Biomedical 3 100 PDF

Biomedical Engineering
Series
Edited by Michael R. Neuman
Published Titles
Electromagnetic Analysis and Design in Magnetic Resonance
Imaging, Jianming Jin
Endogenous and Exogenous Regulation and
Control of Physiological Systems, Robert B. Northrop
Artificial Neural Networks in Cancer Diagnosis, Prognosis,
and Treatment, Raouf N.G. Naguib and Gajanan V. Sherbet
Medical Image Registration, Joseph V. Hajnal, Derek Hill, and
David J. Hawkes
Introduction to Dynamic Modeling of Neuro-Sensory Systems,
Robert B. Northrop
Noninvasive Instrumentation and Measurement in Medical
Diagnosis, Robert B. Northrop
Handbook of Neuroprosthetic Methods, Warren E. Finn
and Peter G. LoPresti
Signals and Systems Analysis in Biomedical Engineering,
Robert B. Northrop
Angiography and Plaque Imaging: Advanced Segmentation
Techniques, Jasjit S. Suri and Swamy Laxminarayan
Analysis and Application of Analog Electronic Circuits to
Biomedical Instrumentation, Robert B. Northrop
Biomedical Image Analysis, Rangaraj M. Rangayyan
The BIOMEDICAL ENGINEERING Series
Series Editor Michael R. Neuman
Biomedical
Image Analysis
Rangaraj M. Rangayyan
University of Calgary
Calgary, Alberta, Canada
CRC PR E S S
Boca Raton London New York Washington, D.C.
9695_disclaimer Page 1 Wednesday, November 17, 2004 1:57 PM
Library of Congress Cataloging-in-Publication Data
Catalog record is available from the Library of Congress
This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microfilming, and recording, or by any information storage or
retrieval system, without prior permission in writing from the publisher.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation, without intent to infringe.
Visit the CRC Press Web site at www.crcpress.com
© 2005 by CRC Press LLC
No claim to original U.S. Government works

International Standard Book Number 0-8493-9695-6
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
To
my wife Mayura,
my daughter Vidya,
and my son Adarsh...
for etching in my mind
the most beautiful images
that I will treasure forever!
v
Preface
Background and Motivation

The science of medical imaging owes much of its existence to the discovery of
X rays by W.C. Roentgen over 100 years ago, in 1895. However, it was the
development of practical computed tomography scanners in the early 1970s by
G. Houns eld and others that brought computers into medical imaging and
clinical practice. Since then, computers have become integral components of
modern medical imaging systems and hospitals, performing a variety of tasks
from data acquisition and image generation to image display and analysis.
With the widespread acceptance of computed tomography came an implicit
invitation to apply computers and computing to a host of other medical imag-
ing situations. As new imaging modalities were developed, the need for com-
puting in image generation, manipulation, display, and analysis grew by many
folds. Computers are now found in virtually every medical imaging system,
including radiography, ultrasound, nuclear medicine, and magnetic resonance
imaging systems. The strengths of computer applications in medical imaging
have been recognized to such an extent that radiology departments in many
hospitals are changing over to \totally digital" departments, using comput-
ers for image archival and communication as well. The humble X-ray lm
that launched the eld of radiology may soon vanish, thereby contributing to
better management of the environment.
The increase in the number of modalities of medical imaging and in their
practical use has been accompanied by an almost natural increase in the
scope and complexity of the associated problems, requiring further advanced
techniques for their solution. For example, physiological imaging with radio-
isotopes in nuclear medicine imaging comes with a host of problems such as
noise due to scatter, e ects of attenuation along the path of propagation of
the gamma rays through the body, and severe blurring due to the collimators
used. Radiation dose concerns limit the strength and amount of the isotopes
that may be used, contributing to further reduction in image quality. Along
with the increase in the acceptance of mammography as a screening tool has
come the need to eciently process such images using computer vision tech-
niques. The use of high-resolution imaging devices for digital mammography
and digital radiography, and the widespread adoption of picture archival and
vii
viii Biomedical Image Analysis
communication systems, have created the need for higher levels of lossless
data compression. The use of multiple modalities of medical imaging for im-
proved diagnosis of a particular type of disease or disorder has raised the need
to combine diverse images of the same organ, or the results thereof, into a
readily comprehensible visual display.
The major strength in the application of computers to medical imaging lies
in the potential use of image processing and computer vision techniques for
quantitative or objective analysis. (See the July 1972 and May 1979 issues
of the Proceedings of the IEEE for historical reviews and articles on digital
image processing.) Medical images are primarily visual in nature however,
visual analysis of images by human observers is usually accompanied by lim-
itations associated with interpersonal variations, errors due to fatigue, errors
due to the low rate of incidence of a certain sign of abnormality in a screening
application, environmental distractions, etc. The interpretation of an image
by an expert bears the weight of the experience and expertise of the ana-
lyst however, such analysis is almost always subjective. Computer analysis
of image features, if performed with the appropriate logic, has the potential
to add objective strength to the interpretation of the expert. It thus becomes
possible to improve the diagnostic con dence and accuracy of even an expert
with many years of experience.
Developing an algorithm for medical image analysis, however, is not an easy
task quite often, it might not even be a straightforward process. The engi-
neer or computer analyst is often bewildered by the variability of features in
biomedical signals, images, and systems that is far higher than that encoun-
tered in physical systems or observations. Benign diseases often mimic the
features of malignant diseases malignancies may exhibit characteristic pat-
terns, which, however, are not always guaranteed to appear. Handling all of
the possibilities and the degrees of freedom in a biomedical system is a major
challenge in most applications. Techniques proven to work well with a certain
system or set of images may not work in another seemingly similar situation.
The Problem-solving Approach

The approach I have taken in presenting the material in this book is primarily
that of problem solving. Engineers are often said to be (with admiration, I
believe) problem solvers. However, the development of a problem statement
and gaining of a good understanding of the problem could require a signi cant
amount of preparatory work. I have selected a logical series of problems, from
the many I have encountered in my research work, for presentation in this
book. Each chapter deals with a certain type of problem with biomedical
images. Each chapter begins with a statement of the problem, and includes
Preface ix
illustrations of the problem with real-life images. Image processing or analysis
techniques are presented, starting with relatively simple \textbook methods",
followed by more sophisticated methods directed at speci c problems. Each
chapter concludes with applications to signi cant and practical problems. The
book is illustrated copiously, in due consideration of the visual nature of the
subject matter.
The methods presented in the book are at a fairly high level of technical
and mathematical sophistication. A good background in one-dimensional sig-
nal and system analysis 1, 2, 3] is very much required in order to follow the
procedures and analyses. Familiarity with the theory of linear systems, sig-
nals, and transforms such as the Laplace and Fourier, in both continuous and
discrete versions, will be assumed. We shall only briey study a few represen-
tative medical imaging techniques. We will study in more detail the problems
present with medical images after they have been acquired, and concentrate
on how to solve the problems. Some preparatory reading on medical imaging
equipment and techniques 3, 4, 5, 6] may be useful, but not always essential.
The Intended Audience

The book is primarily directed at engineering students in their nal year
of undergraduate studies or in their (post-)graduate studies. Electrical and
Computer Engineering students with a rich background in signals and systems
1, 2, 3] will be well prepared for the material in the book. Students in
other engineering disciplines or in computer science, physics, mathematics,
or geophysics should also be able to appreciate the material in this book. A
course on digital signal processing or digital lters 7] would form a useful
link, but a capable student without this topic may not face much diculty.
Additional study of a book on digital image processing 8, 9, 10, 11, 12, 13]
could assist in developing a good understanding of general image processing
methods, but is not required.
Practicing engineers, computer scientists, information technologists, medi-
cal physicists, and data-processing specialists working in diverse areas such as
telecommunications, seismic and geophysical applications, biomedical applica-
tions, hospital information systems, remote sensing, mapping, and geomatics
may nd this book useful in their quest to learn advanced techniques for image
analysis. They could draw inspiration from other applications of data process-
ing or analysis, and satisfy their curiosity regarding computer applications in
medicine and computer-aided medical diagnosis.
x Biomedical Image Analysis
Teaching and Learning Plans

An introduction to the nature of biomedical images is provided in Chap-
ter 1. The easy-to-read material in this chapter gives a general overview of
the imaging techniques that are commonly used to acquire biomedical images
for detailed treatment of medical imaging, refer to Macovski 5], Robb 14],
Barrett and Swindell 3], Huda and Slone 6], and Cho et al. 4]. A good
understanding of the basics of image data acquisition procedures is essential
in order to develop appropriate methods for further treatment of the images.
Several concepts related to image quality and information content are de-
scribed in Chapter 2, along with the related basics of image processing such
as the Fourier transform and the modulation transfer function. The notions,
techniques, and measures introduced in this chapter are extensively used in
the book and in the eld of biomedical image analysis a clear understanding
of this material is an important prerequisite to further study of the subject.
Most of the images acquired in practice su er loss of quality due to arti-
facts and practical limitations. Several methods for the characterization and
removal of artifacts and noise are presented in Chapter 3. Preprocessing of
images to remove artifacts without causing distortion or loss of the desired
information is an important step in the analysis of biomedical images.
Imaging and image processing techniques aimed toward the improvement of
the general quality or the desired features in images are described in Chapter 4.
Methods for contrast enhancement and improvement of the visibility of the
details of interest are presented with illustrative examples.
The important task of detecting regions of interest is the subject of Chap-
ter 5, the largest chapter in the book. Several approaches for the segmentation
and extraction of parts of images are described, along with methods to im-
prove initial approximations or results.
Objective analysis of biomedical images requires the extraction of numerical
features that characterize the most signi cant properties of the regions of
interest. Methods to characterize shape, texture, and oriented patterns are
described in Chapters 6, 7, and 8, respectively. Speci c features are required
for each application, and the features that have been found to be useful in one
application may not suit a new application under investigation. Regardless,
a broad understanding of this subject area is essential in order to possess the
arsenal of feature extraction techniques that is required when attacking a new
problem.
The material in the book through Chapter 8 provides resources that are
more than adequate for a one-semester course with 40 to 50 hours of lectures.
Some of the advanced and specialized topics in these chapters may be omitted,
depending upon the methods and pace of presentation, as well as the level of
comprehension of the students.
Preface xi
The specialized topic of image reconstruction from projections is dealt with
in Chapter 9. The mathematical details related to the derivation of tomo-
graphic images are presented, along with examples of application. This chap-
ter may be skipped in an introductory course, but included in an advanced
course.
Chapter 10 contains descriptions of methods for the restoration of images
with known models of image degradation. The advanced material in this
chapter may be omitted in an introductory course, but forms an important
subject area for those who wish to explore the subject to its full depth.
The subject of image data compression and coding is treated in detail in
Chapter 11. With due regard to the importance of quality and delity in the
treatment of health-related information, the focus of the chapter is on lossless
compression. This subject may also be considered to be an advanced topic of
specialized interest, and limited to an advanced course.
Finally, the most important and signi cant tasks in biomedical image anal-
ysis | pattern analysis, pattern classi cation, and diagnostic decision | are
described in Chapter 12. The mathematical details of pattern classi cation
techniques are presented, along with procedures for their incorporation in
medical diagnosis and clinical assessment. Since this subject forms the cul-
mination of biomedical image analysis, it is recommended that parts of this
chapter be included even in an introductory course.
The book includes adequate material for two one-semester courses or a
full-year course on biomedical image analysis. The subject area is still a
matter of research and development: instructors should endeavor to augment
their courses with material selected from the latest developments published in
advanced journals such as the IEEE Transactions on Medical Imaging as well
as the proceedings of the SPIE series of conferences on medical imaging. The
topics of biometrics, multimodal imaging, multisensor fusion, image-guided
therapy and surgery, and advanced visualization, which are not dealt with in
this book, may also be added if desired.
Each chapter includes a number of study questions and problems to facili-
tate preparation for tests and examinations. Several laboratory exercises are
also provided at the end of each chapter, which could be used to formulate
hands-on exercises with real-life and/or synthetic images. Selected data les
related to some of the problems and exercises at the end of each chapter are
available at the site
www.enel.ucalgary.ca/People/Ranga/enel697
It is strongly recommended that the rst one or two laboratory sessions
in the course be visits to a local hospital, health sciences center, or clinical
laboratory to view biomedical image acquisition and analysis in a practical
(clinical) setting. Images acquired from local sources (with the permissions
and approvals required) could form interesting and motivating material for
laboratory exercises, and should be used to supplement the data les pro-
vided. A few invited lectures and workshops by physiologists, radiologists,
xii Biomedical Image Analysis
pathologists, and other medical professionals should also be included in the
course so as to provide the students with a nonengineering perspective on the
subject.
Practical experience with real-life images is a key element in understanding
and appreciating biomedical image analysis. This aspect could be dicult and
frustrating at times, but provides professional satisfaction and educational
fun!
It is my humble hope that this book will assist students and researchers
who seek to enrich their lives and those of others with the wonderful powers
of biomedical image analysis. Electrical and Computer Engineering is indeed
a great eld in the service of humanity.
Rangaraj Mandayam Rangayyan
November, 2004
About the Author
Rangaraj (Raj) Mandayam Rangayyan was born in Mysore, Karnataka, In-

dia, on July 21, 1955. He received the Bachelor of Engineering degree in
Electronics and Communication in 1976 from the University of Mysore at the
People's Education Society College of Engineering, Mandya, Karnataka, In-
dia, and the Ph.D. degree in Electrical Engineering from the Indian Institute
of Science, Bangalore, Karnataka, India, in 1980. He was with the University
of Manitoba, Winnipeg, Manitoba, Canada, from 1981 to 1984. He joined the
University of Calgary, Calgary, Alberta, Canada, in 1984.
He is, at present, a Professor with the Department of Electrical and Com-
puter Engineering (and an Adjunct Professor of Surgery and Radiology) at the
University of Calgary. His research interests are in the areas of digital signal
and image processing, biomedical signal analysis, medical imaging and image
analysis, and computer vision. His research projects have addressed topics
such as mammographic image enhancement and analysis for computer-aided
diagnosis of breast cancer region-based image processing knee-joint vibration
signal analysis for noninvasive diagnosis of articular cartilage pathology di-
rectional analysis of collagen bers and blood vessels in ligaments restoration
of nuclear medicine images analysis of textured images by cepstral ltering
and soni cation and several other applications of biomedical signal and image
analysis.
He has lectured extensively in many countries, including India, Canada,
United States, Brazil, Argentina, Uruguay, Chile, United Kingdom, The Ne-
therlands, France, Spain, Italy, Finland, Russia, Romania, Egypt, Malaysia,
Singapore, Thailand, Hong Kong, China, and Japan. He has collaborated
with many research groups in Brazil, Spain, France, and Romania.
He was an Associate Editor of the IEEE Transactions on Biomedical Engi-
neering from 1989 to 1996 the Program Chair and Editor of the Proceedings
of the IEEE Western Canada Exhibition and Conference on \Telecommuni-
cation for Health Care: Telemetry, Teleradiology, and Telemedicine", July
1990, Calgary, Alberta, Canada the Canadian Regional Representative to
the Administrative Committee of the IEEE Engineering in Medicine and Bi-
ology Society (EMBS), 1990 to 1993 a Member of the Scienti c Program
Committee and Editorial Board, International Symposium on Computerized
Tomography, Novosibirsk, Siberia, Russia, August 1993 the Program Chair
and Co-editor of the Proceedings of the 15th Annual International Conference
of the IEEE EMBS, October 1993, San Diego, CA and Program Co-chair,
xiii
xiv Biomedical Image Analysis
20th Annual International Conference of the IEEE EMBS, Hong Kong, Oc-
tober 1998.
His research work was recognized with the 1997 and 2001 Research Excel-
lence Awards of the Department of Electrical and Computer Engineering, the
1997 Research Award of the Faculty of Engineering, and by appointment as a
\University Professor" in 2003, at the University of Calgary. He was awarded
the Killam Resident Fellowship in 2002 by the University of Calgary in sup-
port of writing this book. He was recognized by the IEEE with the award
of the Third Millennium Medal in 2000, and was elected as a Fellow of the
IEEE in 2001, Fellow of the Engineering Institute of Canada in 2002, Fellow
of the American Institute for Medical and Biological Engineering in 2003, and
Fellow of SPIE: the International Society for Optical Engineering in 2003.
Photo by Trudie Lee.

