. SIcalc and SIexp
the magnitude image are shown, both expressed in units of σ. Thus at are the theoretical (calculated) and measured (experimental) signal
high SNR (A >> σ), the mean intensity in the magnitude image equals its intensities for a gradient echo image; their similarity gives confidence
value in a real image (MAVE = A), whilst at low SNR it exceeds the value in the modelling. (From Collins, C.M., et al., Magn. Reson. Med., 47(5),
in a real image, reaching an asymptotic value of 1.253σ in the absence 1026–1028, 2002.)
of any signal. At high SNR, the standard deviation equals that in a real
image (MSD = σ), whilst in the absence of any signal it decreases to MSD =
0.655σ. (From Henkelman, R.M., Med. Phys., 12(2), 232–233, 1985.) The Distant Dipolar Field method may offer a solution to
­ easuring PD (Gutteridge 2002). This collects a small signal that
m
Later it became clear that the theorem is not valid at higher is proportional to PD2, and combined with a conventional PD
fields (3T and above),29 that is B1− ≠ B1+ (Sled and Pike 1998; Hoult image, the quotient image can give absolute PD without errors
2000; Ibrahim 2005; Collins and Wang 2011), and also that as from unknown B1 + or B1 − . The 1st edition has more discussion
transmit and receive coils improved, single transmit/receive and results on page 96.
coils would go out of use. Now in the ‘brave new technoworld’
of higher fields and separate transmit and receive coils, the reci-
procity theorem is probably consigned to the past, and we have 2.1.11 Non-Uniformity correction
to travel on without knowledge of B1− (Figure 2.7). Early workers found that images were often visibly non-­
PD and MRS measurements have become more difficult with uniform, and much effort was devoted to measuring and cor-
the loss of the principle of reciprocity. It is almost impossible recting non-uniformity.30 Image non-uniformity (NU) arises
to distinguish whether a change in signal voltage is caused by from three sources: (1) Transmit field (B1+) non-uniformity
a change in proton concentration or a change in coil sensitivity gives a magnetisation M xy NU that also depends on T1 (unless
(see also Chapter 4, Section 4.3.2 and Chapter 12, Section 12.6.5). a fully relaxed sequence is used). Receive (B1−) NU depends on
The solution may lie in using a dedicated ‘reciprocity-friendly both; (2) the receive coil characteristics; and on (3) the head
scanner’ that works at 1.5T or less, with a single transmit/ electromagnetic characteristics. Thus a simple image of a uni-
receive coil. Such a PD map could be matched to a higher quality form object, or simple smoothing, cannot correct for these
­proton-density weighted image from another scanner to provide three factors in a quantitative way. Transmit non-uniformity
improved spatial resolution and SNR. is usually minimised by using the body coil, and measured if
necessary (see Section 2.1.7).
29 As the wavelength of the B1 excitation decreases, the approximation of a
quasistatic excitation field becomes less valid (from Sled and Pike, 1998).
At 3T, the wavelength in water is 260 mm (see Chapter 3, Section 3.5.2). 30 See first edition, p 37.
22 Quantitative MRI of the Brain

% difference of consecutive points

6

–2

–4

–6

–8
0 300 600 900 1200 0 300 600 900 1200
Acquisition time (mins)
(a) (b)

16

14

12

10
%/pu

0
35 45 55 35 45 55
MTR (percentage units)
(c) (d)

FIGURE 2.8 Unsuspected scanner instability – an invisible problem. MTR histograms showed a large within-subject variation (c). Repeated
scanning of a phantom overnight showed large random variation (a). After changing transmitter boards, the scanner was stable (b) and MTR
histograms were reproducible (d). (Data from NG Dowell, originally presented in (Haynes et al., 2010) (From Haynes, B.I., et al., Measuring scan-
rescan reliability in quantitative brain imaging reveals instability in an apparently healthy imager and improves statistical power in a clinical study.
ISMRM annual scientific meeting, Stockholm, p. 2999, 2010.)

2.1.12 Scanner stability imaging and looking at the variation of the raw image data.
If transmitter instability is suspected, then imaging at the Ernst
Despite having followed all good practice, image data can be angle32 ought to produce a stable signal. If receiver instability is
unstable for unknown reasons. Repeated imaging of a phantom suspected, then slowly varying changes would affect the signals
may show a short-term variation in signal that is greater than at all FAs equally.
that predicted by image noise (Weisskoff 1996), in addition to
long-term drift (see Figure 2.8). Such instability can have two
major effects: (1) it degrades reproducibility (i.e. instrumen- 2.2 Image analysis, statistics
tal SD [ISD] – see Chapter 3, Section 3.3.2.1) and (2) in a DCE and classification
series acquisition, it introduces time-dependent variation, which
may mask a subtle signal enhancement caused by a genuine T1 2.2.1 Types of image analysis
change. Thus if a parameter (e.g. T1) shows an unexpectedly Here an introduction to image analysis concepts is given; a
large31 within-subject variation, it is worth carrying out repeated higher-level viewpoint is given in Chapter 17. There are three

31 The effect of image noise σ on repeated measurements of an ROI mean is 32 In a spoilt gradient echo, the signal is maximal and independent of FA θ
to give variation with a standard deviation equal to the standard error of at cos(θE) = exp(-TR/T1), where θE is the Ernst angle; e.g. for T1 = 800 ms,
the mean of the ROI (i.e. sem = σ/ √[no. of pixels]). TR = 10 ms, then θE = 9o.
Measurement Process: MR Data Collection and Image Analysis 23

main ways of extracting relevant image intensities from a set of Thus any conclusions about map values in the lesions would be
images that may cover many slices, several tissue parameters biased, since these values had been used to define which pixels
and many subjects: ROI’s, histograms and group mapping. would be included in the region. If serial measurements are made,
Before analysis, the image dataset may be spatially registered. fixed ROIs should be used for each time point, if conclusions are
Images from a single subject may be registered, to reduce the to be drawn about changes in a parameter value over time.35 Large
effects of movement during the examination. Images from differ- regions of most of the n ­ ormal-appearing white or grey matter
ent subjects may be registered to a standard space.33 Registration have been generated using T1 maps (Parkes and Tofts 2002) or
can produce subtle changes in image intensity that may thwart fractional diffusion anisotropy (Cercignani et al., 2001).
attempts at quantification (e.g. in a DCE series, small changes In studies of diffuse disease that affects large parts of the
in intensity can sometimes be observed that are caused by the brain, instead of creating large regions to study this, two other
registration process, not by changes in T1). approaches are available: histograms and voxel-based group
mapping.
2.2.1.1 Region of interest analysis
The study is focused on a particular part or parts of the brain, for 2.2.1.2 Histogram analysis
example visible lesions or large volumes of normal-appearing white A solution to the problem of ROI placement, and possible bias
matter, where intensities are to be measured. One or several ROIs arising from this process, is to test all of a tissue type (e.g. white
are drawn for each subject. Regions can be circular, oval, square, matter). This is particularly appropriate for diseases where the
rectangular or irregular. Regions may be defined in a single slice biological effects are diffuse and widespread. Histograms do
or extend over several slices (then the set is a volume of interest have the disadvantage that localisation information has been
[VOI]). They are often created using a semi-automatic technique, lost, and if disease only affects part of the region sensitivity will
which speeds up the process and improves the reproducibility. ROI be reduced by pooling data from the whole region.
size is a compromise between reducing noise (which favours large Histograms from different centres have sometimes varied; by
ROIs) and reducing partial volume error (which favours small standardising their generation, multicentre studies are possible
ROIs). Alternatively, if the image datasets are in standard (stereo- (Tofts et al., 2006). The bin width should be chosen to be small
tactic) space, then a number of standard VOIs will be available. The enough to capture any fine structure, yet large enough to not dis-
process of creating the ROIs takes some time to learn, and different play statistical fluctuations caused by a small number of pixels in
observers will develop different approaches. Inter-observer34 varia- the bin (typically 5 ms for T1). The bin should be labelled by its
tion can be reduced by carefully defining the procedure to be used. centre value (not the left- or right-hand edge). A normalised bin
This may include factors such as how the image is to be displayed, amplitude should be calculated as follows: find the percentage
and a detailed description of what anatomical cues are to be used in of the total pixels in that bin, then divide by the bin width. The
positioning the ROI. Usually the intra-observer variation is lower, total area under the histogram curve is then 100%, regardless of
and many studies accept that a single observer should be used to bin width (Figure 2.9).36
analyse the whole dataset. Even for a single observer, the analysis Histogram analysis by Principle Components Analysis (PCA)
should be repeated after a few days to ensure the reproducibility and linear discriminant analysis can be very powerful; MTR his-
is reasonable. A formal measurement of reproducibility can be tograms have predicted clinical score and also separated disease
undertaken (see Chapter 3, Section 3.3). subtypes (Dehmeshki et al., 2001, 2002b) – see Figure 2.10.
Unbiased ROI generation: There may be multiple MR images or
maps, for example conventional MRI (showing lesions in PD- or 2.2.1.3 Voxel-based group mapping –
T2-weighted images) and a MTR map. The appearance of lesions beyond rois and histograms
may be different on the conventional MRI and the map. If the Analysis of group-mapped images provides a way of combining
MTR values of lesions are to be measured and tested, the ROIs the ability of ROIs to be spatially specific (and thus sensitive)
should be defined on the conventional MR images (after spa- with the ability of histograms to be unbiased. In essence, a com-
tial registration), then transferred to the maps. If the ROIs were plete set of ROIs is automatically generated at all locations across
defined directly on the maps, the map intensity would influence the whole brain, without any bias in where they are located. The
where the ROI boundary was placed. ROIs tend to be attracted to image datasets are first spatially normalised to all lie in the same
locations of abnormal intensity (as a result of the process of their space. Appropriate statistical tests are then carried out on all the
creation, where an observer tends to draw around distinct objects). ROIs. Further information is in Chapter 17.