Acknowledgments
Writing this book on the multifaceted subject of biomedical image analysis

has been challenging, yet yielding more knowledge tiring, yet stimulating the
thirst to understand and appreciate more of the subject matter and dicult,
yet satisfying when a part was brought to a certain stage of completion.
A number of very important people have shaped me and my educational
background. My mother, Srimati Padma Srinivasan Rangayyan, and my fa-
ther, Sri Srinivasan Mandayam Rangayyan, encouraged me to keep striving to
gain higher levels of education and to set and achieve higher goals all the time.
I have been very fortunate to have been taught and guided by a number of
dedicated teachers, the most important of them being Professor Ivaturi Surya
Narayana Murthy, my Ph.D. supervisor, who introduced me to the topic of
biomedical signal analysis at the Indian Institute of Science, Bangalore, Kar-
nataka, India. I o er my humble prayers, respect, and admiration to their
spirits.
My basic education was imparted by many inuential teachers at Saint
Joseph's Convent, Saint Joseph's Indian High School, and Saint Joseph's Col-
lege in Mandya and Bangalore, Karnataka, India. My engineering educa-
tion was provided by the People's Education Society College of Engineering,
Mandya, aliated with the University of Mysore. I was initiated into research
in biomedical engineering at the Indian Institute of Science | India's premier
research institute and one of the very highly acclaimed research institutions
in the world. I express my gratitude to all of my teachers.
My postdoctoral training with Richard Gordon at the University of Mani-
toba, Winnipeg, Manitoba, Canada, made a major contribution to my com-
prehension of the eld of biomedical imaging and image analysis I express
my sincere gratitude to him. My association with clinical researchers and
practitioners at the University of Calgary and the University of Manitoba has
been invaluable in furthering my understanding of the subject matter of this
book. I express my deep gratitude to Cyril Basil Frank, Gordon Douglas Bell,
Joseph Edward Leo Desautels, Leszek Hahn, and Reinhard Kloiber of the
University of Calgary.
My understanding and appreciation of the subject of biomedical signal and
image analysis has been boosted by the collaborative research and studies
performed with my many graduate students, postdoctoral fellows, research as-
sociates, and colleagues. I place on record my gratitude to Fabio Jose Ayres,
xv
xvi Biomedical Image Analysis
Sridhar Krishnan, Naga Ravindra Mudigonda, Margaret Hilary Alto, Han-
ford John Deglint, Thanh Minh Nguyen, Ricardo Jose Ferrari, Liang Shen,
Roseli de Deus Lopes, Antonio Cesar Germano Martins, Marcelo Knorich
Zu o, Bego~na Acha Pi~nero, Carmen Serrano Gotarredona, Laura Roa, Annie
France Frere, Graham Stewart Boag, Vicente Odone Filho, Marcelo Valente,
Silvia Delgado Olabarriaga, Christian Roux, Basel Solaiman, Olivier Menut,
Denise Guliato, Fabricio Adorno, Mario Ribeiro, Mihai Ciuc, Vasile Buzuloiu,
Titus Zaharia, Constantin Vertan, Margaret Sarah Rose, Salahuddin Elka-
diki, Kevin Eng, Nema Mohamed El-Faramawy, Arup Das, Farshad Faghih,
William Alexander Rolston, Yiping Shen, Zahra Marjan Kazem Moussavi,
Joseph Provine, Hieu Ngoc Nguyen, Djamel Boulfelfel, Tamer Farouk Rabie,
Katherine Olivia Ladly, Yuanting Zhang, Zhi-Qiang Liu, Raman Bhalachan-
dra Paranjape, Joseph Andre Rodrigue Blais, Robert Charles Bray, Gopinath
Ramaswamaiah Kuduvalli, Sanjeev Tavathia, William Mark Morrow, Tim-
othy Chi Hung Hon, Subhasis Chaudhuri, Paul Soble, Kirby Jaman, Atam
Prakash Dhawan, and Richard Joseph Lehner. In particular, I thank Liang,
Naga, Ricardo, Gopi, Djamel, Hilary, Tamer, Antonio, Bill Rolston, Bill Mor-
row, and Joseph for permitting me to use signi cant portions of their theses
Naga for producing the cover illustration and Fabio, Hilary, Liang, Mihai,
Gopi, Joseph, Ricardo, and Hanford for careful proofreading of drafts of the
book. Sections of the book were reviewed by Cyril Basil Frank, Joseph Edward
Leo Desautels, Leszek Hahn, Richard Frayne, Norm Bartley, Randy Hoang
Vu, Ilya Kamenetsky, Vijay Devabhaktuni, and Sanjay Srinivasan. I express
my gratitude to them for their comments and advice. I thank Leonard Bruton
and Abu Sesay for discussions on some of the topics described in the book.
I also thank the students of my course ENEL 697 Digital Image Processing
over the past several years for their comments and feedback.
The book has bene ted signi cantly from illustrations and text provided
by a number of researchers worldwide, as identi ed in the references and per-
missions cited. I thank them all for enriching the book with their gifts of
knowledge and kindness. Some of the test images used in the book were ob-
tained from the Center for Image Processing Research, Rensselaer Polytechnic
Institute, Troy, NY, www.ipl.rpi.edu I thank them for the resource.
The research projects that have provided me with the background and expe-
rience essential in order to write the material in this book have been supported
by many agencies. I thank the Natural Sciences and Engineering Research
Council of Canada, the Alberta Heritage Foundation for Medical Research,
the Alberta Breast Cancer Foundation, Control Data Corporation, Kids Can-
cer Care Foundation of Alberta, the University of Calgary, the University
of Manitoba, and the Indian Institute of Science for supporting my research
projects.
I thank the Killam Trusts and the University of Calgary for awarding me
a Killam Resident Fellowship to facilitate work on this book. I gratefully ac-
knowledge support from the Alberta Provincial Biomedical Engineering Grad-
uate Programme, funded by a grant from the Whitaker Foundation, toward
Acknowledgments xvii
student assistantship for preparation of some of the exercises and illustrations
for this book and the related courses ENEL 563 Biomedical Signal Analysis
and ENEL 697 Digital Image Processing at the University of Calgary. I am
pleased to place on record my gratitude for the generous support from the
Department of Electrical and Computer Engineering and the Faculty of En-
gineering at the University of Calgary in terms of supplies, services, and relief
from other duties.
I thank Steven Leikeim for help with computer-related issues and problems.
My association with the IEEE Engineering in Medicine and Biology Society
(EMBS) in many positions has bene ted me considerably in numerous ways.
In particular, the period as an Associate Editor of the IEEE Transactions on
Biomedical Engineering was rewarding, as it provided me with a wonderful
opportunity to work with many leading researchers and authors of scienti c
articles. I thank IEEE EMBS and SPIE for lending professional support to
my career on many fronts.
Writing this book has been a monumental task, often draining me of all
of my energy. The in nite source of inspiration and recharging of my energy
has been my family | my wife Mayura, my daughter Vidya, and my son
Adarsh. While supporting me with their love and a ection, they have had to
bear the loss of my time and e ort at home. I express my sincere gratitude to
my family for their love and support, and place on record their contribution
toward the preparation of this book.
I thank CRC Press and its associates for inviting me to write this book and
for completing the publication process in a friendly and ecient manner.
Rangaraj Mandayam Rangayyan
November, 2004
Contents
Preface vii
About the Author xiii
Acknowledgments xv
Symbols and Abbreviations xxix
1 The Nature of Biomedical Images 1
1.1 Body Temperature as an Image . . . . . . . . . . . . . . . . . 2
1.2 Transillumination . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Light Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Electron Microscopy . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 X-ray Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5.1 Breast cancer and mammography . . . . . . . . . . . 22
1.6 Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.7 Nuclear Medicine Imaging . . . . . . . . . . . . . . . . . . . . 36
1.8 Ultrasonography . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.9 Magnetic Resonance Imaging . . . . . . . . . . . . . . . . . . 47
1.10 Objectives of Biomedical Image Analysis . . . . . . . . . . . 53
1.11 Computer-aided Diagnosis . . . . . . . . . . . . . . . . . . . . 55
1.12 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.13 Study Questions and Problems . . . . . . . . . . . . . . . . . 57
1.14 Laboratory Exercises and Projects . . . . . . . . . . . . . . . 58
2 Image Quality and Information Content 61
2.1 Diculties in Image Acquisition and Analysis . . . . . . . . . 61
2.2 Characterization of Image Quality . . . . . . . . . . . . . . . 64
2.3 Digitization of Images . . . . . . . . . . . . . . . . . . . . . . 65
2.3.1 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . 66
2.3.3 Array and matrix representation of images . . . . . . 69
2.4 Optical Density . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.5 Dynamic Range . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.6 Contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.7 Histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.8 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.9 Blur and Spread Functions . . . . . . . . . . . . . . . . . . . 90
2.10 Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xix
xx Biomedical Image Analysis
2.11 The Fourier Transform and Spectral Content . . . . . . . . . 99
2.11.1 Important properties of the Fourier transform . . . . 110
2.12 Modulation Transfer Function . . . . . . . . . . . . . . . . . 122
2.13 Signal-to-Noise Ratio . . . . . . . . . . . . . . . . . . . . . . 131
2.14 Error-based Measures . . . . . . . . . . . . . . . . . . . . . . 138
2.15 Application: Image Sharpness and Acutance . . . . . . . . . 139
2.16 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
3 Removal of Artifacts 151
3.1 Characterization of Artifacts . . . . . . . . . . . . . . . . . . 151
3.1.1 Random noise . . . . . . . . . . . . . . . . . . . . . . 151
3.1.2 Examples of noise PDFs . . . . . . . . . . . . . . . . . 159
3.1.3 Structured noise . . . . . . . . . . . . . . . . . . . . . 164
3.1.4 Physiological interference . . . . . . . . . . . . . . . . 165
3.1.5 Other types of noise and artifact . . . . . . . . . . . . 166
3.1.6 Stationary versus nonstationary processes . . . . . . . 166
3.1.7 Covariance and cross-correlation . . . . . . . . . . . . 168
3.1.8 Signal-dependent noise . . . . . . . . . . . . . . . . . 169
3.2 Synchronized or Multiframe Averaging . . . . . . . . . . . . . 171
3.3 Space-domain Local-statistics-based Filters . . . . . . . . . . 174
3.3.1 The mean lter . . . . . . . . . . . . . . . . . . . . . . 176
3.3.2 The median lter . . . . . . . . . . . . . . . . . . . . . 177
3.3.3 Order-statistic lters . . . . . . . . . . . . . . . . . . . 181
3.4 Frequency-domain Filters . . . . . . . . . . . . . . . . . . . . 193
3.4.1 Removal of high-frequency noise . . . . . . . . . . . . 194
3.4.2 Removal of periodic artifacts . . . . . . . . . . . . . . 199
3.5 Matrix Representation of Image Processing . . . . . . . . . . 202
3.5.1 Matrix representation of images . . . . . . . . . . . . 203
3.5.2 Matrix representation of transforms . . . . . . . . . . 206
3.5.3 Matrix representation of convolution . . . . . . . . . . 212
3.5.4 Illustrations of convolution . . . . . . . . . . . . . . . 215
3.5.5 Diagonalization of a circulant matrix . . . . . . . . . . 218
3.5.6 Block-circulant matrix representation of a 2D lter . . 221
3.6 Optimal Filtering . . . . . . . . . . . . . . . . . . . . . . . . 224
3.6.1 The Wiener lter . . . . . . . . . . . . . . . . . . . . . 225
3.7 Adaptive Filters . . . . . . . . . . . . . . . . . . . . . . . . . 228
3.7.1 The local LMMSE lter . . . . . . . . . . . . . . . . . 228
3.7.2 The noise-updating repeated Wiener lter . . . . . . . 234
3.7.3 The adaptive 2D LMS lter . . . . . . . . . . . . . . . 235
3.7.4 The adaptive rectangular window LMS lter . . . . . 237
3.7.5 The adaptive-neighborhood lter . . . . . . . . . . . . 241
3.8 Comparative Analysis of Filters for Noise Removal . . . . . . 251
3.9 Application: Multiframe Averaging in Confocal Microscopy . 270
Table of Contents xxi
3.10 Application: Noise Reduction in Nuclear Medicine Imaging . 271
3.11 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
4 Image Enhancement 285
4.1 Digital Subtraction Angiography . . . . . . . . . . . . . . . . 286
4.2 Dual-energy and Energy-subtraction X-ray Imaging . . . . . 287
4.3 Temporal Subtraction . . . . . . . . . . . . . . . . . . . . . . 291
4.4 Gray-scale Transforms . . . . . . . . . . . . . . . . . . . . . . 291
4.4.1 Gray-scale thresholding . . . . . . . . . . . . . . . . . 291
4.4.2 Gray-scale windowing . . . . . . . . . . . . . . . . . . 292
4.4.3 Gamma correction . . . . . . . . . . . . . . . . . . . . 294
4.5 Histogram Transformation . . . . . . . . . . . . . . . . . . . 301
4.5.1 Histogram equalization . . . . . . . . . . . . . . . . . 301
4.5.2 Histogram speci cation . . . . . . . . . . . . . . . . . 305
4.5.3 Limitations of global operations . . . . . . . . . . . . 310
4.5.4 Local-area histogram equalization . . . . . . . . . . . 310
4.5.5 Adaptive-neighborhood histogram equalization . . . . 311
4.6 Convolution Mask Operators . . . . . . . . . . . . . . . . . . 314
4.6.1 Unsharp masking . . . . . . . . . . . . . . . . . . . . . 314
4.6.2 Subtracting Laplacian . . . . . . . . . . . . . . . . . . 316
4.6.3 Limitations of xed operators . . . . . . . . . . . . . . 323
4.7 High-frequency Emphasis . . . . . . . . . . . . . . . . . . . . 325
4.8 Homomorphic Filtering for Enhancement . . . . . . . . . . . 328
4.8.1 Generalized linear ltering . . . . . . . . . . . . . . . 328
4.9 Adaptive Contrast Enhancement . . . . . . . . . . . . . . . . 338
4.9.1 Adaptive-neighborhood contrast enhancement . . . . 338
4.10 Objective Assessment of Contrast Enhancement . . . . . . . 346
4.11 Application: Contrast Enhancement of Mammograms . . . . 350
4.11.1 Clinical evaluation of contrast enhancement . . . . . . 354
4.12 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
5 Detection of Regions of Interest 363
5.1 Thresholding and Binarization . . . . . . . . . . . . . . . . . 364
5.2 Detection of Isolated Points and Lines . . . . . . . . . . . . . 365
5.3 Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 367
5.3.1 Convolution mask operators for edge detection . . . . 367
5.3.2 The Laplacian of Gaussian . . . . . . . . . . . . . . . 370
5.3.3 Scale-space methods for multiscale edge detection . . 380
5.3.4 Canny's method for edge detection . . . . . . . . . . . 390
5.3.5 Fourier-domain methods for edge detection . . . . . . 390
5.3.6 Edge linking . . . . . . . . . . . . . . . . . . . . . . . 392
xxii Biomedical Image Analysis
5.4 Segmentation and Region Growing . . . . . . . . . . . . . . . 393
5.4.1 Optimal thresholding . . . . . . . . . . . . . . . . . . 395
5.4.2 Region-oriented segmentation of images . . . . . . . . 396
5.4.3 Splitting and merging of regions . . . . . . . . . . . . 397
5.4.4 Region growing using an additive tolerance . . . . . . 397
5.4.5 Region growing using a multiplicative tolerance . . . . 400
5.4.6 Analysis of region growing in the presence of noise . . 401
5.4.7 Iterative region growing with multiplicative tolerance 402
5.4.8 Region growing based upon the human visual system 405
5.4.9 Application: Detection of calci cations by multitoler-
ance region growing . . . . . . . . . . . . . . . . . . . 410
5.4.10 Application: Detection of calci cations by linear pre-
diction error . . . . . . . . . . . . . . . . . . . . . . . 414
5.5 Fuzzy-set-based Region Growing to Detect Breast Tumors . . 417
5.5.1 Preprocessing based upon fuzzy sets . . . . . . . . . . 419
5.5.2 Fuzzy segmentation based upon region growing . . . . 421
5.5.3 Fuzzy region growing . . . . . . . . . . . . . . . . . . 429
5.6 Detection of Objects of Known Geometry . . . . . . . . . . . 434
5.6.1 The Hough transform . . . . . . . . . . . . . . . . . . 435
5.6.2 Detection of straight lines . . . . . . . . . . . . . . . . 437
5.6.3 Detection of circles . . . . . . . . . . . . . . . . . . . . 440
5.7 Methods for the Improvement of Contour or Region Estimates 444
5.8 Application: Detection of the Spinal Canal . . . . . . . . . . 449
5.9 Application: Detection of the Breast Boundary in Mammo-
grams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
5.9.1 Detection using the traditional active deformable con-
tour model . . . . . . . . . . . . . . . . . . . . . . . . 456
5.9.2 Adaptive active deformable contour model . . . . . . 464
5.9.3 Results of application to mammograms . . . . . . . . 476
5.10 Application: Detection of the Pectoral Muscle in Mammo-
grams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
5.10.1 Detection using the Hough transform . . . . . . . . . 481
5.10.2 Detection using Gabor wavelets . . . . . . . . . . . . . 487
5.10.3 Results of application to mammograms . . . . . . . . 495
5.11 Application: Improved Segmentation of Breast Masses by
Fuzzy-set-based Fusion of Contours and Regions . . . . . . . 500
5.12 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
6 Analysis of Shape 529
6.1 Representation of Shapes and Contours . . . . . . . . . . . . 529
6.1.1 Signatures of contours . . . . . . . . . . . . . . . . . . 530
6.1.2 Chain coding . . . . . . . . . . . . . . . . . . . . . . . 530
6.1.3 Segmentation of contours . . . . . . . . . . . . . . . . 534
Table of Contents xxiii
6.1.4 Polygonal modeling of contours . . . . . . . . . . . .. 537
6.1.5 Parabolic modeling of contours . . . . . . . . . . . .. 543
6.1.6 Thinning and skeletonization . . . . . . . . . . . . .. 548
6.2 Shape Factors . . . . . . . . . . . . . . . . . . . . . . . . .. 549
6.2.1 Compactness . . . . . . . . . . . . . . . . . . . . . .. 551
6.2.2 Moments . . . . . . . . . . . . . . . . . . . . . . . .. 555
6.2.3 Chord-length statistics . . . . . . . . . . . . . . . . .. 560
6.3 Fourier Descriptors . . . . . . . . . . . . . . . . . . . . . . .. 562
6.4 Fractional Concavity . . . . . . . . . . . . . . . . . . . . . .. 569
6.5 Analysis of Spicularity . . . . . . . . . . . . . . . . . . . . .. 570
6.6 Application: Shape Analysis of Calci cations . . . . . . . .. 575
6.7 Application: Shape Analysis of Breast Masses and Tumors . 578
6.8 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 581
6.9 Study Questions and Problems . . . . . . . . . . . . . . . .. 581
6.10 Laboratory Exercises and Projects . . . . . . . . . . . . . .. 582
7 Analysis of Texture 583
7.1 Texture in Biomedical Images . . . . . . . . . . . . . . . . . . 584
7.2 Models for the Generation of Texture . . . . . . . . . . . . . 584
7.2.1 Random texture . . . . . . . . . . . . . . . . . . . . . 589
7.2.2 Ordered texture . . . . . . . . . . . . . . . . . . . . . 589
7.2.3 Oriented texture . . . . . . . . . . . . . . . . . . . . . 590
7.3 Statistical Analysis of Texture . . . . . . . . . . . . . . . . . 596
7.3.1 The gray-level co-occurrence matrix . . . . . . . . . . 597
7.3.2 Haralick's measures of texture . . . . . . . . . . . . . 600
7.4 Laws' Measures of Texture Energy . . . . . . . . . . . . . . . 603
7.5 Fractal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 605
7.5.1 Fractal dimension . . . . . . . . . . . . . . . . . . . . 608
7.5.2 Fractional Brownian motion model . . . . . . . . . . . 609
7.5.3 Fractal analysis of texture . . . . . . . . . . . . . . . . 609
7.5.4 Applications of fractal analysis . . . . . . . . . . . . . 611
7.6 Fourier-domain Analysis of Texture . . . . . . . . . . . . . . 612
7.7 Segmentation and Structural Analysis of Texture . . . . . . . 621
7.7.1 Homomorphic deconvolution of periodic patterns . . . 623
7.8 Audi cation and Soni cation of Texture in Images . . . . . . 625
7.9 Application: Analysis of Breast Masses Using Texture and
Gradient Measures . . . . . . . . . . . . . . . . . . . . . . . . 627
7.9.1 Adaptive normals and ribbons around mass margins . 629
7.9.2 Gradient and contrast measures . . . . . . . . . . . . 632
7.9.3 Results of pattern classi cation . . . . . . . . . . . . . 635
7.10 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637
xxiv Biomedical Image Analysis
8 Analysis of Oriented Patterns 639
8.1 Oriented Patterns in Images . . . . . . . . . . . . . . . . . . 639
8.2 Measures of Directional Distribution . . . . . . . . . . . . . 641
8.2.1 The rose diagram . . . . . . . . . . . . . . . . . . . . 641
8.2.2 The principal axis . . . . . . . . . . . . . . . . . . . . 641
8.2.3 Angular moments . . . . . . . . . . . . . . . . . . . . 642
8.2.4 Distance measures . . . . . . . . . . . . . . . . . . . . 643
8.2.5 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . 643
8.3 Directional Filtering . . . . . . . . . . . . . . . . . . . . . . . 644
8.3.1 Sector ltering in the Fourier domain . . . . . . . . . 646
8.3.2 Thresholding of the component images . . . . . . . . . 649
8.3.3 Design of fan lters . . . . . . . . . . . . . . . . . . . 651
8.4 Gabor Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 657
8.4.1 Multiresolution signal decomposition . . . . . . . . . . 660
8.4.2 Formation of the Gabor lter bank . . . . . . . . . . . 664
8.4.3 Reconstruction of the Gabor lter bank output . . . . 665
8.5 Directional Analysis via Multiscale Edge Detection . . . . . . 666
8.6 Hough-Radon Transform Analysis . . . . . . . . . . . . . . . 671
8.6.1 Limitations of the Hough transform . . . . . . . . . . 671
8.6.2 The Hough and Radon transforms combined . . . . . 673
8.6.3 Filtering and integrating the Hough-Radon space . . 676
8.7 Application: Analysis of Ligament Healing . . . . . . . . . . 679
8.7.1 Analysis of collagen remodeling . . . . . . . . . . . . . 680
8.7.2 Analysis of the microvascular structure . . . . . . . . 684
8.8 Application: Detection of Breast Tumors . . . . . . . . . . . 699
8.8.1 Framework for pyramidal decomposition . . . . . . . . 707
8.8.2 Segmentation based upon density slicing . . . . . . . . 710
8.8.3 Hierarchical grouping of isointensity contours . . . . . 712
8.8.4 Results of segmentation of masses . . . . . . . . . . . 712
8.8.5 Detection of masses in full mammograms . . . . . . . 719
8.8.6 Analysis of mammograms using texture ow- eld . . . 726
8.8.7 Adaptive computation of features in ribbons . . . . . 732
8.8.8 Results of mass detection in full mammograms . . . . 735
8.9 Application: Bilateral Asymmetry in Mammograms . . . . . 742
8.9.1 The broglandular disc . . . . . . . . . . . . . . . . . 743
8.9.2 Gaussian mixture model of breast density . . . . . . . 744
8.9.3 Delimitation of the broglandular disc . . . . . . . . . 747
8.9.4 Motivation for directional analysis of mammograms . 755
8.9.5 Directional analysis of broglandular tissue . . . . . . 757
8.9.6 Characterization of bilateral asymmetry . . . . . . . . 766
8.10 Application: Architectural Distortion in Mammograms . . . 775
8.10.1 Detection of spiculated lesions and distortion . . . . . 775
8.10.2 Phase portraits . . . . . . . . . . . . . . . . . . . . . 779
8.10.3 Estimating the orientation eld . . . . . . . . . . . . 780
8.10.4 Characterizing orientation elds with phase portraits 782
Table of Contents xxv
8.10.5 Feature extraction for pattern classi cation . . . . . . 785
8.10.6 Application to segments of mammograms . . . . . . . 785
8.10.7 Detection of sites of architectural distortion . . . . . . 786
8.11 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791
9 Image Reconstruction from Projections 797
9.1 Projection Geometry . . . . . . . . . . . . . . . . . . . . . . . 797
9.2 The Fourier Slice Theorem . . . . . . . . . . . . . . . . . . . 798
9.3 Backprojection . . . . . . . . . . . . . . . . . . . . . . . . . . 801
9.3.1 Filtered backprojection . . . . . . . . . . . . . . . . . 804
9.3.2 Discrete ltered backprojection . . . . . . . . . . . . . 806
9.4 Algebraic Reconstruction Techniques . . . . . . . . . . . . . . 813
9.4.1 Approximations to the Kaczmarz method . . . . . . . 820
9.5 Imaging with Di racting Sources . . . . . . . . . . . . . . . . 825
9.6 Display of CT Images . . . . . . . . . . . . . . . . . . . . . . 825
9.7 Agricultural and Forestry Applications . . . . . . . . . . . . . 829
9.8 Microtomography . . . . . . . . . . . . . . . . . . . . . . . . 831
9.9 Application: Analysis of the Tumor in Neuroblastoma . . . . 834
9.9.1 Neuroblastoma . . . . . . . . . . . . . . . . . . . . . . 834
9.9.2 Tissue characterization using CT . . . . . . . . . . . . 838
9.9.3 Estimation of tissue composition from CT images . . 839
9.9.4 Results of application to clinical cases . . . . . . . . . 844
9.9.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 845
9.10 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854
10 Deconvolution, Deblurring, and Restoration 857
10.1 Linear Space-invariant Restoration Filters . . . . . . . . . . . 857
10.1.1 Inverse ltering . . . . . . . . . . . . . . . . . . . . . . 858
10.1.2 Power spectrum equalization . . . . . . . . . . . . . . 860
10.1.3 The Wiener lter . . . . . . . . . . . . . . . . . . . . . 863
10.1.4 Constrained least-squares restoration . . . . . . . . . 872
10.1.5 The Metz lter . . . . . . . . . . . . . . . . . . . . . . 874
10.1.6 Information required for image restoration . . . . . . 875
10.1.7 Motion deblurring . . . . . . . . . . . . . . . . . . . . 875
10.2 Blind Deblurring . . . . . . . . . . . . . . . . . . . . . . . . . 877
10.2.1 Iterative blind deblurring . . . . . . . . . . . . . . . . 878
10.3 Homomorphic Deconvolution . . . . . . . . . . . . . . . . . . 885
10.3.1 The complex cepstrum . . . . . . . . . . . . . . . . . . 885
10.3.2 Echo removal by Radon-domain cepstral ltering . . . 886
10.4 Space-variant Restoration . . . . . . . . . . . . . . . . . . . . 891
10.4.1 Sectioned image restoration . . . . . . . . . . . . . . . 893
xxvi Biomedical Image Analysis
10.4.2 Adaptive-neighborhood deblurring . . . . . . ... . . 894
10.4.3 The Kalman lter . . . . . . . . . . . . . . . ... . . 898
10.5 Application: Restoration of Nuclear Medicine Images .. . . 919
10.5.1 Quality control . . . . . . . . . . . . . . . . . ... . . 922
10.5.2 Scatter compensation . . . . . . . . . . . . . ... . . 922
10.5.3 Attenuation correction . . . . . . . . . . . . . ... . . 923
10.5.4 Resolution recovery . . . . . . . . . . . . . . ... . . 924
10.5.5 Geometric averaging of conjugate projections ... . . 926
10.5.6 Examples of restoration of SPECT images . . ... . . 934
10.6 Remarks . . . . . . . . . . . . . . . . . . . . . . . . ... . . 949
10.7 Study Questions and Problems . . . . . . . . . . . . ... . . 953
10.8 Laboratory Exercises and Projects . . . . . . . . . . ... . . 954
11 Image Coding and Data Compression 955
11.1 Considerations Based on Information Theory . . . . . . . . . 956
11.1.1 Noiseless coding theorem for binary transmission . . . 957
11.1.2 Lossy versus lossless compression . . . . . . . . . . . . 957
11.1.3 Distortion measures and delity criteria . . . . . . . . 959
11.2 Fundamental Concepts of Coding . . . . . . . . . . . . . . . . 960
11.3 Direct Source Coding . . . . . . . . . . . . . . . . . . . . . . 961
11.3.1 Hu man coding . . . . . . . . . . . . . . . . . . . . . 961
11.3.2 Run-length coding . . . . . . . . . . . . . . . . . . . . 969
11.3.3 Arithmetic coding . . . . . . . . . . . . . . . . . . . . 969
11.3.4 Lempel{Ziv coding . . . . . . . . . . . . . . . . . . . . 974
11.3.5 Contour coding . . . . . . . . . . . . . . . . . . . . . . 977
11.4 Application: Source Coding of Digitized Mammograms . . . 978
11.5 The Need for Decorrelation . . . . . . . . . . . . . . . . . . . 980
11.6 Transform Coding . . . . . . . . . . . . . . . . . . . . . . . . 984
11.6.1 The discrete cosine transform . . . . . . . . . . . . . . 987
11.6.2 The Karhunen{Loeve transform . . . . . . . . . . . . 989
11.6.3 Encoding of transform coecients . . . . . . . . . . . 992
11.7 Interpolative Coding . . . . . . . . . . . . . . . . . . . . . . . 1001
11.8 Predictive Coding . . . . . . . . . . . . . . . . . . . . . . . . 1004
11.8.1 Two-dimensional linear prediction . . . . . . . . . . . 1005
11.8.2 Multichannel linear prediction . . . . . . . . . . . . . 1009
11.8.3 Adaptive 2D recursive least-squares prediction . . . . 1026
11.9 Image Scanning Using the Peano-Hilbert Curve . . . . . . . . 1033
11.9.1 De nition of the Peano-scan path . . . . . . . . . . . 1035
11.9.2 Properties of the Peano-Hilbert curve . . . . . . . . . 1040
11.9.3 Implementation of Peano scanning . . . . . . . . . . . 1040
11.9.4 Decorrelation of Peano-scanned data . . . . . . . . . . 1041
11.10 Image Coding and Compression Standards . . . . . . . . . . 1043
11.10.1 The JBIG Standard . . . . . . . . . . . . . . . . . . . 1046
11.10.2 The JPEG Standard . . . . . . . . . . . . . . . . . . . 1049
11.10.3 The MPEG Standard . . . . . . . . . . . . . . . . . . 1050
Table of Contents xxvii
11.10.4 The ACR/ NEMA and DICOM Standards . . . . . . 1050
11.11 Segmentation-based Adaptive Scanning . . . . . . . . . . . . 1051
11.11.1 Segmentation-based coding . . . . . . . . . . . . . . . 1051
11.11.2 Region-growing criteria . . . . . . . . . . . . . . . . . 1052
11.11.3 The SLIC procedure . . . . . . . . . . . . . . . . . . . 1055
11.11.4 Results of image data compression with SLIC . . . . . 1055
11.12 Enhanced JBIG Coding . . . . . . . . . . . . . . . . . . . . . 1062
11.13 Lower-limit Analysis of Lossless Data Compression . . . . . . 1066
11.13.1 Memoryless entropy . . . . . . . . . . . . . . . . . . . 1070
11.13.2 Markov entropy . . . . . . . . . . . . . . . . . . . . . 1071
11.13.3 Estimation of the true source entropy . . . . . . . . . 1071
11.14 Application: Teleradiology . . . . . . . . . . . . . . . . . . . 1079
11.14.1 Analog teleradiology . . . . . . . . . . . . . . . . . . . 1080
11.14.2 Digital teleradiology . . . . . . . . . . . . . . . . . . . 1082
11.14.3 High-resolution digital teleradiology . . . . . . . . . . 1084
11.15 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1086
12 Pattern Classication and Diagnostic Decision 1089
12.1 Pattern Classi cation . . . . . . . . . . . . . . . . . . . . . . 1091
12.2 Supervised Pattern Classi cation . . . . . . . . . . . . . . . . 1095
12.2.1 Discriminant and decision functions . . . . . . . . . . 1095
12.2.2 Distance functions . . . . . . . . . . . . . . . . . . . . 1097
12.2.3 The nearest-neighbor rule . . . . . . . . . . . . . . . . 1104
12.3 Unsupervised Pattern Classi cation . . . . . . . . . . . . . . 1104
12.3.1 Cluster-seeking methods . . . . . . . . . . . . . . . . . 1105
12.4 Probabilistic Models and Statistical Decision . . . . . . . . . 1110
12.4.1 Likelihood functions and statistical decision . . . . . . 1110
12.4.2 Bayes classi er for normal patterns . . . . . . . . . . . 1118
12.5 Logistic Regression . . . . . . . . . . . . . . . . . . . . . . . . 1120
12.6 The Training and Test Steps . . . . . . . . . . . . . . . . . . 1125
12.6.1 The leave-one-out method . . . . . . . . . . . . . . . . 1125
12.7 Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . 1126
12.8 Measures of Diagnostic Accuracy . . . . . . . . . . . . . . . . 1132
12.8.1 Receiver operating characteristics . . . . . . . . . . . 1135
12.8.2 McNemar's test of symmetry . . . . . . . . . . . . . . 1138
12.9 Reliability of Features, Classi ers, and Decisions . . . . . . . 1140
12.9.1 Statistical separability and feature selection . . . . . . 1141
12.10 Application: Image Enhancement for Breast Cancer Screen-
ing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1143
12.10.1 Case selection, digitization, and presentation . . . . . 1145
12.10.2 ROC and statistical analysis . . . . . . . . . . . . . . 1147
12.10.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . 1159
xxviii Biomedical Image Analysis
12.11 Application: Classi cation of Breast Masses and Tumors via
Shape Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 1160
12.12 Application: Content-based Retrieval and Analysis of Breast
Masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1166
12.12.1 Pattern classi cation of masses . . . . . . . . . . . . . 1167
12.12.2 Content-based retrieval . . . . . . . . . . . . . . . . . 1169
12.12.3 Extension to telemedicine . . . . . . . . . . . . . . . . 1177
12.13 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1182
References 1187
Index 1262
Symbols and Abbreviations
Note: Bold-faced letters represent the vector or matrix form of the variable
in the corresponding plain letters.
Variables or symbols used within limited contexts are not listed here they
are described within their context.
The mathematical symbols listed may stand for other entities or variables
in di erent applications only the common associations used in this book are
listed for ready reference.
a(p q) a autoregressive model or lter coecients
arctan inverse tangent, tan;1
arg argument of
atan inverse tangent, tan;1
au arbitrary units
AADCM adaptive active deformable contour model
ACF autocorrelation function
ACR American College of Radiology
ADC analog-to-digital converter
ALZ adaptive Lempel-Ziv coding
AMTA AMT acutance
ANCE adaptive-neighborhood contrast enhancement
AND adaptive-neighborhood deblurring
ANN arti cial neural network
ANNS adaptive-neighborhood noise subtraction
AR autoregressive model or lter
ARMA autoregressive, moving-average model or lter
ARW adaptive rectangular window
Az area under the ROC curve
b background intensity
b bit
b(m n) moving-average model or lter coecients
bps bits per second
B byte
BIBO bounded-input { bounded-output stability
BI-RADSTM Breast Imaging Reporting and Data System
BP backprojection
cd candela
xxix
xxx Biomedical Image Analysis
cm centimeter
C contrast
C covariance matrix
Ci Curie
cf Co compactness
Ci the ith class in a pattern classi cation problem
Cxy covariance between x and y
CAD computer-aided diagnosis
CBP convolution backprojection
CBIR content-based image retrieval
CC cranio-caudal
CCD charge-coupled device
CCF cross-correlation function
CCITT Comite Consultatif International Telephonique et Telegraphique
CD compact disk
CLS constrained least squares
CMTA cascaded modulation transfer acutance
CMYK cyan, magenta, yellow, black] representation of color
CNR contrast-to-noise ratio
CNS central nervous system
CR computed radiography
CREW compression with reversible embedded wavelets
CRT cathode ray tube
CSD cross-spectral density, cross-spectrum
CT computed tomography
CV coecient of variation
dB decibel
dE Euclidean dimension
df fractal dimension
dpi dots per inch
DAC digital-to-analog converter
DC direct current zero frequency
DCT discrete cosine transform
DFT discrete Fourier transform
DICOM Digital Imaging and Communications in Medicine
DoG di erence of Gaussians
DPCM di erential pulse code modulation
DR digital radiography
DSA digital subtraction angiography
DWT directional wavelet transform
e(n), E (!) model or estimation error
eV electron volt
exp (x) exponential function, ex
ECG electrocardiogram, electrocardiography
Symbols and Abbreviations xxxi
EEG electroencephalogram
EM electromagnetic
EM expectation-maximization
Ex total energy of the signal x
E ] statistical expectation operator
f foreground intensity
fc cuto frequency (usually at ;3 dB ) of a lter
fcc fractional concavity
ff shape factor obtained using Fourier descriptors
fps frames per second
fs sampling frequency
f (m n) a digital image, typically original or undistorted
f (x y) an image, typically original or undistorted
FBP ltered backprojection
FFT fast Fourier transform
FID free-induction decay
FIR nite impulse response ( lter)
FM frequency modulation
FN false negative
FNF false-negative fraction
FOM gure of merit
FP false positive
FPF false-positive fraction
FT Fourier transform
FWHM full width at half the maximum
g ( m n) a digital image, typically processed or distorted
g (x y ) an image, typically processed or distorted
h(m n) impulse response of a system
h(p) measure of information
h(x y) impulse response of a system
H hydrogen
H Hurst coecient
H magnetic eld strength
H entropy
Hf g joint entropy of f and g
Hf jg conditional entropy of f given g
H Hermitian (complex-conjugate) transposition of a matrix
H (k l) discrete Fourier transform of h(m n)
H (u v) frequency response of a lter, Fourier transform of h(x y)
HINT hierarchical interpolation
HU Houns eld unit
HVS human visual system
Hz Hertz
i index of a series
xxxii Biomedical Image Analysis
I the identity matrix
If jg mutual information
IEPA image edge-pro le acutance
IFT inverse Fourier transform
IIR in nite impulse response ( lter)
ISO International
p Organization for Standardization
j ;1
JBIG Joint Bi-level Image (experts) Group
JM Je ries-Matusita distance
JND just-noticeable di erence
JPEG Joint Photographic Experts Group
k kilo (1,000)
(k l) indices in the discrete Fourier (frequency) domain
kV p kilo-volt peak
K kilo (1,024)
KESF knife-edge spread function
KLT Karhunen-Loeve transform
ln natural logarithm (base e)
lp=mm line pairs per millimeter
L an image processing operator or transform in matrix form
Lij loss function in pattern classi cation
LEAP low-energy all-purpose collimator
LEGP low-energy general-purpose collimator
LLMMSE local linear minimum mean-squared error
LMMSE linear minimum mean-squared error
LMS least mean squares
LMSE Laplacian mean-squared error
LoG Laplacian of Gaussian
LP linear prediction (model)
LSF line spread function
LSI linear shift-invariant
LUT look-up table
LZW Lempel-Ziv-Welch code
m meter
m mean
m mean vector of a pattern class
max maximum
mA milliampere
mf shape factor using moments
min minimum
mm millimeter
(m n) indices in the discrete space (image) domain
mod modulus or modulo
modem modulator { demodulator
Symbols and Abbreviations xxxiii
M number of samples or pixels
MA moving average ( lter)
MAP maximum-a-posteriori probability
MCL medial collateral ligament
MIAS Mammographic Image Analysis Society, London, England
MDL minimum description length
ME maximum entropy
MLO medio-lateral oblique
MMSE minimum mean-squared error
MPEG Moving Picture Experts Group
MR magnetic resonance
MRI magnetic resonance imaging
MS mean-squared
MSE mean-squared error
MTF modulation (magnitude) transfer function
n an index
nm nanometer
N number of samples or pixels
NE normalized error
NEMA National Electrical Manufacturers Association
NMR nuclear magnetic resonance
NMSE normalized mean-squared error
NPV negative predictive value
NSHP nonsymmetric half plane
OD optical density
OTF optical transfer function
pf (l) normalized histogram or PDF of image f
pf g (l1 l2 ) joint PDF of images f and g
pf jg (l1 l2 ) conditional PDF of f given g
pixel picture cell or element
pm mth ray sum in ART
pps pulses per second
(p q) indices of a 2D array
p (t) projection (Radon transform) of an image at angle
p(x) probability density function of the random variable x
p(xjCi ) likelihood function of class Ci or state-conditional PDF of x
P model order
P (x) probability of the event x
Pf (l) histogram of image f
P (Ci jx) posterior probability that x belongs to the class Ci
P (w) Fourier transform of the projection p (t)
P a predicate
PA posterior{anterior
PACS picture archival and communication system
xxxiv Biomedical Image Analysis
PCA principal-component analysis
PCG phonocardiogram (heart sound signal)
PDF probability density function
PET positron emission tomography
PMSE perceptual mean-squared error
PMT photomultiplier tube
PPV positive predictive value
PSD power spectral density, power spectrum
PSE power spectrum equalization
PSF point spread function
PSV prediction selection values in JPEG
q (t) ltered projection of an image at angle
Q model order
QP quarter plane
rj (x) average risk or loss in pattern classi cation
(r s) temporary indices of a 2D array
R+ the set of nonnegative real numbers
RBST rubber-band straightening transform
RD relative dispersion
RDM radial distance measures
RF radio-frequency
RGB red, green, blue] color representation
RLS recursive least-squares
RMS root mean-squared
ROC receiver operating characteristics
ROI region of interest
ROS region of support
RUKF reduced-update Kalman lter
s second
s space variable in the projection (Radon) space
Sf (u v) power spectral density of the image f
SAR synthetic-aperture radar
SD standard deviation
SEM scanning electron microscope
SI spiculation index
SLIC segmentation-based lossless image coding
SMTA system modulation transfer acutance
SNR signal-to-noise ratio
SPECT single-photon emission computed tomography
SQF subjective quality factor
SQRI square-root integral
STFT short-time Fourier transform
SVD singular value decomposition
S+ sensitivity of a test
S; speci city of a test
Symbols and Abbreviations xxxv
t time variable
t space variable in the projection (Radon) space
T Tesla (strength of a magnetic eld)
T a threshold
T as a superscript, vector or matrix transposition
Tc technetium
Tl thallium
T1 longitudinal relaxation time constant in MRI
T2 transverse magnetization time constant in MRI
T+ positive test result
T; negative test result
TEM transmission electron microscope
Th threshold
TN true negative
TNF true-negative fraction
TP true positive
TPF true-positive fraction
Tr trace of a matrix
TSE total squared error
TV television
u(x y) unit step function
(u v) frequency coordinates in the continuous Fourier domain
UHF ultra high frequency
voxel volume cell or element
V volt
VLSI very-large-scale integrated circuit
w lter tap weight weighting function
w frequency variable related to projections
w lter or weight vector
WHT Walsh-Hadamard transform ;
WN Fourier transform kernel function WN = exp ;j 2N
W Fourier transform operator in matrix form
(x y ) image coordinates in the space domain
x a feature vector in pattern classi cation
YIQ luminance, in-phase, quadrature] color representation
z a prototype feature vector in pattern classi cation
Z the set of all integers
null set
1D one-dimensional
2D two-dimensional
3D three-dimensional
4D four-dimensional
xy correlation coecient between x and y
;xy coherence between x and y: Fourier transform of xy
xxxvi Biomedical Image Analysis
;(p) a fuzzy membership function
Dirac delta (impulse) function
$x $y sampling intervals along the x and y axes
" model error, total squared error
a random variable or noise process
scale factor in fractal analysis
an angle
a threshold
, % cross-correlation function
( t) the Radon (projection) space
forgetting factor in the RLS lter
the mean (average) of a random variable
X-ray attenuation coecient
m micrometer
CT micro-computed tomography
correlation coecient
the standard deviation of a random variable
2 the variance of a random variable
fg covariance between images f and g
fg covariance between images f and g in matrix form
' basis function of a transform
f autocorrelation of image f in array form
f autocorrelation of image f in matrix form
fg cross-correlation between images f and g in array form
fg cross-correlation between images f and g in matrix form
&f Fourier transform of f power spectral density of f
r gradient operator
hi dot product
! factorial
when in-line, convolution
as a superscript, complex conjugation
# number of
( average or normalized version of the variable under the bar
( complement of the variable under the bar
^ complex cepstrum of the signal (function of space)
^ complex logarithm of the signal (function of frequency)
~ estimate of the variable under the symbol
~ a variant of a function
0 00 000 rst, second, and third derivatives of the preceding function
0 a variant of a function
cross product when the related entities are vectors
8 for all
2 belongs to or is in (the set)
fg a set
subset
Symbols and Abbreviations xxxvii