35 A recent example showed that, after treatment of tumours, regions of Gd

33 The most common is ‘MNI-space’ (MNI = Montreal Neurological enhancement unexpectedly retained the same mean intensity. On being
Institute). probed, the author disclosed that the ROIs had been drawn and redrawn
34 Inter-observer means ‘between observer’, that is the difference between on the Gd images and had become smaller after treatment. Thus fixed
measurements made by different observers on the same image data. ROIs would have shown a reduction in enhancement.
Intra-observer means ‘within observer’, that is the difference if the same 36 Examples are shown in the first edition, Chapter 18, along with examples

observer repeats the measurement. of classification of clinical data.

24 Quantitative MRI of the Brain

10 very expensive to acquire (both financially and in terms of the

effort by the patients and controls); this drives an imperative to
B (segD)
make the best possible use of clinical data by using appropriate
8 D (652)
analysis techniques (see Chapter 1, Table 1.2).
In this section the basic concepts relevant to statistical analy-
6 sis of MR data are summarised; however for full treatment the
% vol pu-1

reader should go to the books (see Table 2.2). Suitable software

packages are SPSS (originally ‘Statistical Package for the Social
4 Sciences’ but now more general), SAS and STATA. These have
comprehensive manuals associated with them, and the software
is often available through academic sources. With such pack-
2
ages, ‘clickity-click’ analysis can be a hazard; a few mouse clicks
can unleash a sophisticated analysis of a large dataset that the
user does not understand and may misinterpret.
0 10 20 30 40 50
MTR (pu) 2.2.2.1 Group comparisons – t-test
The simplest test of a MR parameter is to compare measure-
FIGURE 2.9 Matching MTR group histograms from two centres with ments in two groups (typically diseased and ‘normal’). Many MR
1.5T scanners from different manufacturers. By using body coil excita- parameters may have been measured. To test their usefulness, it is
tion and standardised histogram generation, inter-centre differences tempting to carry out multiple t-tests, for example to see if any of
were eliminated. (From Tofts, P.S., et al., Magma, 19(4), 209–222, 2006.) the parameters differ between clinical groups. If 20 tests are car-
ried out at the p = 0.05 significance level, on average one test will
come out positive by chance (this is known as a type I error). Thus
9
the results of multiple comparisons must be treated with caution.
8 × The Bonferroni correction for multiple comparisons (Bland and
Altman 1995) allow for this by suggesting that the appropriate
7 p-value is the value that would have been used for a single test (e.g.
× ×
p = 0.05), divided by the number of comparisons (e.g. 20). A much-
Estimated EDSS

6 × ×
×
× + BE
× reduced p-value (e.g. 0.002) is then used, and the chances of a type
5 × PP
RR I error are reduced (in this example to 0.04). The Bonferroni cor-
4 × SP rection can be unnecessarily cautious, missing an effect that is
present (i.e. a type II error). If the outcome variables being tested
3 + are correlated, then the reduction of the p-value by the number of
+ +
2
+ + tests is too extreme. Conversely, if several of the tests show signifi-
cance (which does imply correlation between the tested variables),
1 then the probability of this occurring by chance are much lower
1 2 3 4 5 6 7 8 9
Actual EDSS than the probability of just one occurring by chance.
A useful distinction can be made between two kinds of study.
FIGURE 2.10 Correlation of Expanded Disability Status Scale (EDSS,
A fishing expedition looks at many parameters, tests them at
a clinical score used in multiple sclerosis, MS) with principle compo- p = 0.05, accepting that some type I errors will occur, and uses this
nents enables an estimated EDSS to be calculated for each subgroup of to gain insight or guide further studies. A strict hypothesis-driven
people with MS, solely from the MTR histogram and knowledge of the study sets up the hypothesis before-analysis, makes only one test
subgroup. These are usually within one point of the actual clinically and is thus able to control type I and type II errors better. Thus
measured EDSS value. Compared with conventional features, PCA a fishing expedition might be used to set up a hypothesis-driven
gives better correlation coefficients. (From Dehmeshki, J., et al., Magn. study, which must be carried out on separate data. Alternatively,
Reson. Med., 46(3), 600–609, 2001.) the data could be divided into two parts, the first used for fishing
and the second for a strict test (although the power of the study
would be reduced by the smaller sample sizes).
2.2.2 Types of statistical analysis
Negative results in a group comparison: A negative result has two
Statistical analysis of the measurements is an increasingly com- possible explanations: firstly, the genuine biological spread in each
plex procedure. Involvement with statisticians at an early stage group may be too large to pick out a significant different between
in study design and analysis is advisable. Failure to be aware of them; secondly, the effect of measurement error may have broad-
statistical pitfalls can lead to embarrassing rejections by journal ened the spread in the groups, beyond its genuine biological value,
referees; sometimes re-analysis can address the problems, some- enough to obscure genuine biological difference (see Chapter 3,
times data collection is fatally flawed. Clinical datasets can be Figure 3.5), a kind of ‘false-negative’. Thus a negative result can be
Measurement Process: MR Data Collection and Image Analysis 25

TABLE 2.2 Statistics books

Title Authors Date Published Number of Pages Description
Statistical Methods in Medical Peter Armitage, Geoffrey 2001 832 Classic from statisticians, 4th edition,
Research Berry, JNS Matthews, hardback.
Practical Statistics for Medical Research Douglas Altman 2000 254 2nd edition, paperback.
An Introduction to Medical Statistics Martin Bland 2015 448 4th edition, paperback.
Essential Medical Statistics B Kirkwood, J Sterne 2003 512
Medical Statistics: A Textbook for the MJ Campbell, 2007 344 4th edition, paperback.
Health Sciences DJ Machin, SJ Walters
Health Measurement Scales: A Practical DL Streiner, 2014 416 5th edition, good on clinical scales, paperback.
Guide to Their Development and Use GR Norman, J Cairney
Analyzing Multivariate Data J Lattin, JD Carroll, 2006 556 High level hardback, covers principal components
PE Green analysis, analysis of variance, clustering,
discriminant analysis and much more.
An Introduction to Error Analysis: The JR Taylor 1997 488 A clear description of eternal truths, from a
Study of Uncertainties in Physical physics point of view.
Measurements