T superset
S intersection
union
equivalent to
j given, conditional upon
! maps to
( gets (updated as)
) leads to
, transform pair
] closed interval, including the limits
() open interval, not including the limits
jj absolute value or magnitude
jj determinant of a matrix
kk norm of a vector or matrix
dxe ceiling operator the smallest integer x
bxc oor operator the largest integer x
1
The Nature of Biomedical Images
The human body is composed of many systems, such as the cardiovascular

system, the musculo-skeletal system, and the central nervous system. Each
system is made up of several subsystems that carry on many physiological pro-
cesses. For example, the visual system performs the task of focusing visual or
pictorial information on to the retina, transduction of the image information
into neural signals, and encoding and transmission of the neural signals to the
visual cortex. The visual cortex is responsible for interpretation of the image
information. The cardiac system performs the important task of rhythmic
pumping of blood through the arterial network of the body to facilitate the
delivery of nutrients, as well as pumping of blood through the pulmonary sys-
tem for oxygenation of the blood itself. The anatomical features of the organs
related to a physiological system often demonstrate characteristics that reect
the functional aspects of its processes as well as the well-being or integrity of
the system itself.
Physiological processes are complex phenomena, including neural or hor-
monal stimulation and control inputs and outputs that could be in the form
of physical material or information and action that could be mechanical,
electrical, or biochemical. Most physiological processes are accompanied by
or manifest themselves as signals that reect their nature and activities. Such
signals could be of many types, including biochemical in the form of hor-
mones or neurotransmitters, electrical in the form of potential or current, and
physical in the form of pressure or temperature.
Diseases or defects in a physiological system cause alterations in its nor-
mal processes, leading to pathological processes that a ect the performance,
health, and general well-being of the system. A pathological process is typ-
ically associated with signals and anatomical features that are di erent in
some respects from the corresponding normal patterns. If we possess a good
understanding of a system of interest, it becomes possible to observe the cor-
responding signals and features and assess the state of the system. The task
is not dicult when the signal is simple and appears at the outer surface of
the body. However, most systems and organs are placed well within the body
and enclosed in protective layers (for good reason!). Investigating or probing
such systems typically requires the use of some form of penetrating radiation
or invasive procedure.
1
2 Biomedical Image Analysis
1.1 Body Temperature as an Image

Most infections cause a rise in the temperature of the body, which may be
sensed easily, albeit in a relative and qualitative manner, via the palm of
one's hand. Objective or quantitative measurement of temperature requires
an instrument, such as a thermometer.
A single measurement f of temperature is a scalar, and represents the ther-
mal state of the body at a particular physical location in or on the body
denoted by its spatial coordinates (x y z ) and at a particular or single in-
stant of time t. If we record the temperature continuously in some form, such
as a strip-chart record, we obtain a signal as a one-dimensional (1D) func-
tion of time, which may be expressed in the continuous-time or analog form
as f (t). The units applicable here are o C (degrees Celsius) for the temper-
ature variable, and s (seconds) for the temporal variable t. If some means
were available to measure the temperature of the body at every spatial posi-
tion, we could obtain a three-dimensional (3D) distribution of temperature as
f (x y z). Furthermore, if we were to perform the 3D measurement at every
instant of time, we would obtain a 3D function of time as f (x y z t) this
entity may also be referred to as a four-dimensional (4D) function.
When oral temperature, for example, is measured at discrete instants of
time, it may be expressed in discrete-time form as f (nT ) or f (n), where n
is the index or measurement sample number of the array of values, and T
represents the uniform interval between the time instants of measurement. A
discrete-time signal that can take amplitude values only from a limited list of
quantized levels is called a digital signal this distinction between discrete-time
and digital signals is often ignored.
If one were to use a thermal camera and take a picture of a body, a two-
dimensional (2D) representation of the heat radiated from the body would
be obtained. Although the temperature distribution within the body (and
even on the surface of the body) is a 3D entity, the picture produced by the
camera is a 2D snapshot of the heat radiation eld. We then have a 2D spatial
function of temperature | an image | which could be represented as f (x y).
The units applicable here are o C for the temperature variable itself, and mm
(millimeters) for the spatial variables x and y. If the image were to be sam-
pled in space and represented on a discrete spatial grid, the corresponding
data could be expressed as f (m$x n$y), where $x and $y are the sam-
pling intervals along the horizontal and vertical axes, respectively (in spatial
units such as mm). It is common practice to represent a digital image simply
as f (m n), which could be interpreted as a 2D array or a matrix of values.
It should be noted at the outset that, while images are routinely treated as
arrays, matrices, and related mathematical entities, they are almost always
representative of physical or other measures of organs or of physiological pro-
The Nature of Biomedical Images 3
cesses that impose practical limitations on the range, degrees of freedom, and
other properties of the image data.
Examples: In intensive-care monitoring, the tympanic (ear drum) temper-
ature is often measured using an infrared sensor. Occasionally, when catheters
are being used for other purposes, a temperature sensor may also be intro-
duced into an artery or the heart to measure the core temperature of the
body. It then becomes possible to obtain a continuous measurement of tem-
perature, although only a few samples taken at intervals of a few minutes
may be stored for subsequent analysis. Figure 1.1 illustrates representations
of temperature measurements as a scalar, an array, and a signal that is a func-
tion of time. It is obvious that the graphical representation facilitates easier
and faster comprehension of trends in the temperature than the numerical
format. Long-term recordings of temperature can facilitate the analysis of
temperature-regulation mechanisms 15, 16].
Infrared (with wavelength in the range 3 000;5 000 nm) or thermal sensors
may also be used to capture the heat radiated or emitted from a body or a
part of a body as an image. Thermal imaging has been investigated as a
potential tool for the detection of breast cancer. A tumor is expected to be
more vascularized than its neighboring tissues, and hence could be at a slightly
higher temperature. The skin surface near the tumor may also demonstrate a
relatively high temperature. Temperature di erences of the order of 2o C have
been measured between surface regions near breast tumors and neighboring
tissues. Figure 1.2 shows thermal images of a patient with benign brocysts
and a patient with breast cancer the local increase in temperature due to a
tumor is evident in the latter case. Thermography can help in the diagnosis
of advanced cancer, but has limited success in the detection of early breast
cancer 17, 18]. Recent improvements in detectors and imaging techniques
have created a renewed interest in the application of thermography for the
detection of breast cancer 19, 20, 21, 22, 23].
Infrared imaging via a telethermographic camera has been applied to the
detection of varicocele, which is the most common cause of infertility in
men 24, 25, 26]. In normal men, the testicular temperature is about 3 ; 4 o C
below the core body temperature. In the case of varicocele, dilation of the tes-
ticular veins reduces the venous return from the scrotum, causes stagnation of
blood and edema, and leads to increased testicular temperature. In the exper-
iments conducted by Merla et al. 25], a cold patch was applied to the subject's
scrotum, and the thermal recovery curves were analyzed. The results obtained
showed that the technique was successful in detecting subclinical varicocele.
Vlaisavljevi*c 26] showed that telethermography can provide better diagnostic
accuracy in the detection of varicocele than contact thermography.
33.5 o C
(a)
Time 08 10 12 14 16 18 20 22 24
(hours)
Temperature 33.5 33.3 34.5 36.2 37.3 37.5 38.0 37.8 38.0
(o C )
(b)
39
38
37
Temperature in degrees Celsius
36
35
34
33
32
8 10 12 14 16 18 20 22 24
Time in hours
(c)
FIGURE 1.1
Measurements of the temperature of a patient presented as (a) a scalar with
one temperature measurement f at a time instant t (b) an array f (n) made
up of several measurements at di erent instants of time and (c) a signal f (t)
or f (n). The horizontal axis of the plot represents time in hours the vertical
axis gives temperature in degrees Celsius. Data courtesy of Foothills Hospital,
Calgary.
(a) (b)
FIGURE 1.2
Body temperature as a 2D image f (x y) or f (m n). The images illustrate the distribution of surface temperature measured
using an infrared camera operating in the 3 000 ; 5 000 nm wavelength range. (a) Image of a patient with pronounced
vascular features and benign brocysts in the breasts. (b) Image of a patient with a malignant mass in the upper-outer
quadrant of the left breast. Images courtesy of P. Hoekstra, III, Therma-Scan, Inc., Huntington Woods, MI.
5
The thermal images shown in Figure 1.2 serve to illustrate an important
distinction between two major categories of medical images:
anatomical or physical images, and
functional or physiological images.
The images illustrate the notion of body temperature as a signal or image.
Each point in the images in Figure 1.2 represents body temperature, which
is related to the ongoing physiological or pathological processes at the cor-
responding location in the body. A thermal image is, therefore, a functional
image. An ordinary photograph obtained with reected light, on the other
hand, would be a purely anatomical or physical image. More sophisticated
techniques that provide functional images related to circulation and various
physiological processes are described in the following sections.
1.2 Transillumination
Transillumination, diaphanography, and diaphanoscopy involve the shining of
visible light or near-infrared radiation through a part of the body, and viewing
or imaging the transmitted radiation. The technique has been investigated
for the detection of breast cancer, the attractive feature being the use of
nonionizing radiation 27]. The use of near-infrared radiation appears to have
more potential than visible light, due to the observation that nitrogen-rich
compounds preferentially absorb (or attenuate) infrared radiation. The fat
and broglandular tissue in the mature breast contain much less nitrogen than
malignant tissues. Furthermore, the hemoglobin in blood has a high nitrogen
content, and tumors are more vascularized than normal tissues. For these
reasons, breast cancer appears as a relatively dark region in a transilluminated
image.
The e ectiveness of transillumination is limited by scatter and ine ective
penetration of light through a large organ such as the breast. Transillumina-
tion has been found to be useful in di erentiating between cystic (uid- lled)
and solid lesions however, the technique has had limited success in distin-
guishing malignant tumors from benign masses 18, 28, 29].
1.3 Light Microscopy