the consequence of poor measurement technique (high ISD – see 2.2.2.2 Correlation with clinical score
Chapter 3, Section 3.3.2.1). Another centre (with better technique) In many studies, MR parameters are tested for correlation with a
might succeed in showing a group difference in a similar group of clinical measure (in MS, often the Expanded Disability Status Scale,
subjects. The original centre might succeed with a larger sample EDSS), in an attempt to investigate or demonstrate their clinical
(i.e. the study was underpowered). Failure to observe a difference utility (or lack of it). A parameter with a high correlation coeffi-
does not mean that none exists. cient is thought to be a good candidate for a surrogate MR marker
In the case of a negative result, insight can be obtained in two of the disease in a clinical trial. In MS, low correlation coefficients
ways: first, estimates of measurement error and within-group r are reported (typically 0.3–0.6, sometimes 0.8; Dehmeshki et al.,
variance should be made (see Chapter 3, Section 3). This enables 2001). Significance values p are also given. Correlation values
the intraclass correlation coefficient (ICC) to be calculated; a are attenuated by the imperfect reliability (i.e. scatter) in the MR
good ICC means that the groups are genuinely indistinguish- and clinical scores (thus even if the two measures were intrinsi-
able; a poor ICC means that the failure to distinguish may be cally perfectly correlated, the correlation plot would show scat-
caused by poor instrumentation. Second, the confidence limits ter about the line describing this relationship). Correlation does
on the group means, and the minimum detectable group differ- not imply causality, only association, and the association may be
ence, should be reported. Other workers can then judge whether weakened by the introduction of another causative factor (such
improved technique (i.e. reduced measurement error) might as treatment). Thus a good correlation between an MR parameter
enable them to obtain a positive result. and a clinical score does not necessarily imply the parameter is a
good MR surrogate in a treatment trial; more evidence is needed
The excessive use of hypothesis testing at the of a direct causal relationship between the biological changes that
expense of more informative approaches ... is an happen in the disease and the MR parameter.
unsatisfactory way of assessing ... findings from An alternative way to think of the linear regression implied
medical studies. We prefer the use of confidence in correlation is: how well can the MR parameter predict the
intervals .... (Altman et al., 2008). current value of the clinical parameter? Thus a high correlation
implies that the MR parameter provides a good estimate of the
Positive results: If a positive result is obtained, the confidence current clinical status (Dehmeshki et al., 2001)– see Figure 2.10.
limits on group means and group difference should be given; then This establishes the relevance of the MR measure, although the
other groups can estimate whether their measurement errors are ultimate goal is to predict future clinical status. The fraction of
low enough to repeat the positive observation. False positive results the variance in the clinical score explained by the MR parameter
can be obtained if there is another (confounding) factor that dif- is r2, and this is a useful interpretation of r.
fers between the groups. This could be the time of scanning (if Age and gender correlations with the principle variables can
one group is scanned before a change in measurement procedure, cause problems; they should be included in the correlation, as
and the other after the change) or other uncontrolled variables, for covariates (Chard et al., 2002).
example age, gender, lifestyle differences or even head size.37
2.2.2.3 Clinical scores
37 A study comparing MS patients and controls found a difference that was
actually caused by head size (this gave a different spatial normalisation Clinical status can be quantified using scores. For example in
and hence intensity for the two groups). MS the EDSS is used to measure disability. In tumours a grading
26 Quantitative MRI of the Brain

system is used, based on histology of biopsy samples. In psychi- only attempts to choose between two classes (binary classi-
atric illness, a battery of psychometric tests, including cognitive fication). A linear discriminant is often constructed, and a
and emotional, are in use. Newly introduced MR parameters threshold used to assign the class (see e.g. Dehmeshki et al.,
have often been judged by how well they correlate with exist- 2002a). Such binary classification techniques have been used
ing clinical scores; intensive effort has gone into characterising in spectroscopy to classify tumours into several types (Tate
and improving the performance of the MR parameter. In turn it et al., 1998). The choice of threshold in a classifier is crucial in
has been accepted that the same intensive study should go into balancing false positive and false negative errors, and receiver
the clinical scores, and in fact they do have some serious short- operating characteristic (ROC) formalism is an ideal way to
comings. EDSS is non-linear; mixes impairments of ambulation, view and optimise this balance. ROC analysis is also used to
fine motor skill and cognition; and has limited reproducibility characterise how well a radiologist can identify a lesion on a
(Hobart et al., 2000). difficult background.
Scores that are more appropriate are being designed. In MS Receiver operating characteristic (ROC) curves (Zweig and
the functional composite score (MSFC score) is increasingly Campbell 1993; Altman and Bland 1994b; Armitage et al., 2001;
popular (Fischer et al., 1999; Cohen et al., 2000). It consists of Dendy and Heaton 2002; Huo et al., 2002) originate from stud-
three components, which measure different aspects of impair- ies to characterise screens used by radar operators, and spe-
ment. Leg function and ambulation are measured by the timed cifically recognised that the number of objects reported would
25-foot walk,38 arm function by the nine-hole peg test and cogni- depend on how a particular operator was making their deci-
tion by the Paced Auditory Serial Addition Test. Correlations of sions (see Table 2.3). A low threshold of abnormality, that is
MR parameters with these individual components may be supe- reporting all objects that could possibly be real, would result in
rior, since they reflect different impairments, which may occur a large number of positive decisions. The proportion of actual
at different times in the evolution of the disease and originate objects detected (i.e. true positives) would be large, but at the
from different locations in the CNS. There is still controversy expense of many false positives (which are in fact noise on
over how reliable these measurements are, since there may be the screen). A higher threshold (only reporting objects judged
learning effects in the subjects. A correlation plot may show an to be certainly real) would result in fewer positive decisions,
approximately linear dependence of a MR parameter on EDSS, more missed objects (false negatives) and fewer false positives.
but a line will often not go through the normal point (normal MR Thus the choice of decision threshold allows true positives to
value, EDSS = 0), possibly because subclinical changes happen to be traded off against false positives, and a particular thresh-
the MR parameter, before any clinical disability is apparent. old, corresponding to a particular point on the curve, can be
Tumour grade scoring based on sampling tissue is vulnerable chosen according to the relative benefit of true decisions versus
to missing high-grade tissue in a heterogeneous tumour, and the cost of false decisions. For example in a screening program,
multicentre studies can show appreciable differences in grad- false negatives are expensive, since missing a tumour may
ing between pathologists, which can limit how well MRS clas- result in death, and false positives are also costly (although less
sifiers can work. The psychological tests are generally better, so), since they produce unnecessary worry in the subjects.
because there is a longer-standing tradition of test design in In the radar screen context, an ROC curve can be gener-
that field, although learning, floor and ceiling effects and tired- ated by asking the observer to give a score with each object
ness are still important limitations. reported – for example 0: object absent (corresponding to a
In view of these shortcomings in clinical scores, and the com- low threshold), 1: object possibly present, 2: may be present,
plex relationship between biological changes and the ensuing 3: probably present, 4: almost certainly present, 5: certainly
clinical changes, a failure of MR parameters to closely predict present (corresponding to a high threshold). The scores can
these scores is hardly surprising. A more realistic test may be also be defined by their anticipated probabilities (e.g. 0: <10%
to look at how well the MR correlates with the biology and with of being present, 1: 10%–30% of being present, etc.). The prob-
scores of simple human functions. abilities need not be correct; they just allow the observer to
behave consistently. Combinations of scores of True Positive
2.2.2.4 Classification of individual subjects and Rate (TPR) and False Positive Rate (FPR) are summed to give
receiver operating characteristic curves points in the ROC space corresponding to different thresholds
If a measurement is performing well in separating two groups
of subjects (Section 2.2.2.1), then its performance on individ- TABLE 2.3 Decision matrix, in ROC formalism
ual subjects is worth investigating. In computer science meth-
Object reported not to Object reported to be
odology, the MR measurements may be seen as an example of be present (N) present (P)
a classifier. A classifier is a software tool for deciding which
Object absent (N) True negative (TN) False positive (FP)
class a number of subjects belong to, based on measurements
Object present (P) False negative (FN) True positive (TP)
made on each subject. Classification in its simplest case
Note: The test is to report whether an object is present or not; it
could equally well be to determine whether a disease or a lesion is present
38 Often replaced by the 10 m walk in Europe. or not.
Measurement Process: MR Data Collection and Image Analysis 27

80 1.0

No. of cases per unit x-interval

70
60 Negatives
Positives
50

TPR
40 0.5
TPR (d' = 2)
30
thr = 0
20 thr = 10
TPR (d' = 1)
10 TPR (d' = 3)
0
–10 –5 0 5 10 15 20 25 0 0.5 1.0
x-Value FPR
(a) (b)

1.0
1.0

Positive predictive value

Sensitivity

0.5 0.5

0.0
0 0.5 1.0 0.8 0.9 1.0
Specificity Negative predictive value
(c) (d)

FIGURE 2.11 A receiver operating characteristic (ROC) computer simulation: (a) negative (N) and positive (P) normal distributions. There are
1000 cases, a prevalence of 1 in 10 (so 100 positive cases). Mean values are x = 0 and x = 10. Standard deviation is 5 for both groups (b) ROC curve
(labelled d' = 2, meaning that the distributions are separated by 2 SD). Points from thresholds at the centre of each distribution are shown in parts b,
c and d (thr = 0, centre of negatives; thr = 10, centre of positives). (Cases to the right-hand side of the threshold are reported positive. Thus the part of
the negative tail to the right of the threshold contributes false positives, which decrease with threshold value. Conversely, the part of the positive tail
to the left of the threshold gives false negatives; as the threshold is increased these increase, reducing the true positive rate TPR.) Curves for other
separations (d' = 1, i.e. 5 units, and d' = 3, i.e. 15 units) are shown. The decision criterion to generate the N and P results is that an x-value greater
than the threshold indicates that the sample comes from the P distribution. (c) Sensitivity versus specificity plot, which is the left–right reflection
of the ROC curve. (d) Positive and negative predictive values both depend on the threshold (and also on the prevalence).