Studies of the ne structure of biological cells and tissues require signi cant
magni cation for visualization of the details of interest. Useful magni cation
of up to 1 000 may be obtained via light microscopy by the use of combina-
tions of lenses. However, the resolution of light microscopy is reduced by the
following factors 30]:
Diraction: The bending of light at edges causes blurring the image
of a pinhole appears as a blurred disc known as the Airy disc.
Astigmatism: Due to nonuniformities in lenses, a point may appear

as an ellipse.
Chromatic aberration: Electromagnetic (EM) waves of di erent wave-

length or energy that compose the ordinarily used white light converge
at di erent focal planes, thereby causing enlargement of the focal point.
This e ect may be corrected for by using monochromatic light. See
Section 3.9 for a description of confocal microscopy.
Spherical aberration: The rays of light arriving at the periphery

of a lens are refracted more than the rays along the axis of the lens.
This causes the rays from the periphery and the axis not to arrive at a
common focal point, thereby resulting in blurring. The e ect may be
reduced by using a small aperture.
Geometric distortion: Poorly crafted lenses may cause geometric

distortion such as the pin-cushion e ect and barrel distortion.
Whereas the best resolution achievable by the human eye is of the order
of 0:1 ; 0:2 mm, light microscopes can provide resolving power up to about
0:2
m.
Example: Figure 1.3 shows a rabbit ventricular myocyte in its relaxed
state as seen through a light microscope at a magni cation of about 600.
The experimental setup was used to study the contractility of the myocyte
with the application of electrical stimuli 31].
Example: Figure 1.4 shows images of three-week-old scar tissue and forty-
week-old healed tissue samples from rabbit ligaments at a magni cation of
about 300. The images demonstrate the alignment patterns of the nuclei of
broblasts (stained to appear as the dark objects in the images): the three-
week-old scar tissue has many broblasts that are scattered in di erent direc-
tions, whereas the forty-week-old healed sample has fewer broblasts that are
well-aligned along the length of the ligament (the horizontal edge of the im-
age). The appearance of the forty-week-old sample is closer to that of normal
samples than that of the three-week-old sample. Images of this nature have
been found to be useful in studying the healing and remodeling processes in
ligaments 32].
FIGURE 1.3
A single ventricular myocyte (of a rabbit) in its relaxed state. The width
(thickness) of the myocyte is approximately 15
m. Image courtesy of R.
Clark, Department of Physiology and Biophysics, University of Calgary.
(a)
(b)
FIGURE 1.4
(a) Three-week-old scar tissue sample, and (b) forty-week-old healed tissue
sample from rabbit medial collateral ligaments. Images courtesy of C.B.
Frank, Department of Surgery, University of Calgary.
1.4 Electron Microscopy

Accelerated electrons possess EM wave properties, with the wavelength
given by = mv h , where h is Planck's constant, m is the mass of the electron,
and v is the electron's velocity this relationship reduces to = 1p:23
V , where
V is the accelerating voltage 30]. At a voltage of 60 kV , an electron beam
has an e ective wavelength of about 0:005 nm, and a resolving power limit
of about 0:003 nm. Imaging at a low kV provides high contrast but low
resolution, whereas imaging at a high kV provides high resolution due to
smaller wavelength but low contrast due to higher penetrating power. In
addition, a high-kV beam causes less damage to the specimen as the faster
electrons pass through the specimen in less time than with a low-kV beam.
Electron microscopes can provide useful magni cation of the order of 106 ,
and may be used to reveal the ultrastructure of biological tissues. Electron
microscopy typically requires the specimen to be xed, dehydrated, dried,
mounted, and coated with a metal.
Transmission electron microscopy: A transmission electron microscope
(TEM) consists of a high-voltage electron beam generator, a series of EM
lenses, a specimen holding and changing system, and a screen- lm holder, all
enclosed in vacuum. In TEM, the electron beam passes through the specimen,
is a ected in a manner similar to light, and the resulting image is captured
through a screen- lm combination or viewed via a phosphorescent viewing
screen.
Example: Figure 1.5 shows TEM images of collagen bers (in cross-
section) in rabbit ligament samples. The images facilitate analysis of the
diameter distribution of the bers 33]. Scar samples have been observed to
have an almost uniform distribution of ber diameter in the range 60 ; 70 nm,
whereas normal samples have an average diameter of about 150 nm over a
broader distribution. Methods for the detection and analysis of circular ob-
jects are described in Sections 5.6.1, 5.6.3, and 5.8.
Example: In patients with hematuria, the glomerular basement membrane
of capillaries in the kidney is thinner (< 200 nm) than the normal thickness
of the order of 300 nm 34]. Investigation of this feature requires needle-core
biopsy of the kidney and TEM imaging. Figure 1.6 shows a TEM image
of a capillary of a normal kidney in cross-section. Figure 1.7 (a) shows an
image of a sample with normal membrane thickness Figure 1.7 (b) shows
an image of a sample with reduced and variable thickness. Although the
ranges of normal and abnormal membrane thickness have been established by
several studies 34], the diagnostic decision process is subjective methods for
objective and quantitative analysis are desired in this application.
(a)
(b)
FIGURE 1.5
TEM images of collagen bers in rabbit ligament samples at a magni cation
of approximately 30 000. (a) Normal and (b) scar tissue. Images courtesy
of C.B. Frank, Department of Surgery, University of Calgary.
FIGURE 1.6
TEM image of a kidney biopsy sample at a magni cation of approximately
3 500. The image shows the complete cross-section of a capillary with nor-
mal membrane thickness. Image courtesy of H. Benediktsson, Department of
Pathology and Laboratory Medicine, University of Calgary.
(a) (b)
FIGURE 1.7
TEM images of kidney biopsy samples at a magni cation of approximately 8 000. (a) The sample shows normal capillary
membrane thickness. (b) The sample shows reduced and varying membrane thickness. Images courtesy of H. Benediktsson,
Department of Pathology and Laboratory Medicine, University of Calgary.
13
Scanning electron microscopy: A scanning electron microscope (SEM)
is similar to a TEM in many ways, but uses a nely focused electron beam
with a diameter of the order of 2 nm to scan the surface of the specimen. The
electron beam is not transmitted through the specimen, which could be fairly
thick in SEM. Instead, the beam is used to scan the surface of the specimen
in a raster pattern, and the secondary electrons that are emitted from the
surface of the sample are detected and ampli ed through a photomultiplier
tube (PMT), and used to form an image on a cathode-ray tube (CRT). An
SEM may be operated in di erent modes to detect a variety of signals emitted
from the sample, and may be used to obtain images with a depth of eld of
several mm.
Example: Figure 1.8 illustrates SEM images of collagen bers in rabbit

ligament samples (freeze-fractured surfaces) 35]. The images are useful in an-
alyzing the angular distribution of bers and the realignment process during
healing after injury. It has been observed that collagen bers in a normal lig-
ament are well aligned, that bers in scar tissue lack a preferred orientation,
and that organization and alignment return toward their normal patterns dur-
ing the course of healing 36, 37, 35]. Image processing methods for directional
analysis are described in Chapter 8.
(a) (b)
FIGURE 1.8
SEM images of collagen bers in rabbit ligament samples at a magni cation
of approximately 4 000. (a) Normal and (b) scar tissue. Reproduced with
permission from C.B. Frank, B. MacFarlane, P. Edwards, R. Rangayyan, Z.Q.
Liu, S. Walsh, and R. Bray, \A quantitative analysis of matrix alignment
in ligament scars: A comparison of movement versus immobilization in an
immature rabbit model", Journal of Orthopaedic Research, 9(2): 219 { 227,
1991. c Orthopaedic Research Society.
1.5 X-ray Imaging

The medical diagnostic potential of X rays was realized soon after their dis-
covery by Roentgen in 1895. (See Robb 38] for a review of the history of
X-ray imaging.) In the simplest form of X-ray imaging or radiography, a 2D
projection (shadow or silhouette) of a 3D body is produced on lm by irra-
diating the body with X-ray photons 4, 3, 5, 6]. This mode of imaging is
referred to as projection or planar imaging. Each ray of X-ray photons is
attenuated by a factor depending upon the integral of the linear attenuation
coecient along the path of the ray, and produces a corresponding gray level
(or signal) at the point hit on the lm or the detecting device used.
Considering the ray path marked as AB in Figure 1.9, let Ni denote the
number of X-ray photons incident upon the body being imaged, within a
speci ed time interval. Let us assume that the X rays are mutually parallel,
with the X-ray source at a large distance from the subject or object being
imaged. Let No be the corresponding number of photons exiting the body.
Then, we have Z
No = Ni exp ;
(x y) ds (1.1)
rayAB
or Z
Ni :

(x y) ds = ln N (1.2)
rayAB o
The equations above are modi ed versions of Beer's law (also known as the
Beer-Lambert law) on the attenuation of X rays due to passage through a
medium. The ray AB lies in the sectional plane PQRS the mutually parallel
rays within the plane PQRS are represented by the coordinates (t s) that are
at an angle with respect to the (x y) coordinates indicated in Figure 1.9,
with the s axis being parallel to the rays. Then, s = ;x sin + y cos . The
variable of integration ds represents the elemental distance along the ray, and
the integral is along the ray path AB from the X-ray source to the detector.
(See Section 9.1 for further details on this notation.) The quantities Ni and
No are Poisson variables it is assumed that their values are large for the
equations above to be applicable. The function
(x y) represents the linear
attenuation coecient at (x y) in the sectional plane PQRS. The value of
(x y) depends upon the density of the object or its constituents along the
ray path, as well as the frequency (or wavelength or energy) of the radiation
used. Equation 1.2 assumes the use of monochromatic or monoenergetic X
rays.
A measurement of the exiting X rays (that is, No , and Ni for reference) thus
gives us only an integral of
(x y) over the ray path. The internal details of
the body along the ray path are compressed onto a single point on the lm
or a single measurement. Extending the same argument to all ray paths,
z
z
Q’ Q R
No ds
S A Ni
B P
P’ y
y X rays
x
2D projection 3D object
FIGURE 1.9
An X-ray image or a typical radiograph is a 2D projection or planar image
of a 3D object. The entire object is irradiated with X rays. The projection
of a 2D cross-sectional plane PQRS of the object is a 1D pro le P'Q' of the
2D planar image. See also Figures 1.19 and 9.1. Reproduced, with permis-
sion, from R.M. Rangayyan and A. Kantzas, \Image reconstruction", Wiley
Encyclopedia of Electrical and Electronics Engineering, Supplement 1, Editor:
John G. Webster, Wiley, New York, NY, pp 249{268, 2000. c This material
is used by permission of John Wiley & Sons, Inc.
we see that the radiographic image so produced is a 2D planar image of the

3D object, where the internal details are superimposed. In the case that the
rays are parallel to the x axis (as in Figure 1.9), we have = 90o , s = ;x,
ds = ;dx, and the planar image
Z
g(y z) = ;
(x y z) dx: (1.3)
Ignoring the negative sign, we see that the 3D object is reduced to (or inte-
grated into) a 2D planar image by the process of radiographic imaging.
The most commonly used detector in X-ray imaging is the screen- lm com-
bination 5, 6]. The X rays exiting from the body being imaged strike a
uorescent (phosphor) screen made of compounds of rare-earth elements such
as lanthanum oxybromide or gadolinium oxysul de, where the X-ray photons
are converted into visible-light photons. A light-sensitive lm that is placed in
contact with the screen (in a light-tight cassette) records the result. The lm
contains a layer of silver-halide emulsion with a thickness of about 10
m.
The exposure or blackening of the lm depends upon the number of light
photons that reach the lm.
A thick screen provides a high eciency of conversion of X rays to light,
but causes loss of resolution due to blurring (see Figure 1.10). The typical
thickness of the phosphor layer in screens is in the range 40 ; 100
m. Some
receiving units make use of a lm with emulsion on both sides that is sand-
wiched between two screens: the second screen (located after the lm along
the path of propagation of the X rays) converts the X-ray photons not af-
fected by the rst screen into light, and thereby increases the eciency of
the receiver. Thin screens may be used in such dual-screen systems to achieve
higher conversion eciency (and lower dose to the patient) without sacri cing
resolution.
X rays
A
light B screen
film
FIGURE 1.10
Blur caused by a thick screen. Light emanating from point A in the screen is
spread over a larger area on the lm than that from point B.
A uoroscopy system uses an image intensi er and a video camera in place

of the lm to capture the image and display it on a monitor as a movie or
video 5, 6]. Images are acquired at a rate of 2 ; 8 frames=s (fps), with
the X-ray beam pulsed at 30 ; 100 ms per frame. In computed radiography
(CR), a photo-stimulable phosphor plate (made of europium-activated barium
uorohalide) is used instead of lm to capture and temporarily hold the image
pattern. The latent image pattern is then scanned using a laser and digitized.
In digital radiography (DR), the lm or the entire screen- lm combination is
replaced with solid-state electronic detectors 39, 40, 41, 42].
Examples: Figures 1.11 (a) and (b) show the posterior-anterior (PA, that
is, back-to-front) and lateral (side-to-side) X-ray images of the chest of a
patient. Details of the ribs and lungs, as well as the outline of the heart,
are visible in the images. Images of this type are useful in visualizing and
discriminating between the air- lled lungs, the uid- lled heart, the ribs, and
vessels. The size of the heart may be assessed in order to detect enlargement
of the heart. The images may be used to detect lesions in the lungs and
fracture of the ribs or the spinal column, and to exclude the presence of uid
in the thoracic cage. The use of two views assists in localizing lesions: use
of the PA view only, for example, will not provide information to decide if a
tumor is located toward the posterior or anterior of the patient.
18
Biomedical Image Analysis
(a) (b)
FIGURE 1.11
(a) Posterior-anterior and (b) lateral chest X-ray images of a patient. Images courtesy of Foothills Hospital, Calgary.
The following paragraphs describe some of the physical and technical con-
siderations in X-ray imaging 4, 5, 6, 43, 44].
Target and focal spot: An electron beam with energy in the range
of 20 ; 140 keV is used to produce X rays for diagnostic imaging. The
typical target materials used are tungsten and molybdenum. The term
\focal spot" refers to the area of the target struck by the electron beam
to generate X rays however, the nominal focal spot is typically ex-
pressed in terms of its diameter in mm as observed in the imaging plane
(on the lm). A small focal spot is desired in order to obtain a sharp
image, especially in magni cation imaging. (See also Section 2.9 and
Figure 2.18.) Typical focal spot sizes in radiography lie in the range of
0:1 ; 2 mm. A focal spot size of 0:1 ; 0:3 mm is desired in mammogra-
phy.
Energy: The penetrating capability of an X-ray beam is mainly de-
termined by the accelerating voltage applied to the electron beam that
impinges the target in the X-ray generator. The commonly used indi-
cator of penetrating capability (often referred to as the \energy" of the
X-ray beam) is kV p, standing for kilo-volt-peak. The higher the kV p,
the more penetrating the X-ray beam will be. The actual unit of en-
ergy of an X-ray photon is the electron volt or eV , which is the energy
gained by an electron when a potential of 1 V is applied to it. The kV p
measure relates to the highest possible X-ray photon energy that may
be achieved at the voltage used.
Low-energy X-ray photons are absorbed at or near the skin surface, and
do not contribute to the image. In order to prevent such unwanted
radiation, a lter is used at the X-ray source to absorb low-energy X
rays. Typical lter materials are aluminum and molybdenum.
Imaging of soft-tissue organs such as the breast is performed with low-
energy X rays in the range of 25 ; 32 kV p 45]. The use of a higher kV p
would result in low di erential attenuation and poor tissue-detail visibil-
ity or contrast. A few other energy levels used in projection radiography
are, for imaging the abdomen: 60 ; 100 kV p chest: 80 ; 120 kV p and
skull: 70 ; 90 kV p. The kV p to be used depends upon the distance
between the X-ray source and the patient, the size (thickness) of the
patient, the type of grid used, and several other factors.
Exposure: For a given tube voltage (kV p), the total number of X-ray
photons released at the source is related to the product of the tube cur-
rent (mA) and the exposure time (s), together expressed as the product
mAs. As a result, for a given body being imaged, the number of pho-
tons that arrive at the lm is also related to the mAs quantity. A low
mAs results in an under-exposed lm (faint or light image), whereas a
high mAs results in an over-exposed or dark image (as well as increased
X-ray dose to the patient). Typical exposure values lie in the range
of 2 ; 120 mAs. Most imaging systems determine automatically the
required exposure for a given mode of imaging, patient size, and kV p
setting. Some systems use an initial exposure of the order of 5 ms to
estimate the penetration of the X rays through the body being imaged,
and then determine the required exposure.
Beam hardening: The X rays used in radiographic imaging are typi-
cally not monoenergetic that is, they possess X-ray photons over a cer-
tain band of frequencies or EM energy levels. As the X rays propagate
through a body, the lower-energy photons get absorbed preferentially,
depending upon the length of the ray path through the body and the
attenuation characteristics of the tissues along the path. Thus, the X
rays that pass through the object at longer distances from the source
will possess relatively fewer photons at lower-energy levels than at the
point of entry into the object (and hence a relatively higher concentra-
tion of higher-energy photons). This e ect is known as beam hardening,
and leads to incorrect estimation of the attenuation coecient in com-
puted tomography (CT) imaging. The e ect of beam hardening may be
reduced by pre ltering or prehardening the X-ray beam and narrowing
its spectrum. The use of monoenergetic X rays from a synchrotron or a
laser obviates this problem.
Scatter and the use of grids: As an X-ray beam propagates through
a body, photons are lost due to absorption and scattering at each point
in the body. The angle of the scattered photon at a given point along the
incoming beam is a random variable, and hence the scattered photon
contributes to noise at the point where it strikes the detector. Fur-
thermore, scattering results in the loss of contrast of the part of the
object where X-ray photons were scattered from the main beam. The
noise e ect of the scattered radiation is signi cant in gamma-ray emis-
sion imaging, and requires speci c methods to improve the quality of
the image 4, 46]. The e ect of scatter may be reduced by the use of
grids, collimation, or energy discrimination due to the fact that the scat-
tered (or secondary) photons usually have lower energy levels than the
primary photons.
A grid consists of an array of X-ray absorbing strips that are mutually
parallel if the X rays are in a parallel beam, as in chest imaging (see
Figures 1.12 and 1.13), or are converging toward the X-ray source in the
case of a diverging beam (as in breast imaging, see Figure 1.15). Lattice
or honeycomb grids with parallel strips in criss-cross patterns are also
used in mammography. X-ray photons that arrive via a path that is not
aligned with the grids will be stopped from reaching the detector.
A typical grid contains thin strips of lead or aluminum with a strip den-
sity of 25 ; 80 lines=cm and a grid height:strip width ratio in the range
parallel X rays
A
F
E parallel grid
B C
screen-film
A’ D
FIGURE 1.12
Use of parallel grids to reduce scatter. X rays that are parallel to the grids
reach the lm for example, line AA'. Scattered rays AB, AC, and AE have
been blocked by the grids however, the scattered ray AD has reached the lm
in the illustration.
of 5:1 to 12:1. The space between the grids is lled with low-attenuation
material such as wood. A stationary grid produces a line pattern that is
superimposed upon the image, which would be distracting. Figure 1.13
(a) shows a part of an image of a phantom with the grid artifact clearly
visible. (An image of the complete phantom is shown in Figure 1.14.)
Grid artifact is prevented in a reciprocating grid, where the grid is moved
about 20 grid spacings during exposure: the movement smears the grid
shadow and renders it invisible on the image. Figure 1.13 (b) shows
an image of the same object as in part (a), but with no grid artifact.
Low levels of grid artifact may appear in images if the bucky that holds
the grid does not move at a uniform pace or starts moving late or ends
movement early with respect to the X-ray exposure interval. A major
disadvantage of using grids is that it requires approximately two times
the radiation dose required for imaging techniques without grids. Fur-
thermore, the contrast of ne details is reduced due to the smeared
shadow of the grid.
Photon detection noise: The interaction between an X-ray beam
and a detector is governed by the same rules as for interaction with
any other matter: photons are lost due to scatter and absorption, and
some photons may pass through una ected (or undetected). The small
size of the detectors in DR and CT imaging reduces their detection
eciency. Scattered and undetected photons cause noise in the mea-
surement for detailed analysis of noise in X-ray detection, refer to Bar-
rett and Swindell 3], Macovski 5], and Cho et al. 4]. More details on
noise in medical images and techniques to remove noise are presented in
Chapter 3.
Ray stopping by heavy implants: If the body being imaged contains
extremely heavy parts or components, such as metal screws or pins in
bones and surgical clips that are nearly X-ray-opaque and entirely stop
the incoming X-ray photons, no photons would be detected at the cor-
responding point of exit from the body. The attenuation coecient for
the corresponding path would be inde nite, or within the computational
context, in nity. Then, a reconstruction algorithm would not be able
to redistribute the attenuation values over the points along the corre-
sponding ray path in the reconstructed image. This leads to streaking
artifacts in CT images.
Two special techniques for enhanced X-ray imaging | digital subtraction
angiography (DSA) and dual-energy imaging | are described in Sections 4.1
and 4.2, respectively.
1.5.1 Breast cancer and mammography