(see Figure 2.11).39 More generally, an ROC curve can be gener- The false positive rate, equal to the fraction of the negative dis-
ated from any model that predicts binary status with a certain tribution that is to the right of the threshold (see Figure 2.11) is
probability, depending on one or more predictor variables, known in statistics as the significance, the α-value, p-value or the
such as MR parameters. ROC curves can be fitted to analytic chance of a type I error. The false negative rate, equal to the frac-
functions, to obtain a measure that enables ROC curves to be tion of the positive distribution that is to the left of the threshold,
compared and to provide an element of smoothing (Constable is the β-value, q-value, or the chance of a type II error (Haacke
et al., 1995; Sorenson and Wang 1996). et al., 1999; Armitage et al., 2001). 1-β is the sensitivity or true pos-
The terms sensitivity, specificity, positive predictive value, itive rate (TPR) and is known as the power of the test (the prob-
negative predictive value, accuracy and prevalence are used in ability of detecting a true association where one exists). Note that
the context of how well a test performs (Altman and Bland parameters such as the sensitivity are all estimated from samples
1994a, 1994c; Dalton et al., 2002). These can be defined in terms and therefore should be reported with confidence limits.
of the number of true and false positive and negative reports
(see Table 2.4). 2.2.2.5 Multiparametric analysis
39 Thus Threshold 1 corresponds to objects detected with scores 1–5 (i.e. all
Multiparametric studies can be powerful. Several MR parame-
objects), Threshold 2 is object scores 2–5 (i.e. certainty of ‘probably’ or ters are measured and combined in such a way to produce a bet-
more), etc., and each threshold gives a point on the ROC curve. ter predictor of biological change or clinical outcome. If several
28 Quantitative MRI of the Brain

TABLE 2.4 Radiological terms summarising the performance of a test

Dependent on
Term Formal Definition Formula Comment Prevalence?
Sensitivity The probability of the test finding the disease TP/(TP + FN) Sensitivity characterises the false No
amongst those who have the disease negatives.
Specificity The probability of the test finding no disease TN/(TN + FP) Specificity characterises the false No
amongst those who do not have the disease positives.
Positive predictive The fraction of people with a positive test TP/(TP + FP) Fraction of positive reports that are Yes
value result who do have the disease correct.
Negative predictive The fraction of people with a negative test TN/(TN + FN) Fraction of negative reports that are Yes
value result who do not have the disease correct.
Accuracy The fraction of test results that are correct (TP + TN)/(TP + TN + Fraction of all reports that are Yes
FP + FN) correct.
Prevalence The fraction of the sample that has the disease (TP + FN)/(TP + TN + Distinguish from incidence.a –
FP + FN)
True positive rate Fraction of positive cases that are detected TP/(TP + FN) Equals sensitivity. The fractional No
(TPR) area under the positive curve that
is to the right of the threshold.
False positive rate Fraction of negative cases that are reported FP/(TN + FP) Equals 1-specificity. The fractional No
(FPR) positive area under the negative curve that
is to the right of the threshold.
Note: In this case the test is to find out whether subjects have a disease or not. Negative indicates that they do not, positive that they do. Sensitivity and speci-
ficity define the performance of the test (regardless of prevalence), whilst predictive value and accuracy depend on prevalence (i.e. what fraction of the sample
has the disease). Thus the latter quantities are very different for an asymptomatic screened population and a symptomatic hospital population.
a Thus the prevalence is the fraction of people that have a particular disease at any one time. The incidence is the number of new cases of disease that arise in

a given period (usually 1 year) for a given size of population.

MR parameters all correlate with clinical scores and do not cor- extracted from histograms of each MR parameter and from
relate strongly with each other, then there is a case for measur- lesion values. Statistical techniques such as multiple linear
ing them all. A plea in the journal Neurology, on the subject of discriminant analysis and cluster analysis are appropriate
MRI techniques to monitor MS evolution (Filippi and Grossman (Tintore et al., 2001). Multiparametric analysis is discussed
2002), asked that further in Chapter 18.

1. Metrics from magnetisation transfer MRI, diffu- References

sion-weighted MRI and proton MRS should be
implemented to obtain reliable in vivo quantifi- Alecci M, Collins CM, Smith MB, Jezzard P. Radio frequency
cation of MS pathology. magnetic field mapping of a 3 Tesla birdcage coil: experi-
2. Multiparametric MRI should be used in all pos- mental and theoretical dependence on sample properties.
sible clinical circumstances and trials. Magn Reson Med 2001; 46(2): 379–85.
3. Reproducible quantitative MR measures should Altman DG, Bland JM. Diagnostic tests 2: predictive values.
ideally be used for the assessment of patients BMJ 1994a; 309(6947): 102.
and are essential for trials. Altman DG, Bland JM. Diagnostic tests 3: receiver operating
characteristic plots. BMJ 1994b; 309(6948): 188.
Diagnosis at an early stage of a disease with a long and vari- Altman DG, Bland JM. Diagnostic tests. 1: sensitivity and speci-
able course (e.g. multiple sclerosis or Alzheimer’s disease) and ficity. BMJ 1994c; 308(6943): 1552.
prediction of disease progression can also be viewed as a multi- Altman DG, Machin D, Bryant TN, Gardner MJ. Statistics with
parametric problem. Early symptoms, MRI and other biochemi- confidence: confidence intervals and statistical guidelines
cal data are available; their combination can be optimised to (2nd ed). BMJ books, London, pp. 24; 2000.
provide the best prediction of future clinical status, measured Andersen AH. On the Rician distribution of noisy MRI data.
according to various clinical scores. The optimisation can be Magn Reson Med 1996; 36(2): 331–3.
driven by criteria such as sensitivity, specificity and predictive Armitage P, Matthews JNS, Berry G. Statistical Methods in
value (see Table 2.4). Medical Research. Blackwell; 2001.
The effects of disease are similarly multidimensional and Barker GJ, Simmons A, Arridge SR, Tofts PS. A simple method
need several scores to properly characterise them, covering for investigating the effects of non-uniformity of radio-
both clinical symptoms and measurable biological changes. frequency transmission and radiofrequency reception in
To assemble multiparametric MR data, features can be MRI. Br J Radiol 1998; 71(841): 59–67.
Measurement Process: MR Data Collection and Image Analysis 29

Bland JM, Altman DG. Multiple significance tests: the Bonfer­ Dehmeshki J, Ruto AC, Arridge S, Silver NC, Miller DH,
roni method. BMJ 1995; 310(6973): 170. Tofts PS. Analysis of MTR histograms in multiple sclero-
Bottomley PA, Andrew ER. RF magnetic field penetration, phase sis using principal components and multiple discriminant
shift and power dissipation in biological tissue: implica- analysis. Magn Reson Med 2001; 46(3): 600–9.
tions for NMR imaging. Phys Med Biol 1978; 23(4): 630–43. Dehmeshki J, Van Buchem MA, Bosma GP, Huizinga TW,
Boudreau M, Tardif CL, Stikov N, Sled JG, Lee W, Pike GB. Tofts PS. Systemic lupus erythematosus: diagnostic appli-
B1 mapping for bias-correction in quantitative T1 imaging cation of magnetization transfer ratio histograms in
of the brain at 3T using standard pulse sequences. J Magn patients with neuropsychiatric symptoms--initial results.
Reson Imaging. 2017; 46:1673–1682. Radiology 2002b; 222(3): 722–8.
Brown RW, Cheng N, Haacke EM, Thompson MR, Dendy PP, Heaton B. Physics for Diagnostic Radiology. Institute
Venkatesan R. Magnetic Resonance Imaging: Physical of Physics, London; 2002.
Principles and Sequence Design (2nd edition): Wiley- Dowell NG, Tofts PS. Fast, accurate, and precise mapping of the
Blackwell, New Jersey; 2014. RF field in vivo using the 180 degrees signal null. Magn
Cercignani M, Inglese M, Siger-Zajdel M, Filippi M. Segmenting Reson Med 2007; 58(3): 622–30.
brain white matter, gray matter and cerebro-spinal fluid Duan Q, van Gelderen P, Duyn J. Improved Bloch-Siegert based
using diffusion tensor-MRI derived indices. Magn Reson B1 mapping by reducing off-resonance shift. NMR Biomed
Imaging 2001; 19(9): 1167–72. 2013; 26(9): 1070–8.
Chard DT, Griffin CM, Parker GJ, Kapoor R, Thompson AJ, Edelstein WA, Bottomley PA, Pfeifer LM. A signal-to-noise cali-
Miller DH. Brain atrophy in clinically early relapsing- bration procedure for NMR imaging systems. Med Phys
remitting multiple sclerosis. Brain 2002; 125(Pt 2): 327–37. 1984; 11(2): 180–5.
Cohen JA, Fischer JS, Bolibrush DM, Jak AJ, Kniker JE, Fernandez-Seara MA, Song HK, Wehrli FW. Trabecular bone
Mertz LA, et al. Intrarater and interrater reliability of the volume fraction mapping by low-resolution MRI. Magn
MS functional composite outcome measure. Neurology Reson Med 2001; 46(1): 103–13.
2000; 54(4): 802–6. Filippi M, Grossman RI. MRI techniques to monitor MS evo-
Collins CM, Liu W, Swift BJ, Smith MB. Combination of opti- lution: the present and the future. Neurology 2002; 58(8):
mized transmit arrays and some receive array reconstruc- 1147–53.
tion methods can yield homogeneous images at very high Fischer JS, Rudick RA, Cutter GR, Reingold SC. The Multiple
frequencies. Magn Reson Med 2005; 54(6): 1327–32. Sclerosis Functional Composite Measure (MSFC): an
Collins CM, Smith MB. Signal-to-noise ratio and absorbed integrated approach to MS clinical outcome assessment.
power as functions of main magnetic field strength, and National MS Society Clinical Outcomes Assessment Task
definition of “90 degrees” RF pulse for the head in the bird- Force. Mult Scler 1999; 5(4): 244–50.
cage coil. Magn Reson Med 2001; 45(4): 684–91. Goerner FL, Clarke GD. Measuring signal-to-noise ratio in par-
Collins CM, Wang Z. Calculation of radiofrequency electro- tially parallel imaging MRI. Med Phys 2011; 38(9): 5049–57.
magnetic fields and their effects in MRI of human subjects. Gudbjartsson H, Patz S. The Rician distribution of noisy MRI
Magn Reson Med 2011; 65(5): 1470–82. data. Magn Reson Med 1995; 34(6): 910–4.
Collins CM, Yang QX, Wang JH, Zhang X, Liu H, Michaeli S, Gutteridge S, Ramanathan C, Bowtell R. Mapping the absolute
et al. Different excitation and reception distributions with value of M0 using dipolar field effects. Magn Reson Med
a single-loop transmit-receive surface coil near a head- 2002; 47(5): 871–9.
sized spherical phantom at 300 MHz. Magn Reson Med Haacke EM, Brown RW, Thompson MR, Venkatesan R. Magnetic
2002; 47(5): 1026–8. Resonance Imaging; 1999. Physical principles and sequence
Constable RT, Skudlarski P, Gore JC. An ROC approach for eval- design.
uating functional brain MR imaging and postprocessing Haynes BI, Dowell NG, Tofts PS. Measuring scan-rescan reli-
protocols. Magn Reson Med 1995; 34(1): 57–64. ability in quantitative brain imaging reveals instability
Cunningham CH, Pauly JM, Nayak KS. Saturated double-angle in an apparently healthy imager and improves statistical
method for rapid B1+ mapping. Magn Reson Med 2006; power in a clinical study. ISMRM annual scientific meet-
55(6): 1326–33. ing; 2010; Stockholm; 2010. p. 2999.
Dalton CM, Brex PA, Miszkiel KA, Hickman SJ, MacManus DG, Henkelman RM. Measurement of signal intensities in the pres-
Plant GT, et al. Application of the new McDonald criteria ence of noise in MR images. Med Phys 1985; 12(2): 232–3.
to patients with clinically isolated syndromes suggestive of Henkelman RM. Erratum: measurement of signal intensities in
multiple sclerosis. AnnNeurol 2002; 52(1): 47–53. the presence of noise [Med. Phys. 12, 232 (1985)]. Med Phys
Dehmeshki J, Barker GJ, Tofts PS. Classification of disease sub- 1986; 13(4): 544.
group and correlation with disease severity using magnetic Hetherington HP, Chu WJ, Gonen O, Pan JW. Robust fully
resonance imaging whole-brain histograms: application to automated shimming of the human brain for high-field
magnetization transfer ratios and multiple sclerosis. IEEE 1 H spectroscopic imaging. Magn Reson Med 2006; 56(1):

Trans Med Imaging 2002a; 21(4): 320–31. 26–33.

30 Quantitative MRI of the Brain

Hobart J, Freeman J, Thompson A. Kurtzke scales revisited: the Miller AJ, Joseph PM. The use of power images to perform quan-
application of psychometric methods to clinical intuition. titative analysis on low SNR MR images. Magn Reson
Brain 2000; 123 (Pt 5): 1027–40. Imaging 1993; 11(7): 1051–6.
Hoult DI. The NMR receiver: a description and analysis of Moerland MA, Beersma R, Bhagwandien R, Wijrdeman HK,
design. Progress in NMR Spectroscopy 1978; 12: 41–77. Bakker CJ. Analysis and correction of geometric distor-
Hoult DI. The principle of reciprocity in signal strength tions in 1.5 T magnetic resonance images for use in radio-
­calculations – a mathematical guide. Concepts in Magnetic therapy treatment planning. Phys Med Biol 1995; 40(10):
Resonance 2000; 12: 173–87. 1651–4.
Hoult DI, Richards RE. The signal-to-noise ratio of the nuclear Morrell GR. A phase-sensitive method of flip angle mapping.
magnetic resonance experiment. J Magn Reson 1976; 24: Magn Reson Med 2008; 60(4): 889–94.
71–85. Morrell GR, Schabel MC. An analysis of the accuracy of mag-
Huo Z, Giger ML, Vyborny CJ, Metz CE. Breast cancer: effec- netic resonance flip angle measurement methods. Phys
tiveness of computer-aided diagnosis observer study with Med Biol 2010; 55(20): 6157–74.
independent database of mammograms. Radiology 2002; Murphy BW, Carson PL, Ellis JH, Zhang YT, Hyde RJ,
224(2): 560–8. Chenevert TL. Signal-to-noise measures for magnetic res-
Hurley SA, Yarnykh VL, Johnson KM, Field AS, Alexander AL, onance imagers. Magn Reson Imaging 1993; 11(3): 425–8.
Samsonov AA. Simultaneous variable flip angle-actual flip Nehrke K, Börnert P. DREAM – a novel approach for robust,
angle imaging method for improved accuracy and preci- ultrafast, multislice B₁ mapping. Magn Reson Med 2012;
sion of three-dimensional T1 and B1 measurements. Magn 68(5): 1517–26.
Reson Med 2012; 68(1): 54–64. Park DJ, Bangerter NK, Javed A, Kaggie J, Khalighi MM,
Hutton C, Bork A, Josephs O, Deichmann R, Ashburner J, Morrell GR. A statistical analysis of the Bloch-Siegert B1
Turner R. Image distortion correction in fMRI: a quanti- mapping technique. Phys Med Biol 2013; 58(16): 5673–91.
tative evaluation. NeuroImage 2002; 16(1): 217–40. Parker GJ, Barker GJ, Tofts PS. Accurate multislice gradient echo
Ibrahim TS. Analytical approach to the MR signal. Magn Reson T(1) measurement in the presence of non-ideal RF pulse
Med 2005; 54(3): 677–82. shape and RF field nonuniformity. Magn Reson Med 2001;
Ibrahim TS, Lee R, Baertlein BA, Abduljalil AM, Zhu H, 45(5): 838–45.
Robitaille PM. Effect of RF coil excitation on field inhomo- Parkes LM, Tofts PS. Improved accuracy of human cerebral
geneity at ultra high fields: a field optimized TEM resona- blood perfusion measurements using arterial spin label-
tor. Magn Reson Imaging 2001; 19(10): 1339–47. ing: accounting for capillary water permeability. Magn
Insko EK, Bolinger L. Mapping the radiofrequency field. J Magn Reson Med 2002; 48(1): 27–41.
Reson series A 1993; 103: 82–5. Pohmann R, Scheffler K. A theoretical and experimental com-
Jezzard P. Physical basis of spatial distortions in Magnetic parison of different techniques for B₁ mapping at very high
Resonance Images. In: Isaac B, editor. Handbook of fields. NMR Biomed 2013; 26(3): 265–75.
Medical Imaging: Academic Press, Cambridge, MA; 2002. Provencher SW. Estimation of metabolite concentrations from
Jin J, Liu F, Zuo Z, Xue R, Li M, Li Y, et al. Inverse field-based localized in vivo proton NMR spectra. Magn Reson Med
approach for simultaneous B₁ mapping at high fields - a 1993; 30(6): 672–9.
phantom based study. J Magn Reson 2012; 217: 27–35. Provencher SW. Automatic quantitation of localized in vivo 1H
Karlsen OT, Verhagen R, Bovee WM. Parameter estimation spectra with LCModel. NMR Biomed 2001; 14(4): 260–4.
from Rician-distributed data sets using a maximum likeli- Sabati M, Maudsley AA. Fast and high-resolution quantitative
hood estimator: application to T1 and perfusion measure- mapping of tissue water content with full brain coverage
ments. Magn Reson Med 1999; 41(3): 614–23. for clinically-driven studies. Magn Reson Imaging 2013;
Lemieux L, Barker GJ. Measurement of small inter-scan fluctua- 31(10): 1752–9.
tions in voxel dimensions in magnetic resonance images Sacolick LI, Sun L, Vogel MW, Dixon WT, Hancu I. Fast radio-
using registration. Med Phys 1998; 25(6): 1049–54. frequency flip angle calibration by Bloch-Siegert shift.
Lutti A, Hutton C, Finsterbusch J, Helms G, Weiskopf N. Magn Reson Med 2011; 66(5): 1333–8.
Optimization and validation of methods for mapping of Sacolick LI, Wiesinger F, Hancu I, Vogel MW. B1 mapping by
the radiofrequency transmit field at 3T. Magn Reson Med Bloch-Siegert shift. Magn Reson Med 2010; 63(5): 1315–22.
2010; 64(1): 229–38. Sbrizzi A, Raaijmakers AJ, Hoogduin H, Lagendijk JJ,
McRobbie D, Semple S. Quality Control and Artefacts in Luijten PR, van den Berg CA. Transmit and receive
Magnetic Resonance Imaging (IPEM report 112). York: RF fields determination from a single low-tip-angle
Institute of Physics and Engineering in Medicine; 2017. gradient-­echo scan by scaling of SVD data. Magn Reson
Michaelis T, Merboldt KD, Bruhn H, Hanicke W, Frahm J. Med 2014; 72(1): 248–59.
Absolute concentrations of metabolites in the adult human Sijbers J, den Dekker AJ, Van Audekerke J, Verhoye M,
brain in vivo: quantification of localized proton MR spec- Van Dyck D. Estimation of the noise in magnitude MR
tra. Radiology 1993; 187(1): 219–27. images. Magn Reson Imaging 1998; 16(1): 87–90.
Measurement Process: MR Data Collection and Image Analysis 31