Breast cancer: Cancer is caused when a single cell or a group of cells
escapes from the usual controls that regulate cellular growth, and begins to
multiply and spread. This activity results in a mass, tumor, or neoplasm.
Many masses are benign that is, the abnormal growth is restricted to a single,
circumscribed, expanding mass of cells. Some tumors are malignant that is,
the abnormal growth invades the surrounding tissues and may spread, or
metastasize, to distant areas of the body. Although benign masses may lead
to complications, malignant tumors are usually more serious, and it is for
these tumors that the term \cancer" is used. The majority of breast tumors
will have metastasized before reaching a palpable size.
Although curable, especially when detected at early stages, breast cancer
is a major cause of death in women. An important factor in breast cancer
is that it tends to occur earlier in life than other types of cancer and other
major diseases 47, 48]. Although the cause of breast cancer has not yet
been fully understood, early detection and removal of the primary tumor are
essential and e ective methods to reduce mortality, because, at such a point
in time, only a few of the cells that departed from the primary tumor would
have succeeded in forming secondary tumors 49]. When breast tumors are
detected by the a ected women themselves (via self-examination), most of the
tumors would have metastasized 50].
If breast cancer can be detected by some means at an early stage, while it is
clinically localized, the survival rate can be dramatically increased. However,
(a)
(b)
FIGURE 1.13
X-ray images of a part of a phantom: (a) with, and (b) without grid artifact.
Image courtesy of L.J. Hahn, Foothills Hospital, Calgary. See also Figure 1.14.
FIGURE 1.14
X-ray image of the American College of Radiology (ACR) phantom for mam-
mography. The pixel-value range 117 210] has been linearly stretched to the
display range 0 255] to show the details. Image courtesy of S. Bright, Sun-
nybrook & Women's College Health Sciences Centre, Toronto, ON, Canada.
See also Figure 1.13.
such early breast cancer is generally not amenable to detection by physical ex-
amination and breast self-examination. The primary role of an imaging tech-
nique is thus the detection of lesions in the breast 29]. Currently, the most
e ective method for the detection of early breast cancer is X-ray mammogra-
phy. Other modalities, such as ultrasonography, transillumination, thermog-
raphy, CT, and magnetic resonance imaging (MRI) have been investigated for
breast cancer diagnosis, but mammography is the only reliable procedure for
detecting nonpalpable cancers and for detecting many minimal breast cancers
when they appear to be curable 18, 28, 29, 51]. Therefore, mammography
has been recommended for periodic screening of asymptomatic women. Mam-
mography has gained recognition as the single most successful technique for
the detection of early, clinically occult breast cancer 52, 53, 54, 55, 56].
X-ray imaging of the breast: The technique of using X rays to ob-
tain images of the breast was rst reported by Warren in 1930, after he had
examined 100 women using sagital views 57]. Because of the lack of a re-
producible method for obtaining satisfactory images, this technique did not
make much progress until 1960, when Egan 58] reported on high-mA and
low-kV p X-ray sources that yielded reproducible images on industrial lm. It
was in the mid-1960s that the rst modern X-ray unit dedicated to mammog-
raphy was developed. Since then, remarkable advances have led to a striking
improvement in image quality and a dramatic reduction in radiation dose.
A major characteristic of mammograms is low contrast, which is due to
the relatively homogeneous soft-tissue composition of the breast. Many ef-
forts have been focused on developing methods to enhance contrast. In an
alternative imaging method known as xeromammography, a selenium-coated
aluminum plate is used as the detector 6]. The plate is initially charged
to about 1 000 V . Exposure to the X rays exiting the patient creates a
charge pattern on the plate due to the liberation of electrons and ions. The
plate is then sprayed with an ionized toner, the pattern of which is trans-
ferred to plastic-coated paper. Xeromammograms provide wide latitude and
edge enhancement, which lead to improved images as compared to screen- lm
mammography. However, xeromammography results in a higher dose to the
subject, and has not been in much use since the 1980s.
A typical mammographic imaging system is shown schematically in Fig-
ure 1.15. Mammography requires high X-ray beam quality (a narrow-band
or nearly monochromatic beam), which is controlled by the tube target ma-
terial (molybdenum) and beam ltration with molybdenum. E ective breast
compression is an important factor in reducing scattered radiation, creating
as uniform a density distribution as possible, eliminating motion, and sepa-
rating mammary structures, thereby increasing the visibility of details in the
image. The use of grids speci cally designed for mammography can further
reduce scattered radiation and improve subject contrast, which is especially
signi cant when imaging thick, dense breasts 59].
Generally, conventional screen- lm mammography is performed with the
breast directly in contact with the screen- lm cassette, producing essentially
X-ray source (target)

Filter
Collimating
diaphragm
Breast
compression
paddle
Compressed breast
Focused grid
Screen-film
cassette
FIGURE 1.15
A typical mammography setup.
life-size images. The magni cation technique, on the other hand, interposes
an air gap between the breast and the lm, so that the projected radiographic
image is enlarged. Magni cation produces ne-detail images containing addi-
tional anatomical information that may be useful in re ning mammographic
diagnosis, especially in cases where conventional imaging demonstrates un-
certain or equivocal ndings 60]. As in the grid method, the advantages
of magni cation imaging are achieved at the expense of increased radiation
exposure. Therefore, the magni cation technique is not used routinely.
Screen- lm mammography is now the main tool for the detection of early
breast cancer. The risk of radiation is still a matter of concern, although
there is no direct evidence of breast cancer risk from the low-dose radiation
exposure of mammography. Regardless, technological advances in mammog-
raphy continue to be directed toward minimizing radiation exposure while
maintaining the high quality of the images.
Examples: Figures 1.16 (a) and (b) show the cranio-caudal (CC) and
medio-lateral-oblique (MLO) views of the same breast of a subject. The MLO
view demonstrates architectural distortion due to a spiculated tumor near the
upper right-hand corner edge.
Mammograms are analyzed by radiologists specialized in mammography. A
normal mammogram typically depicts converging patterns of broglandular
tissues and vessels. Any feature that causes a departure from or distortion
with reference to the normal pattern is viewed with suspicion and analyzed
with extra attention. Features such as calci cations, masses, localized increase
in density, architectural distortion, and asymmetry between the left and right
breast images are carefully analyzed.
Several countries and states have instituted breast cancer screening pro-
grams where asymptomatic women within a certain age group are invited to
participate in regular mammographic examinations. Screen Test | Alberta
Program for the Early Detection of Breast Cancer 61] is an example of such
a program. Several applications of image processing and pattern analysis
techniques for mammographic image analysis and breast cancer detection are
described in the chapters to follow.
1.6 Tomography
The problem of visualizing the details of the interior of the human body
noninvasively has always been of interest, and within a few years after the
discovery of X rays by Rontgen in 1895, techniques were developed to image
sectional planes of the body. The techniques of laminagraphy, planigraphy,
or \classical" tomography 38, 62] used synchronous movement of the X-ray
source and lm in such a way as to produce a relatively sharp image of a
(a) (b)
FIGURE 1.16
(a) Cranio-caudal (CC) and (b) medio-lateral oblique (MLO) mammograms
of the same breast of a subject. Images courtesy of Screen Test | Alberta
Program for the Early Detection of Breast Cancer 61].
single focal plane of the object, with the images of all other planes being
blurred. Figure 1.17 illustrates a simple linear-motion system, where the X-
ray source and lm cassette move along straight-line paths so as to maintain
the longitudinal (coronal) plane, indicated by the straight line AB, in focus.
It is seen that the X rays along the paths X1-A and X2-A strike the same
physical spot A1 = A2 on the lm, and that the rays along the paths X1-B
and X2-B strike the same spot B1 = B2. On the other hand, for the point C
in a di erent plane, the rays along the paths X1-C and X2-C strike di erent
points C1 6= C2 on the lm. Therefore, the details in the plane AB remain
in focus and cause a strong image, whereas the details in the other planes are
smeared all over the lm. The smearing of information from the other planes
of the object causes loss of contrast in the plane of interest. The development
of CT imaging rendered lm-based tomography obsolete.
Example: Figure 1.18 shows a tomographic image of a patient in a longitu-
dinal (coronal) plane through the chest. Images of this nature provided better
visualization and localization of lesions than regular X-ray projection images,
and permitted the detection of masses in bronchial tubes and air ducts.
X2 X1
Path of source movement X-ray
source
A B Patient
Table
C1 C2
Film cassette
A1 B1 A2 B2
Path of film movement
FIGURE 1.17
Synchronized movement of the X-ray source and lm to obtain a tomographic
image of the focal plane indicated as AB. Adapted from Robb 38].
Computed tomography: The technique of CT imaging was developed

during the late 1960s and the early 1970s, producing images of cross-sections of
the human head and body as never seen before (noninvasively and nondestruc-
FIGURE 1.18
Tomographic image of a patient in a longitudinal sectional plane through the
chest. Reproduced with permission from R.A. Robb, \X-ray computed tomog-
raphy: An engineering synthesis of multiscienti c principles", CRC Critical
Reviews in Biomedical Engineering, 7:264{333, March 1982. c CRC Press.
tively!). In the simplest form of CT imaging, only the desired cross-sectional
plane of the body is irradiated using a nely collimated ray of X-ray photons
(see Figure 1.19), instead of irradiating the entire body with a 3D beam of X
rays as in ordinary radiography (Figure 1.9). The fundamental radiographic
equation for CT is the same as Equation 1.2. Ray integrals are measured at
many positions and angles around the body, scanning the body in the pro-
cess. The principle of image reconstruction from projections, described in
detail in Chapter 9, is then used to compute an image of a section of the
body: hence the name computed tomography. (See Robb 38] for an excellent
review of the history of CT imaging see also Rangayyan and Kantzas 63]
and Rangayyan 64].)
Q’
Q R
No S Ni
B P A
P’ X rays
1D Projection 2D Section
FIGURE 1.19
In the basic form of CT imaging, only the cross-sectional plane of interest is
irradiated with X rays. The projection of the 2D cross-sectional plane PQRS
of the object is the 1D pro le P'Q' shown. Compare this case with the planar
imaging case illustrated in Figure 1.9. See also Figure 9.1. Reproduced, with
permission, from R.M. Rangayyan and A. Kantzas, \Image reconstruction",
Wiley Encyclopedia of Electrical and Electronics Engineering, Supplement 1,
Editor: John G. Webster, Wiley, New York, NY, pp 249{268, 2000. c This
material is used by permission of John Wiley & Sons, Inc.
Figure 1.20 depicts some of the scanning procedures employed: Figure 1.20
(a) shows the translate-rotate scanning geometry for parallel-ray projections
Figure 1.20 (b) shows the translate-rotate scanning geometry with a small fan-
beam detector array Figure 1.20 (c) shows the rotate-only scanning geometry
for fan-beam projections and Figure 1.20 (d) shows the rotate-only scanning
geometry for fan-beam projections using a ring of detectors. A more recently
developed scanner specialized for cardiovascular imaging 65, 66] completely
eliminates mechanical scanning movement to reduce the scanning time by
employing electronically steered X-ray microbeams and rings of detectors, as
illustrated schematically in Figure 1.21.
The fundamental principle behind CT, namely, image reconstruction from
projections, has been known for close to 100 years, since the exposition of
the topic by Radon 67, 68] in 1917. More recent developments in the sub-
ject arose in the 1950s and 1960s from the works of a number of researchers
in diverse applications. Some of the important publications in this area are
the works of Cormack on the representation of a function by its line inte-
grals 69, 70] Bracewell and Riddle on the reconstruction of brightness dis-
tributions of astronomical bodies from fan-beam scans at various angles 71]
Crowther et al. 72] and De Rosier and Klug 73] on the reconstruction of
3D images of viruses from electron micrographs Ramachandran and Laksh-
minarayanan 74] on the convolution backprojection technique and Gordon
et al. 75] on algebraic reconstruction techniques. Pioneering works on the
development of practical scanners for medical applications were performed
by Oldendorf 76], Houns eld 77], and Ambrose 78]. X-ray CT was well
established as a clinical diagnostic tool by the early 1970s.
Once a sectional image is obtained, the process may be repeated to obtain
a series of sectional images of the 3D body or object being investigated. CT
imaging a 3D body may be accomplished by reconstructing one 2D section
at a time through the use of 1D projections. In the Dynamic Spatial Recon-
structor 38, 79], 2D projection images are obtained on a uorescent screen
via irradiation of the entire portion of interest of the body. In single-photon
emission computed tomography (SPECT), several 2D projection images are
obtained using a gamma camera 4, 5, 46, 80, 81]. In these cases, the projec-
tion data of several sectional planes are acquired simultaneously: each row of
a given 2D projection or planar image provides the 1D projection data of a
sectional plane of the body imaged (see Figure 1.9). Many sectional images
may then be reconstructed from the set of 2D planar images acquired.
Other imaging modalities used for projection data collection are ultrasound
(time of ight or attenuation), magnetic resonance (MR), nuclear emission
(gamma rays or positrons), and light 4, 5, 80, 82, 83, 84, 85]. Techniques
using nonionizing radiation are of importance in imaging pregnant women.
Whereas the physical parameter imaged may di er between these modalities,
once the projection data are acquired, the mathematical image reconstruction
procedure could be almost the same. A few special considerations in imaging
with di racting sources are described in Section 9.5. The characteristics of the
data acquired in nuclear medicine, ultrasound, and MR imaging are described
in Sections 1.7, 1.8, and 1.9, respectively.
Examples: Figure 1.22 shows a CT scan of the head of a patient. The
image displays, among other features, the ventricles of the brain. CT images
of the head are useful in the detection of abnormalities in the brain and skull,
including bleeding in brain masses, calci cations, and fractures in the cranial
vault.
FIGURE 1.20
(a) Translate-rotate scanning geometry for parallel-ray projections
(b) translate-rotate scanning geometry with a small fan-beam detector array
(c) rotate-only scanning geometry for fan-beam projections (d) rotate-only
scanning geometry for fan-beam projections using a ring of detectors. Repro-
duced with permission from R.A. Robb, \X-ray computed tomography: An
engineering synthesis of multiscienti c principles", CRC Critical Reviews in
Biomedical Engineering, 7:264{333, March 1982. c CRC Press.
FIGURE 1.21
Electronic steering of an X-ray beam for motion-free scanning and CT imag-
ing. Reproduced with permission from D.P. Boyd, R.G. Gould, J.R. Quinn,
and R. Sparks, \Proposed dynamic cardiac 3-D densitometer for early detec-
tion and evaluation of heart disease", IEEE Transactions on Nuclear Science,
26(2):2724{2727, 1979. c IEEE. See also Robb 38].
FIGURE 1.22
CT image of a patient showing the details in a cross-section through the head
(brain). Image courtesy of Foothills Hospital, Calgary.
Figure 1.23 illustrates the CT image of the abdomen of a patient. The image
shows the stomach, gall bladder, liver, spleen, kidneys, intestines, and the
spinal column. The air-uid interface in the stomach is clearly visible. Images
of this type are useful in detecting several abnormalities in the abdomen,
including gallstones, kidney stones, and tumors in the liver.
Figure 1.24 shows two renditions of the same CT image of the chest of a
patient: image (a) has been scaled (or windowed details provided in Sections
4.4.2 and 9.6) to illustrate the details of the lungs, whereas image (b) has
been scaled to display the mediastinum in relatively increased detail. Image
(a) shows the details of the distal branches of the pulmonary arteries. CT
images of the chest facilitate the detection of the distortion of anatomy due
to intrathoracic or extrapulmonary uid collection, or due to ruptured lungs.
They also aid in the detection of lung tumors and blockage of the pulmonary
arteries due to thrombi.
CT is an imaging technique that has revolutionized the eld of medical
diagnostics. CT has also found applications in many other areas such as non-
destructive evaluation of industrial and biological specimens, radioastronomy,
light and electron microscopy, optical interferometry, X-ray crystallography,
petroleum engineering, and geophysical exploration. Indirectly, it has also
led to new developments in its predecessor techniques in radiographic imag-
FIGURE 1.23
CT image of a patient showing the details in a cross-section through the
abdomen. Image courtesy of Foothills Hospital, Calgary.
ing. Details of the mathematical principles related to reconstruction from