Sled JG, Pike GB. Standing-wave and RF penetration artifacts Tubridy N, McKinstry CS. Neuroradiological history: Sir Joseph
caused by elliptic geometry: an electrodynamic analysis of Larmor and the basis of MRI physics. Neuroradiology
MRI. IEEE Trans Med Imaging 1998; 17(4): 653–62. 2000; 42(11): 852–5.
Sled JG, Pike GB. Correction for B(1) and B(0) variations in Venkatesan R, Lin W, Haacke EM. Accurate determination of
quantitative T(2) measurements using MRI. Magn Reson spin-density and T1 in the presence of RF-field inhomo-
Med 2000; 43(4): 589–93. geneities and flip-angle miscalibration. Magn Reson Med
Sorenson JA, Wang X. ROC methods for evaluation of fMRI 1998; 40(4): 592–602.
techniques. Magn Reson Med 1996; 36(5): 737–44. Volz S, Nöth U, Deichmann R. Correction of systematic errors
Stollberger R, Wach P. Imaging of the active B1 field in vivo. in quantitative proton density mapping. Magn Reson Med
Magn Reson Med 1996; 35(2): 246–51. 2012; 68(1): 74–85.
Tate AR, Griffiths JR, Martinez-Perez I, Moreno A, Barba I, Volz S, Nöth U, Rotarska-Jagiela A, Deichmann R. A fast
Cabanas ME, et al. Towards a method for automated clas- B1-mapping method for the correction and normalization
sification of 1H MRS spectra from brain tumours. NMR of magnetization transfer ratio maps at 3 T. NeuroImage
Biomed 1998; 11(4–5): 177–91. 2010; 49(4): 3015–26.
Tintore M, Rovira A, Brieva L, Grive E, Jardi R, Borras C, Watanabe H, Takaya N, Mitsumori F. Non-uniformity correc-
et al. Isolated demyelinating syndromes: comparison of tion of human brain imaging at high field by RF field map-
CSF oligoclonal bands and different MR imaging crite- ping of B1+ and B1−. J Magn Reson 2011; 212(2): 426–30.
ria to predict conversion to CDMS. Mult Scler 2001; 7(6): Weisskoff RM. Simple measurement of scanner stability for
359–63. functional NMR imaging of activation in the brain. Magn
Tofts PS. Optimal detection of blood-brain barrier defects with Reson Med 1996; 36(4): 643–5.
Gd-DTPA MRI-the influences of delayed imaging and Wheeler-Kingshott CA, Parker GJ, Symms MR, Hickman SJ,
optimised repetition time. Magn Reson Imaging 1996; Tofts PS, Miller DH, et al. ADC mapping of the human
14(4): 373–80. optic nerve: increased resolution, coverage, and reliability
Tofts PS, Kermode AG, MacManus DG, Robinson WH. Nasal with CSF-suppressed ZOOM-EPI. Magn Reson Med 2002;
orientation device to control head movement during CT 47(1): 24–31.
and MR studies. J Comput Assist Tomogr 1990; 14(1): 163–4. Whisenant JG, Dortch RD, Grissom W, Kang H,
Tofts PS, Steens SC, Cercignani M, Admiraal-Behloul F, Arlinghaus LR, Yankeelov TE. Bloch-Siegert
Hofman PA, van Osch MJ, et al. Sources of variation in B1-Mapping improves accuracy and precision of lon-
multi-centre brain MTR histogram studies: body-coil gitudinal relaxation measurements in the breast at 3 T.
transmission eliminates inter-centre differences. Magma Tomography 2016; 2(4): 250–9.
2006; 19(4): 209–22. Yarnykh VL. Actual flip-angle imaging in the pulsed steady
Tofts PS, Wray S. Noninvasive measurement of molar concentra- state: a method for rapid three-dimensional mapping of
tions of 31P metabolites in vivo, using surface coil NMR the transmitted radiofrequency field. Magn Reson Med
spectroscopy. Magn Reson Med 1988; 6(1): 84–6. 2007; 57(1): 192–200.
Tozer DJ, Tofts PS. Removing spikes caused by quantization Zweig MH, Campbell G. Receiver-operating characteristic
noise from high-resolution histograms. Magn Reson Med (ROC) plots: a fundamental evaluation tool in clinical
2003; 50(3): 649–53. medicine. Clin Chem 1993; 39(4): 561–77.
Other documents randomly have
different content
venger l'objet de ses dernières amours, elle prétendit exercer
féodalement le droit de haute justice dans sa maison; elle condamna
le coupable à mort et le fit décapiter sans désemparer, sous ses yeux
affamés de sang, devant le populaire assemblé dans le carrefour, sur
la porte même de l'hôtel de Sens.

La fraise collerette.
 VI

HENRI IV ET LOUIS XIII

Retour à une simplicité relative.—Les femmes-tours.—

Hautes coiffures.—Excommunication du décolletage.—Les
robes à grands ramages de fleurs.—Collets montés et
collets rabattus.—Tailles longues.—Les édits de Richelieu.
—La dame suivant l'édit.—Tailles courtes.

Il y a des siècles qui ont la vie dure, et d'autres qui meurent

avant l'âge, le XVIe siècle, de complexion sans doute particulièrement
robuste, se prolongea jusqu'à la fin du règne du Béarnais, avec ses
idées et ses mœurs, ses façons et ses modes. On verra plus tard le
XVIIe durer de même avec Louis XIV au détriment du XVIIIe, et ce
pauvre et charmant XVIIIe finir tristement avant l'âge, de mort subite
en l'année 89.

Ces années de grâce du XVIe siècle sous le sceptre du roi Henri,

sont une convalescence après les longues années de fièvre chaude;
la France, que la maladie a mise si bas, renaît, le poison qu'elle avait
dans les veines est expulsé, tout se répare, se nettoie et s'assainit.
Après les raffinements ridicules et maladifs du règne de Henri III,
le costume prend un caractère sans façon, un aspect de bonne et
simple franchise, s'il peut y avoir de la franchise dans le costume.
C'est cependant presque le même costume, mais simplifié dans les
lignes et débarrassé de ce qu'il avait de surabondant et de trop
cherché dans les détails.
 Les modes sont moins élégantes, certainement, celles des
femmes comme celles des hommes; elles ont bien des ridicules
aussi, mais ce sont des ridicules naïfs. On est sorti de la prétention
excessive, de la grâce raffinée et corrompue; en allant dans la
simplicité, on est tombé dans la lourdeur et la gaucherie, pourtant
de cette lourdeur inélégante mais saine, se dégagera bientôt la
grâce cavalière du costume Louis XIII. Il ne faut cependant pas
prendre ce mot simplicité au pied de la lettre: hâtons-nous de dire
que cette simplicité n'est que très relative.
Les jours d'apparat, les dames arboraient encore la même
quantité de joailleries et de pierreries que par le passé. La reine qui
a remplacé Marguerite de Valois après le divorce,—une deuxième
alliance Médicis qui ne paraît pas avoir trop réussi au Béarnais, bien
payé déjà pour se souvenir de Catherine—la reine de la main droite
Marie de Médicis et la reine du côté cœur Gabrielle d'Estrées,
duchesse de Verneuil, et les autres belles dames, se montraient
«aux fêtes, ballets, mascarades et collations, richement parées et
magnifiquement atournées et si fort chargées de pierres et pierreries
qu'elles ne pouvaient se remuer».
La reine montra lors d'une grande occasion, une robe, «étoffée
de trente-deux mille perles et trois mille diamants,» et à son
exemple les grandes dames et les dames de moyenne étoffe
dépensaient volontiers plus que leurs revenus, en somptuosités, en
habillements de brocart, satins, damas admirables, ramagés et
passementés d'or, chargés et surchargés de clinquant et de
joailleries diverses.
Voilà une bien étrange simplicité, et pourtant quand on examine
tableaux et estampes du temps, ces documents n'en montrent pas
moins une grande différence entre les suprêmes raffinements des
modes de Henri III et l'élégance un peu mastoque du temps de
Henri IV.
Les coiffures sont plus hautes, les têtes se surchargent de
cheveux achetés chez le coiffeur, à la couleur à la mode.
 Pour un temps les perruques des règnes de Louis XIV et Louis XV
apparaissent, mais sur la tête des dames: perruques brunes ou
blondes, perruques de simple filasse même, pour celles qui ne
pouvaient s'offrir mieux. Et avec les perruques la poudre aussi se
montre. C'est plutôt un empois mélangeant la pommade aux
poudres les plus diverses, depuis les fines poudres parfumées à la
violette et à l'iris, jusqu'à la poudre de chêne pourri, et à la simple
farine pour les naïves campagnardes.
Ce temps voit aussi éclore les mouches qui reparaîtront
également au XVIIIe siècle, mais ce sont d'abord des mouches larges
comme des emplâtres et d'un aspect moins séduisant que les
coquettes «assassines» de plus tard.
Les femmes du peuple et de la petite bourgeoisie ont gardé
l'ancien chaperon, coiffure modeste, pendant que les femmes de la
haute classe, coiffées en cheveux avec perles et bijoux, adoptent
pour sortir le chapeau ou la toque à petit bouquet de plumes.
Voici le portrait d'une dame à la mode:
En ces temps heureux de vivre et de respirer, après tant de
sombres années, une élégante est sanglée et comprimée dans un
corsage dur et rigide, fortement armé de baleines, une véritable
gaine descendant tout d'une pièce, sans indication de modelé, en
longue pointe sur la jupe.
 Toilette de Cour Henri IV.

Il faut dire qu'on se rattrape de cette mise à la gehenne par le

décolletage du corsage, très libéralement échancré en pointe aussi,
trop libéralement même, puisque Sa Sainteté le Pape se croit obligé
d'intervenir et menace d'excommunication les belles qui continueront
à se décolleter dans des proportions exagérées.
 DAME LOUIS XIII.

Cette menace d'excommunication—amende à payer seulement

là-haut—n'a pas beaucoup d'effet, et les grandes fraises, les collets
montés de magnifiques dentelles soutenues de fils d'archal,
continuent à encadrer les opulences du corsage. La fine dentelle va
si bien autour de la chair, elle fait si bien ressortir les épaules et les
épaules font si bien valoir les merveilles des points de Venise ou de
Flandre, cette délicate et si artistique orfèvrerie à l'aiguille!
 La belle Gabrielle.

D'énormes manches qui ne sont pas des manches tiennent au

corsage. Ce sont des ailes ouvertes fendues dès l'épaule, descendant
très bas, garnies de boutons serrés qui ne se boutonnent pas. La
vraie manche paraît en dessous, toujours rembourrée et remontante
aux épaules, terminée par des poignets en dentelles appelés rebras.
Les jupes sont moins ballonnées que jadis, le vertugadin est plus
modeste, c'est une simple cloche lourde et tombant droit, ou plutôt
cela ressemble à la grosse caisse bariolée d'un bataillon de Suisses,
mais les hanches sont renflées en coupole et accusées de façon
grotesque par un rang de tuyaux godronnés de la même étoffe que
la robe.
Il est assez difficile aux femmes d'avoir avec cela une démarche
élégante et légère; cependant les beautés de l'époque tiennent à ces
jupes et l'idéal de la grâce est d'affecter en marchant un
dandinement de canard pour leur donner un balancement
rythmique.
 Une dame élégante a sous la robe trois autres jupes qu'elle doit
montrer en se retroussant élégamment, trois autres jupes
d'ornementation et de couleurs différentes.
Dans la liste des étoffes et des couleurs à la mode, elle a de quoi
choisir, nous avons alors une série de noms aussi drolatiques que
ceux inventés plus tard par le capricieux XVIIIe siècle.
Couleur triste amie, ventre de biche, face grattée, couleur de rat,
fleur mourante, singe mourant, couleur de veuve réjouie, de temps
perdu, de trépassé revenu, Espagnol malade, péché mortel, jambon
commun, racleur de cheminée, etc.
Le temps de la régence de Marie de Médicis est une époque de
transition entre les modes du XVIe et celles du XVIIe siècles; le vrai
costume Louis XIII ne se dégagera complètement des derniers
vestiges des modes de la Renaissance que vers 1630, à l'époque des
édits réformateurs de Richelieu qui, prohibant draps et brocards d'or
et d'argent, broderies et passementeries de fils d'or, dentelles, points
coupés, forcèrent les élégants à se contenter d'étoffes et de lingeries
plus simples et induisirent les tailleurs de robes et d'habits à
chercher des formes nouvelles.
Pendant la première partie du règne, la mode se dégage
lentement de sa lourdeur, le vertugadin diminue peu à peu et le si
disgracieux renflement godronné au-dessus des hanches disparaît,
remplacé par un retroussis à grands plis de la jupe de dessus.
Le vertugadin humilié a passé la frontière, il règne en Espagne où
sous le nom de guarde infante, il prend de si colossales proportions
que l'autorité veut par des édits, comme en France, arrêter leur
développement. A l'amende s'ajoute la saisie et l'exposition publique
des objets prohibés. L'édit, sévèrement appliqué, suscita des
résistances violentes et des émeutes où le sang coula.
Le vertugadin eut la vie si longue de l'autre côté des Pyrénées
que les galants de la cour de Louis XIV le revirent avec surprise
porté par les dames de la cour espagnole lors de l'entrevue dans l'île
de la Conférence pour le mariage de Louis avec Marie-Thérèse.
En France, la recherche, la richesse et le faste, la multiplicité des
ornements, la surcharge de joaillerie se remettent à dominer dans la
mode et toutes les dames, même celles de la plus simple
bourgeoisie donnent dans l'abus des superfluités coûteuses et du
clinquant.

D'après Callot.

Comment «une galante femme en habits se comporte,» un poète

satirique va nous le dire:
 Il lui faut des carcans, chaînes et bracelets,
Diamants, affiquets et montants de collets,
Pour charger un mulet, et voire davantage...
Il lui faut des rabats de la sorte que celles
Qui sont de cinq ou six villages damoiselles;
Cinq collets de dentelle haute de demi-pié
L'un sur l'autre montés...
Si les vertugadins ont diminué, les fraises ont plutôt gagné en
hauteur et développement; les grands portraits de Rubens et ensuite
ceux de Van Dick nous montrent ces fraises de la dernière période,
en demi-circonférences s'évasant derrière la tête.
Mais les estampes de Callot et d'Abraham Bosse vont nous
renseigner sur les modes parisiennes d'avant et d'après les édits de
Richelieu.
Callot qui avant 1630 a dessiné de sa merveilleuse pointe tant
d'élégants et pittoresques cavaliers en pourpoint de soie ou de
buffle, tant d'officiers en hongreline, à petites bottes et grandes
flamberges, de seigneurs bien XVIIe siècle, dans ces costumes si
charmants et d'une si jolie crânerie, portés avec tant de prestance et
de laisser-aller, a gravé aussi quelques costumes de femmes, qui,
bien que de la même époque sont encore un peu dans le style des
modes du siècle précédent.
Ces dames portent encore les robes à taille longue serrée dans le
corps piqué rigide, les manches à bourrelets avec crevés tailladés en
grande ou petite déchiquetade, de couleurs vives, les jupes relevées
sur le vertugadin rétréci.
Elles sont chaussées de souliers à pont-levis, avec attaches sur le
coup de pied, une mode nouvelle.
Les bourgeoises non plus que les dames ne vont
Nulle part maintenant, qu'avec soulier à pont,
Qui aye aux deux côtés une large ouverture
 Pour faire voir leurs bas, et dessus pour parure
Un beau cordon de soie en nœud d'amour lié...
Ceci décrit suffisamment le soulier Louis XIII d'une si cavalière
élégance. Le Musée de Cluny dans sa riche collection de chaussures
en possède d'admirables, très découpés et décorés d'ornements
noirs sur le cuir fauve et d'autres plus simples avec le nœud de
rubans dit nœud d'amour.
Les découpures laissaient voir les bas de soie incarnat, couleur à
la mode; pour sortir on ajoutait à ces souliers des patins de velours
cramoisi à très hautes semelles.

Fraise Médicis.

Les gants des élégantes étaient non moins jolis, ornés de dessins
sur le dos et d'arabesques brodés sur le grand crispin emboîtant le
poignet.
 FIN DU RÈGNE DE LOUIS XIII.