projections and more illustrations of CT imaging are provided in Chapter 9.
1.7 Nuclear Medicine Imaging

The use of radioactivity in medical imaging began in the 1950s nuclear
medicine has now become an integral part of most medical imaging cen-
ters 3, 4, 5, 6, 86]. In nuclear medicine imaging, a small quantity of a radio-
pharmaceutical is administered into the body orally, by intravenous injection,
or by inhalation. The radiopharmaceutical is designed so as to be absorbed by
and localized in a speci c organ of interest. The gamma-ray photons emitted
from the body as a result of radioactive decay of the radiopharmaceutical are
used to form an image that represents the distribution of radioactivity in the
organ.
Nuclear medicine imaging is used to map physiological function such as
perfusion and ventilation of the lungs, and blood supply to the musculature of
the heart, liver, spleen, and thyroid gland. Nuclear medicine has also proven
to be useful in detecting brain and bone tumors. Whereas X-ray images
(a)
(b)
FIGURE 1.24
CT image of a patient scaled to (a) show the details of the lungs and (b) dis-
play the mediastinum in detail | the details of the lungs are not visible in
this rendition. Images courtesy of Alberta Children's Hospital, Calgary.
provide information related to density and may be used to detect altered
anatomy, nuclear medicine imaging helps in examining altered physiological
(or pathological) functioning of speci c organs in a body.
The most commonly used isotopes in nuclear medicine imaging are tech-
netium as 99m Tc which emits gamma-ray photons at 140 keV , and thallium
as 201 Tl at 70 keV or 167 keV . Iodine as 131 I is also used for thyroid imaging.
The rst imaging device used in nuclear medicine was the rectilinear scan-
ner, which consisted of a single-bore collimator connected to a gamma-ray
counter or detector. The scanner was coupled to a mechanical system that
performed a raster scan over the area of interest, making a map of the radi-
ation distribution in the area. The amount of radioactivity detected at each
position was either recorded on lm or on a storage oscilloscope. A major
diculty with this approach is that scanning is time consuming.
The scintillation gamma camera or the Anger camera uses a large thallium-
activated sodium iodide NaI (Tl)] detector, typically 40 cm in diameter and
10 mm in thickness. The gamma camera consists of three major parts: a
collimator, a detector, and a set of PMTs. Figure 1.25 illustrates the Anger
camera in a schematic sectional view. The following paragraphs describe some
of the important components of a nuclear medicine imaging system.
Image computer
Position Pulse
height Image
analysis
analysis
Photomultiplier
tubes (PMTs)
NaI(Tl) crystal detector

* *
Collimator
Gamma rays
emitted
Patient
FIGURE 1.25
Schematic (vertical sectional) representation of a nuclear medicine imaging
system with an Anger camera.
Collimator: A collimator consists of an array of holes separated by
lead septa. The function of the collimator is to allow passage of the
gamma rays that arrive along a certain path of propagation, and to
block (absorb) all gamma rays outside a narrow solid angle of accep-
tance. Collimators are usually made of lead alloys, but other materials
such as tantalum, tungsten, and gold have also been used. Di erent
geometries have been used in the design of collimator holes, including
triangular, square, hexagonal, and round patterns. Hexagonal holes
have been observed to be the most ecient, and are commonly used.
Two key factors in collimator design are geometric eciency, which is
the fraction of the gamma-ray photons from the source that are trans-
mitted through the collimator to the detector, and geometric (spatial)
resolution. In general, for a given type of collimator, the higher the
eciency, the poorer is the resolution. The resolution of a collimator is
increased if the size of the holes is reduced or if the collimator thickness
is increased however, these measures decrease the number of photons
that will reach the crystal, and hence reduce the sensitivity and e-
ciency of the system. The eciency of a typical collimator is about
0:01% that is, only 1 in every 10 000 photons emitted is passed by the
collimator to the crystal.
The most commonly used type of collimator is the parallel-hole collima-
tor, which usually serves for general purpose imaging, particularly for
large organs. Other designs include diverging, converging, fan-beam,
and pin-hole collimators.
Detector: At the back of the collimator is attached a detector, which is
usually a NaI (Tl) crystal of 6 ; 13 mm thickness. The crystal absorbs
the gamma-ray photons that pass through the collimator holes, and
reemits their energy as visible light (scintillation). The thickness of
the crystal determines the absorbed fraction of the gamma-ray photons
by the photoelectric e ect. A thick crystal has better absorption than
a thin crystal however, a thick crystal scatters and absorbs the light
before it reaches the back surface of the crystal. A crystal of thickness
10 mm absorbs about 92% of the photons received at 140 keV .
Photomultiplier tubes: The crystal is optically coupled at its back
surface to an array of PMTs. The PMTs are usually hexagonal in sec-
tion, and are arranged so as to cover the imaging area. Scintillations
within the crystal are converted by the photocathodes at the front of the
PMTs to photoelectrons, which are accelerated toward each of a series
of dynodes held at successively higher potentials until they reach the
anode at the back of the tube. During this process, the photoelectrons
produce a number of secondary electrons at each dynode, leading to a
current gain of the order of 106 .
Image computer: The current pulses produced by the PMTs in re-
sponse to scintillations in the crystal are applied to a resistor matrix
that computes the points of arrival of the corresponding gamma-ray
photons. The amplitudes of the pulses represent the energy deposited
by the gamma rays. A pulse-height analyzer is used to select pulses that
are within a preset energy window corresponding to the peak energy of
the gamma rays. The pulse-selection step reduces the e ect of scattered
rays at energy levels outside the energy window used.
Whereas the major advantage of nuclear medicine imaging lies in its ca-
pability of imaging the functional aspects of the human body rather than
the anatomy, its major disadvantages are poor spatial resolution and high
noise content. The intrinsic resolution of a typical gamma camera (crystal) is
3 ; 5 mm the net resolution including the e ect of the collimator, expressed as
the full width at half the maximum (FWHM) of the image of a thin line source
(the line spread function or LSF) is 7:5 ; 10 mm (see Figure 2.21). The main
causes of noise are quantum mottle due to the low number of photons used
to create images, and the random nature of gamma ray emission. Structured
noise may also be caused by nonuniformities in the gamma camera.
In general, the contrast of nuclear medicine images is high as the radiophar-
maceutical is designed so as to be selectively absorbed by and localized in the
organ of interest. Regardless, other organs and tissues in the body will also
absorb some amounts of the radiopharmaceutical, and emit gamma rays that
will appear as background counts and degrade contrast. Such e ects may be
labeled as physiological or anatomical artifacts. The contrast of an image is
also diminished by septal penetration in the collimator and scatter.
Multi-camera imaging and scanning systems: The data acquired by
a gamma camera represent a projection or planar image, which corresponds
to an integration of the 3D body being imaged along the paths of the gamma
rays. It should be noted that gamma rays are emitted by the body in all
directions during the imaging procedure. Modern imaging systems use two or
three gamma cameras (\heads") to acquire simultaneously multiple projection
images. Projection images acquired at diametrically opposite positions may
be averaged to reduce artifacts 87] see Section 10.5.5.
SPECT imaging: SPECT scanners usually gather 64 or 128 projections
spanning 180o or 360o around the patient, depending upon the application. In-
dividual scan lines from the projection images may then be processed through
a reconstruction algorithm to obtain 2D sectional images, which may fur-
ther be combined to create a 3D image of the patient. Coronal, sagital, and
oblique sections may then be created from the 3D dataset. Circular scanning
is commonly used to acquire projection images of the body at di erent angles.
Circular scanning provides projections at equal angular intervals, as required
by certain reconstruction algorithms. However, some clinical studies use el-
liptical scanning so as to keep the camera close to the body, in consideration
of the fact that the spatial resolution deteriorates rapidly with distance. In
such situations, the CT reconstruction algorithm should be adapted to the
nonuniformly spaced data.
Examples: Figure 1.26 illustrates a nuclear medicine projection (planar)
image of the chest of a patient. The region of high intensity (activity) on the
right-hand side of the image represents the heart. Observe the high level of
noise in the image. Figure 1.27 illustrates several series of SPECT images of
the left ventricle of the patient before and after stress (exercise on a treadmill).
The SPECT images display oblique sectional views of the myocardium in three
orientations: the short axis, the horizontal long axis, and the vertical long
axis of the left ventricle. Ischemic regions demonstrate lower intensity than
normal regions due to reduced blood supply. The generally noisy appearance
of SPECT images as well as the nature of the artifacts in nuclear medicine
images are illustrated by the images.
See Section 3.10 for details regarding gated blood-pool imaging see Sec-
tion 10.5 for several examples of SPECT images, and for discussions on the
restoration of SPECT images.
FIGURE 1.26
A planar or projection image of a patient used for myocardial SPECT imaging.
The two horizontal lines indicate the limits of the data used to reconstruct the
SPECT images shown in Figure 1.27. Image courtesy of Foothills Hospital,
Calgary.
Positron emission tomography (PET): Certain isotopes of carbon

(11 C ), nitrogen (13 N ), oxygen (15 O), and uorine (18 F ) emit positrons and are
suitable for nuclear medicine imaging. PET is based upon the simultaneous
detection of the two annihilation photons produced at 511 keV and emitted
in opposite directions when a positron loses its kinetic energy and combines
with an electron 4, 5, 6, 88]. The process is also known as coincidence detec-
tion. Collimation in PET imaging is electronic, which substantially increases
(a)
(b)
(c)
FIGURE 1.27
SPECT imaging of the left ventricle. (a) Short-axis images. (b) Horizontal
long axis images. (c) Vertical long axis images. In each case, the upper panel
shows four SPECT images after exercise (stress), and the lower panel shows
the corresponding views before exercise (rest). Images courtesy of Foothills
Hospital, Calgary.
the eciency of the detection process as compared to SPECT imaging. A
diaphragm is used only to de ne the plane of sectional (CT) imaging.
Several modes of data collection are possible for PET imaging, including
stationary, rotate-only, and rotate-translate gantries 88]. In one mode of data
collection, a ring of bismuth-germanate detectors is used to gather emission
statistics that correspond to a projection of a transversal section. A recon-
struction algorithm is then used to create 2D and 3D images.
The spatial resolution of PET imaging is typically 5 mm, which is almost
two times better than that of SPECT imaging. PET is useful in functional
imaging and physiological analysis of organs 89, 90].
1.8 Ultrasonography
Ultrasound in the frequency range of 1 ; 20 MHz is used in diagnostic ul-
trasonography 4, 5, 6]. The velocity of propagation of ultrasound through
a medium depends upon its compressibility: lower compressibility results in
higher velocity. Typical velocities in human tissues are 330 m=s in air (the
lungs) 1 540 m=s in soft tissue and 3 300 m=s in bone. A wave of ultra-
sound may get reected, refracted, scattered, or absorbed as it propagates
through a body. Most modes of diagnostic ultrasonography are based upon
the reection of ultrasound at tissue interfaces. A gel is used to minimize the
presence of air between the transducer and the skin in order to avoid reection
at the skin surface. Typically, pulses of ultrasound of about 1
s duration at
a repetition rate of about 1 000 pps (pulses per second) are applied, and the
resulting echoes are used for locating tissue interfaces and imaging.
Large, smooth surfaces in a body cause specular reection, whereas rough
surfaces and regions cause nonspecular reection or di use scatter. The nor-
mal liver, for example, is made up of clusters of parenchyma that are of the
order of 2 mm in size. Considering an ultrasound signal at 1 MHz and as-
suming a propagation velocity of 1 540 m=s, we get the wavelength to be
1:54 mm, which is of the order of the size of the parenchymal clusters. For
this reason, ultrasound is scattered in all directions by the liver, which appears
with a speckled texture in ultrasound scans 3, 6]. Fluid- lled regions such as
cysts have no internal structure, generate no echoes except at their bound-
aries, and appear as black regions on ultrasound images. The almost-complete
absorption of ultrasound by bone causes shadowing in images: tissues and or-
gans past bones and dense objects along the path of propagation of the beam
are not imaged in full and accurate detail. The quality of ultrasonographic
images is also a ected by multiple reections, speckle noise due to scattering,
and spatial distortion due to refraction. The spatial resolution in ultrasound
images is limited to the order of 0:5 ; 3 mm.
Some of the commonly used modes of ultrasonography are briey described
below.
A mode: A single transducer is used in this mode. The amplitude
(hence the name \A") of the echoes received is displayed on the vertical
axis, with the corresponding depth (related to the time of arrival of the
echo) being displayed on the horizontal axis. The A mode is useful in
distance measurement (ranging), with applications in the detection of
retinal detachment and the detection of shift of the midline of the brain.
M mode: This mode produces a display with time on the horizontal
axis and echo depth on the vertical axis. The M mode is useful in the
study of movement or motion (hence the name), with applications in
cardiac valve motion analysis.
B mode: An image of a 2D section or slice of the body is produced
in this mode by using a single transducer to scan the region of interest
or by using an array of sequentially activated transducers. Real-time
imaging is possible in the B mode with 15 ; 40 fps. The B mode is
useful in studying large organs, such as the liver, and in fetal imaging.
Doppler ultrasound: This mode is based upon the change in fre-
quency of the investigating beam caused by a moving target (the Doppler
e ect), and is useful in imaging blood ow. A particular application lies
in the detection of turbulence and retrograde ow, which is useful in
the diagnosis of stenosis or insuciency of cardiac valves and plaques in
blood vessels 91]. Doppler imaging may be used to obtain a combina-
tion of anatomic information with B-mode imaging and ow information
obtained using pulsed Doppler.
Special probes: A variety of probes have been developed for ultra-
sonography of speci c organs and for special applications, some of which
are endovaginal probes for fetal imaging, transrectal probes for imaging
the prostate 92], transesophageal probes for imaging the heart via the
esophagus, and intravascular probes for the study of blood vessels.
Examples: Echocardiography is a major application of ultrasonography
for the assessment of the functional integrity of heart valves. An array of ul-
trasound transducers is used in the B mode in echocardiography, so as to ob-
tain a video illustrating the opening and closing activities of the valve leaets.
Figure 1.28 illustrates two frames of an echocardiogram of a subject with a
normal mitral valve, which is captured in the open and closed positions in
the two frames. Figure 1.29 shows the M-mode ultrasound image of the same
subject, illustrating the valve leaet movement against time. Echocardiogra-
phy is useful in the detection of stenosis and loss of exibility of the cardiac
valves due to calci cation.
(a) (b)
FIGURE 1.28
Two frames of the echocardiogram of a subject with normal function of the
mitral valve. (a) Mitral valve in the fully open position. (b) Mitral valve in
the closed position. Images courtesy of Foothills Hospital, Calgary.
FIGURE 1.29
M-mode ultrasound image of a subject with normal function of the mitral
valve. The horizontal axis represents time. The echo signature of the mitral
valve leaets as they open and close is illustrated. Image courtesy of Foothills
Hospital, Calgary.
In spite of limitations in image quality and resolution, ultrasonography
is an important medical imaging modality due to the nonionizing nature of
the medium. For this reason, ultrasonography is particularly useful in fetal
imaging. Figure 1.30 shows a B-mode ultrasound image of a fetus. The
outline of the head and face as well as the spinal column are clearly visible
in the image. Images of this nature may be used to measure the size of the
head and head-to-sacrum length of the fetus. Ultrasonography is useful in the
detection of abnormalities in fetal development, especially distension of the
stomach, hydrocephalus, and complications due to maternal (or gestational)
diabetes such as the lack of development of the sacrum.
FIGURE 1.30
B-mode ultrasound (3:5 MHz ) image of a fetus (sagital view). Image courtesy
of Foothills Hospital, Calgary.
Ultrasonography is also useful in tomographic imaging 83, 93], discrimi-

nating between solid masses and uid- lled cysts in the breast 53], and tissue
characterization 5].
1.9 Magnetic Resonance Imaging