De vives chamarrures, de grands ramages de fleurs courent sur

toutes les robes comme ils couvrent toutes les étoffes du temps. Le
jardin des plantes, autrefois jardin du Roi, doit sa création à cette
mode; le noyau primitif fut sous Henri IV le jardin d'un horticulteur
avisé où toutes les sortes de plantes françaises ou étrangères étaient
cultivées en vue de fournir des modèles aux dessinateurs d'étoffes
ou de broderies.
 Corsage Louis XIII.

Les coiffures varient. Longtemps à cause des grands collets des

fraises, elles sont restées très hautes, ondées ou frisées en bonnet
d'astrakan et ornées seulement de bijoux. Plus tard les fraises
s'abaissent tout à coup et se séparent en rabats de dentelle de point
coupé, rabattus sur l'échancrure carrée du corsage, et en collets
abaissés, sinon rabattus aussi.
La coiffure peut s'abaisser aussi avec ces fraises basses; on
forme un petit chignon dit culebutte derrière la tête et on encadre la
figure de jolies boucles tombantes ou frisées. Cette mode s'exagère
un peu, les femmes se font avec leur coiffure frisottée et les petites
mèches plaquées sur le front, une tête ronde comme une boule.
Viennent les édits de Richelieu qui veut empêcher l'or de France
de s'en aller, au détriment du commerce français, enrichir les
manufactures étrangères en achats de passementeries de soie de
Milan et de dentelles ou broderies, les édits qui prohibent ensuite les
galons et franges, parfilures et canetilles enrichies d'or et d'argent,
en ne permettant que les galons étroits de simple étoffe; le costume
va changer tout à coup,

Il faut serrer ces belles jupes

Qui brillent de clinquants divers.
On a pris les dames pour dupes,
Leurs habits n'en seront point couverts,

dit une dame dessinée par Abraham Bosse en 1634 après les édits et
la réformation du costume.
 Bourgeoise Louis XIII.

Changement radical, plus de surcharge d'ornements, plus

d'étoffes à ramages, plus de fines dentelles de Venise ou de
Bruxelles. La dame suivant l'édit d'Abraham Bosse porte sur une jupe
plate, à plis tombant droit, sans le moindre soupçon de vertugade,
un corsage à basques, à taille très haute serrée par un simple ruban,
des manches larges, ouvertes sur une manche de dessous très
simple sans la moindre broderie ni garniture.
La grande fraise, le grand collet monté ou rabattu est remplacé
par un grand rabat de lingerie qui monte jusqu'au menton. Il n'y a
plus dans ce costume aucun reste des modes du XVIe siècle
définitivement trépassées.
Mais ce costume extrêmement simple, d'une sobriété qui touche
à l'austérité, restera celui des toutes petites bourgeoises, des bonnes
ménagères à qui les édits somptuaires ne causent pas grand souci ni
douleur; c'est en somme dans les grandes lignes, le costume actuel
des sœurs de Saint-Vincent de Paul, aux couleurs près.
 Les belles dames vont prendre ce modeste costume d'après les
édits et le transformer bien vite et en faire un des ensembles les plus
élégants et les plus charmants que la mode ait inventés, un type
vraiment remarquable de haute distinction, juste au moment où le
costume masculin si dégagé, si cavalier des premiers temps de
Callot, va se modifier en mal, devenir lourd et guindé avec les
justaucorps à taille sous les bras et les hauts de chausses tombant
au mollet.

Fin du règne de Louis XIII.

La robe s'ouvre du haut en bas, laissant voir un devant de

corsage de satin clair orné d'aiguillettes et terminé en pointe
arrondie sur une jupe de dessous de soie ou satin mordoré. La robe
de dessus ainsi largement ouverte et assez longue, a tous ses plis
sur les côtés ou par derrière.
Les manches bouffantes sont coupées en minces bandes du haut
en bas, rattachées sur la saignée par un ruban ou simplement
ouvertes sur une riche manche de dessous et garnies sur l'ouverture
d'aiguillettes ou de nœuds de rubans.
Plus de collets montés, rien que des collets rabattus. Ces grands
collets et rabats de lingerie ont bien vite repris quelques riches
broderies, dont les pointes tombent maintenant très bas sur les
épaules et sur les bras, en même temps que de hautes manchettes
dentelées et découpées de la même broderie montent des poignets
jusqu'au coude.
Et touffes et bouffettes de rubans partout, rosettes au corsage;
guirlandes de rosettes à la ceinture, et colliers de perles tombant
dans le corsage, carcans de bijouterie serrés au cou, diamants et
pierres sur les aiguillettes et les ferrets. Voici la dame à la mode de
1635 qui s'en va promener ses riches atours à la Place Royale parmi
les galants à moustaches retroussées, qui papillonnent sous les
arcades.
Ce sera tout à l'heure le costume des héroïnes de la Fronde, des
duchesses liguées contre Mazarin, et cela deviendra en se modifiant
peu à peu le grand costume des fêtes éblouissantes de la cour de
Louis XIV.
Elégante Louis XIII.
Marion.
 VII

SOUS LE ROI-SOLEIL

Les héroïnes de la Fronde.—De la Vallière à la Maintenon.—

Les robes dites transparentes.—Triomphe de la dentelle.—
Le roman de la mode.—Les Steinquerques.—La coiffure à
la Fontanges.—Le règne de Mme de Maintenon ou trente-
cinq ans de morosité.

Le règne du grand roi. Le règne des architectures étalant une

somptuosité d'apparat, une solennité majestueuse et le règne des
perruques également solennelles et majestueuses, des modes d'un
luxe écrasant, où la superbe écrase un peu l'élégance!
 A LA COUR DU ROI-SOLEIL.

Le grand siècle! la grandeur poussée jusqu'au gonflement et la

splendeur jusqu'à la surcharge, la même lourde magnificence dans le
style des hôtels ou des palais, demeures des nobles seigneurs
emperruqués, dans le mobilier noble et pompeux que dans
l'habillement masculin et féminin et dans les fantaisies raffinées du
costume.
Le grand règne a un prologue légèrement agité, la Fronde, qui
donne occasion aux belles dames de faire un peu de galante
politique et de se donner une petite idée des émotions de leurs
grand'mères du temps de la Ligue. La mort a desserré la forte main
qui tenait les brides du royaume, Richelieu disparu, on peut
caracoler.
Et à l'exemple de messieurs les ducs, les héroïnes de la Fronde
ont caracolé! Ce commencement, quand le grand roi n'est encore
que le petit roi, a une jolie allure romanesque.

Mmes les Duchesses, Mme de Chevreuse, Mme de Montbazon,

Mme de Bouillon, Mme de Longueville et la duchesse de Montpensier,
Mademoiselle, la Grande Mademoiselle, petite-fille d'Henri IV, qui
aide à battre les soldats du roi à coups de canon, en attendant
qu'elle soit, à coups de canne, battue par son mari, le beau Lauzun
pris à défaut de Louis,—les belles et séduisantes rebelles aux libres
allures, aux beaux yeux et aux belles tailles sans aller jusqu'à la
casaque des gardes et la hongreline soldatesque, arborent avec
crânerie des costumes semi-militaires.
Pendant les années de troubles et d'émeutes, de guerre civile à
Paris et de cavalcades armées dans les provinces, n'assistent-elles
pas aux parades des troupes levées par les princes contre les
troupes du Roi, avec Condé ou contre Condé;—ces amazones, du
haut du perron de l'Hôtel de Ville, ne haranguent-elles pas les
Parisiens toujours en goût d'émeute, le populaire hérissé de vieilles
hallebardes et d'arquebuses ligueuses, ne passent-elles pas en revue
dans Paris un peu assiégé les forces de la Fronde, les milices
parisiennes qui traînent bruyamment ce qui reste du pittoresque
bric-à-brac guerrier du temps de M. de Guise, la Cavalerie des portes
cochères et le régiment de Corinthe de M. le Coadjuteur,—et ne
tirent-elles pas vaillamment, quand les affaires se gâtent, le canon
de la Bastille sur l'armée royale? Quel joli prétexte à modes
cavalières.
Tout est à la Fronde, les modes comme le reste. La mode pouvait
avoir quelque motif d'en vouloir au Mazarin qui renouvelait les édits
prohibitifs, ces éternels édits sans doute oubliés ou bravés aussitôt
que publiés et qu'il fallait renouveler toujours, frappant
alternativement les passementeries au profit des guipures, et les
guipures au bénéfice de passementeries.
Louis a grandi, il règne.
Mais le roi est jeune, le grand siècle songe à se divertir, il aime la
gloire, mais il aime aussi le plaisir. C'est sa première manière, plus
tard le siècle et le roi, vieillis tous deux, tout en gardant le culte de
la gloire, songeront à se repentir du plaisir.

Une Duchesse de la Fronde.

Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookball.com