MRI is based on the principle of nuclear magnetic resonance (NMR): the
behavior of nuclei under the inuence of externally applied magnetic and EM
(radio-frequency or RF) elds 4, 6, 84, 94, 95, 96]. A nucleus with an odd
number of protons or an odd number of neutrons has an inherent nuclear spin
and exhibits a magnetic moment such a nucleus is said to be NMR-active.
The commonly used modes of MRI rely on hydrogen (1 H or proton), carbon
(13 C ), uorine (19 F ), and phosphorus (31 P ) nuclei.
In the absence of an external magnetic eld, the vectors of magnetic mo-
ments of active nuclei have random orientations, resulting in no net mag-
netism. When a strong external magnetic eld Ho is applied, some of the
nuclear spins of active nuclei align with the eld (either parallel or antiparal-
lel to the eld). The number of spins that align with the eld is a function
of Ho and inversely related to the absolute temperature. At the commonly
used eld strength of 1:5 T (Tesla), only a relatively small fraction of spins
align with the eld. The axis of the magnetic eld is referred to as the z
axis. Parallel alignment corresponds to a lower energy state than antiparallel
alignment, and hence there will be more nuclei in the former state. This state
of forced alignment results in a net magnetization vector. (At equilibrium
and 1:5 T , only about seven more spins in a million are aligned in the parallel
state hence, MRI is a low-sensitivity imaging technique.)
The magnetic spin vector of each active nucleus precesses about the z axis
at a frequency known as the Larmor frequency, given by
!o = Ho (1.4)
where is the gyromagnetic ratio of the nucleus considered (for protons,
= 42:57 MHz T ;1 ). From the view-point of classical mechanics and for
1 H , this form of precession is comparable to the rotation of a spinning top's
axis around the vertical.
MRI involves controlled perturbation of the precession of nuclear spins, and
measurement of the RF signals emitted when the perturbation is stopped and
the nuclei return to their previous state of equilibrium. MRI is an intrinsi-
cally 3D imaging procedure. The traditional CT scanner requires mechanical
scanning and provides 2D cross-sectional images in a slice-by-slice manner:
slices at other orientations, if required, have to be computed from a set of
2D slices covering the required volume. In MRI, however, images may be
obtained directly in any transversal, coronal, sagital, or oblique section by us-
ing appropriate gradients and pulse sequences. Furthermore, no mechanical
scanning is involved: slice selection and scanning are performed electronically
by the use of magnetic eld gradients and RF pulses.
The main components and principles of MRI are as follows 84]:
A magnet that provides a strong, uniform eld of the order of 0:5 ;
4 T . This causes some active nuclei to align in the direction of the eld
(parallel or antiparallel) and precess about the axis of the eld. The
rate of precession is proportional to the strength of the magnetic eld
Ho. The stronger the magnetic eld, the more spins are aligned in the
parallel state versus the antiparallel state, and the higher will be the
signal-to-noise ratio (SNR) of the data acquired.
An RF transmitter to deliver an RF electromagnetic pulse H1 to the
body being imaged. The RF pulse provides the perturbation mentioned
earlier: it causes the axis of precession of the net spin vector to deviate
or \ip" from the z axis. In order for this to happen, the frequency of
the RF eld must be the same as that of precession of the active nuclei,
such that the nuclei can absorb energy from the RF eld (hence the term
\resonance" in MRI). The frequency of RF-induced rotation is given by
!1 = H1 : (1.5)
When the RF perturbation is removed, the active nuclei return to their
unperturbed states (alignment with Ho ) through various relaxation pro-
cesses, emitting energy in the form of RF signals.
A gradient system to apply to the body a controlled space-variant
and time-variant magnetic eld
h(t x) = G(t) x (1.6)
where x is a vector representing the spatial coordinates, G is the gradi-
ent applied, and t is time. The components of G along the z direction
as well as in the x and y directions (the plane x ; y is orthogonal to
the z axis) are controlled individually however, the component of mag-
netic eld change is only in the z direction. The gradient causes nuclei
at di erent positions to precess at di erent frequencies, and provides
for spatial coding of the signal emitted from the body. The Larmor
frequency at x is given by
!(x) = (Ho + G x): (1.7)
Nuclei at speci c positions or planes in the body may be excited selec-
tively by applying RF pulses of speci c frequencies. The combination of
the gradient elds and the RF pulses applied is called the pulse sequence.
An RF detector system to detect the RF signals emitted from the
body. The RF signal measured outside the body represents the sum of
the RF signals emitted by active nuclei from a certain part or slice of
the body, as determined by the pulse sequence. The spectral spread of
the RF signal due to the application of gradients provides information
on the location of the corresponding source nuclei.
A computing and imaging system to reconstruct images from the
measured data, as well as process and display the images. Depend-
ing upon the pulse sequence applied, the RF signal sensed may be
formulated as the 2D or 3D Fourier transform of the image to be re-
constructed 4, 84, 94, 95]. The data measured correspond to samples
of the 2D Fourier transform of a sectional image at points on concentric
squares or circles 4]. The Fourier or backprojection methods of image
reconstruction from projections (described in Chapter 9) may then be
used to obtain the image. (The Fourier method is the most commonly
used method for reconstruction of MR images.)
Various pulse sequences may be used to measure di erent parameters of

the tissues in the body being imaged. The image obtained is a function
of the nuclear spin density in space and the corresponding parameters of
the relaxation processes involved. Longitudinal magnetization refers to the
component of the magnetization vector along the direction of the external
magnetic eld. The process by which longitudinal magnetization returns to
its state of equilibrium (that is, realignment with the external magnetic eld)
after an excitation pulse is referred to as longitudinal relaxation. The time
constant of longitudinal relaxation is labeled as T1 .
A 90o RF pulse causes the net magnetization vector to be oriented in the
plane perpendicular to the external magnetic eld: this is known as transverse
magnetization. When the excitation is removed, the a ected nuclei return to
their states of equilibrium, emitting a signal, known as the free-induction
decay (FID) signal, at the Larmor frequency. The decay time constant of
transverse magnetization is labeled as T2 . Values of T1 for various types
of tissues range from 200 ms to 2 000 ms T2 values range from 80 ms to
180 ms. Several other parameters may be measured by using di erent MRI
pulse sequences: the resulting images may have di erent appearances and
clinical applications.
MRI is suitable for functional imaging. The increased supply of oxygen
(or oxygenated blood) to certain regions of the brain due to related stimuli
may be recorded on MR images. The di erence between the prestimulus and
post-stimulus images may then be used to analyze the functional aspects of
speci c regions of the brain.
Examples: Figure 1.31 shows a sagital MR image of a patient's knee,
illustrating the bones and cartilages that form the knee joint. Images of this
type assist in the detection of bruised bones, bleeding inside the distal end of
the femur, torn cartilages, and ruptured ligaments.
Figures 1.32 (a){(c) illustrate the sagital, coronal, and transversal (cross-
sectional) views of the MR image of a patient's head. The images show the
details of the structure of the brain. MRI is useful in functional imaging of
the brain.
FIGURE 1.31
Sagital section of the MR image of a patient's knee. Image courtesy of
Foothills Hospital, Calgary.
(a) (b) (c)
FIGURE 1.32
(a) Sagital, (b) coronal, and (c) transversal (cross-sectional) MR images of a patient's head. Images courtesy of Foothills
Hospital, Calgary.
51
52
Physiological
Image data acquisition
system (patient)
Biomedical images
Imaging Analog-to- Picture archival
Transducers
system digital and communication
Probing signal conversion system (PACS)
or radiation
Computer-aided
diagnosis
and therapy
Pattern recognition, Analysis of regions Detection Filtering

classification, and or objects; of regions and image
diagnostic decision feature extraction or objects enhancement
Biomedical Image Analysis

Physician
or medical
Image or pattern analysis Image processing
specialist
FIGURE 1.33
Computer-aided diagnosis and therapy based upon biomedical image analysis.
1.10 Objectives of Biomedical Image Analysis

The representation of biomedical images in electronic form facilitates com-
puter processing and analysis of the data. Figure 1.33 illustrates the typical
steps and processes involved in computer-aided diagnosis (CAD) and therapy
based upon biomedical images analysis.
The human{instrument system: The components of a human{instru-
ment system 97, 98, 99, 100, 101, 102] and some related notions are described
in the following paragraphs.
The subject (or patient): It is important always to bear in mind that
the main purpose of biomedical imaging and image analysis is to provide
a certain bene t to the subject or patient. All systems and procedures
should be designed so as not to cause undue inconvenience to the subject,
and not to cause any harm or danger. In applying invasive or risky
procedures, it is extremely important to perform a risk{bene t analysis
and determine if the anticipated bene ts of the procedure are worth
placing the subject at the risks involved.
Transducers: lms, scintillation detectors, uorescent screens, solid-
state detectors, piezoelectric crystals, X-ray generators, ultrasound gen-
erators, EM coils, electrodes, sensors.
Signal-conditioning equipment: PMTs, ampli ers, lters.
Display equipment: oscilloscopes, strip-chart or paper recorders, com-
puter monitors, printers.
Recording, data processing, and transmission equipment: lms, analog-
to-digital converters (ADCs), digital-to-analog converters (DACs), digi-
tal tapes, compact disks (CDs), diskettes, computers, telemetry systems,
picture archival and communication systems (PACS).
Control devices: power supply stabilizers and isolation equipment, pa-
tient intervention systems.
The major objectives of biomedical instrumentation 97, 98, 99, 100, 101,
102] in the context of imaging and image analysis are:
Information gathering | measurement of phenomena to interpret an
organ, a process, or a system.
Screening | investigating a large asymptomatic population for the in-
cidence of a certain disease (with the aim of early detection).
Diagnosis | detection or con rmation of malfunction, pathology, or
abnormality.
Monitoring | obtaining periodic information about a system.
Therapy and control | modi cation of the behavior of a system based

upon the outcome of the activities listed above to ensure a speci c result.
Evaluation | objective analysis to determine the ability to meet func-

tional requirements, obtain proof of performance, perform quality con-
trol, or quantify the e ect of treatment.
Invasive versus noninvasive procedures: Image acquisition procedures

may be categorized as invasive or noninvasive procedures. Invasive proce-
dures involve the placement of devices or materials inside the body, such as
the insertion of endoscopes, catheter-tip sensors, and X-ray contrast media.
Noninvasive procedures are desirable in order to minimize risk to the subject.
Note that making measurements or imaging with X rays, ultrasound, etc.
could strictly be classi ed as invasive procedures because they involve pene-
tration of the body with externally administered radiation, even though the
radiation is invisible and there is no visible puncturing or invasion of the body.
Active versus passive procedures: Image acquisition procedures may
be categorized as active or passive procedures. Active data acquisition pro-
cedures require external stimuli to be applied to the subject, or require the
subject to perform a certain activity to stimulate the system of interest in
order to elicit the desired response. For example, in SPECT investigations of
myocardial ischemia, the patient performs vigorous exercise on a treadmill.
An ischemic zone is better delineated in SPECT images taken when the car-
diac system is under stress than when at rest. While such a procedure may
appear to be innocuous, it does carry risks in certain situations for some sub-
jects: stressing an unwell system beyond a certain limit may cause pain in
the extreme situation, the procedure may cause irreparable damage or death.
The investigator should be aware of such risks, factor them in a risk{bene t
analysis, and be prepared to manage adverse reactions.
Passive procedures do not require the subject to perform any activity. Ac-
quiring an image of a subject using reected natural light (with no ash from
the camera) or thermal emission could be categorized as a passive and non-
contact procedure.
Most organizations require ethical approval by specialized committees for
experimental procedures involving human or animal subjects, with the aim of
minimizing the risk and discomfort to the subject and maximizing the bene ts
to both the subject and the investigator.
1.11 Computer-aided Diagnosis

Radiologists, physicians, cardiologists, neuroscientists, pathologists, and other
health-care professionals are highly trained and skilled practitioners. Why
then would we want to suggest the use of computers for the analysis of biomed-
ical images? The following paragraphs provide some arguments in favor of the
application of computers to process and analyze biomedical images.
Humans are highly skilled and fast in the analysis of visual patterns, but
are slow (usually) in arithmetic operations with large numbers of values.
A single 64 64-pixel SPECT image contains a total of 4 096 pixels a
high-resolution mammogram may contain as many as 5 000 4 000 =
20 106 pixels. If images need to be processed to remove noise or extract
a parameter, it would not be practical for a person to perform such
computation. Computers can perform millions of arithmetic operations
per second. It should be noted, however, that the recognition of objects
and patterns in images using mathematical procedures typically requires
huge numbers of operations that could lead to slow responses in such
tasks from low-level computers. A trained human observer, on the other
hand, can usually recognize an object or a pattern in an instant.
Humans could be a ected by fatigue, boredom, and environmental fac-
tors, and are susceptible to committing errors. Working with large
numbers of images in one sitting, such as in breast cancer screening,
poses practical diculties. A human observer could be distracted by
other events in the surrounding areas and may miss uncommon signs
present in some images. Computers, being inanimate but mathemat-
ically accurate and consistent machines, can be designed to perform
computationally speci c and repetitive tasks.
Analysis by humans is usually subjective and qualitative. When com-
parative analysis is required between an image of a subject and another
or a reference pattern, a human observer would typically provide a qual-
itative response. Speci c or objective comparison | for example, the
comparison of the volume of two regions to the accuracy of the order
of even a milliliter | would require the use of a computer. The deriva-
tion of quantitative or numerical features from images would certainly
demand the use of computers.
Analysis by humans is subject to interobserver as well as intraobserver
variations (with time). Given that most analyses performed by hu-
mans are based upon qualitative judgment, they are liable to vary with
time for a given observer, or from one observer to another. The former
could be due to lack of diligence or due to inconsistent application of
knowledge, and the latter due to variations in training and the level of
understanding or competence. Computers can apply a given procedure
repeatedly and whenever recalled in a consistent manner. Furthermore,
it is possible to encode the knowledge (to be more speci c, the logical
processes) of many experts into a single computational procedure, and
thereby enable a computer with the collective \intelligence" of several
human experts in an area of interest.
One of the important points to note in the discussion above is that quantita-
tive analysis becomes possible by the application of computers to biomedical
images. The logic of medical or clinical diagnosis via image analysis could
then be objectively encoded and consistently applied in routine or repetitive
tasks. However, it should be emphasized at this stage that the end-goal of
biomedical image analysis should be computer-aided diagnosis and not auto-
mated diagnosis. A physician or medical specialist typically uses a signi cant
amount of information in addition to images, including the general physical
appearance and mental state of the patient, family history, and socio-economic
factors a ecting the patient, many of which are not amenable to quanti ca-
tion and logical rule-based processes. Biomedical images are, at best, indirect
indicators of the state of the patient many cases may lack a direct or unique
image-to-pathology relationship. The results of image analysis need to be
integrated with other clinical signs, symptoms, and information by a physi-
cian or medical specialist. Above all, the intuition of the medical specialist
plays an important role in arriving at the nal diagnosis. For these reasons,
and keeping in mind the realms of practice of various licensed and regulated
professions, liability, and legal factors, the nal diagnostic decision is best left
to the physician or medical specialist. It could be expected that quantitative
and objective analysis facilitated by the application of computers to biomed-
ical image analysis will lead to a more accurate diagnostic decision by the
physician.
On the importance of quantitative analysis:
\When you can measure what you are speaking about, and ex-
press it in numbers, you know something about it but when you
cannot measure it, when you cannot express it in numbers, your
knowledge is of a meagre and unsatisfactory kind: it may be the
beginning of knowledge, but you have scarcely, in your thoughts,
advanced to the stage of science."
{ Lord Kelvin (William Thomson, 1824 { 1907) 103]
On assumptions made in quantitative analysis:
\Things do not in general run around with their measure stamped
on them like the capacity of a freight car it requires a certain
amount of investigation to discover what their measures are ...
What most experimenters take for granted before they begin their
experiments is in nitely more interesting than any results to which
their experiments lead."
{ Norbert Wiener (1894 { 1964)
1.12 Remarks
We have taken a general look at the nature of biomedical images in this chap-
ter, and seen a few images illustrated for the purpose of gaining familiarity
with their appearance and typical features. Speci c details of the characteris-
tics of several types of biomedical images and their processing or analysis are
described in subsequent chapters, along with more examples.
We have also stated the objectives of biomedical imaging and image ana-
lysis. Some practical diculties that arise in biomedical investigations based
upon imaging were discussed in order to draw attention to the relevant prac-
tical issues. The suitability and desirability of the application of computers
for biomedical image analysis were discussed, with emphasis on objective and
quantitative analysis toward the end-goal of CAD. The following chapters
provide the descriptions of several image processing and analysis techniques
for various biomedical applications.
1.13 Study Questions and Problems

1. Give three reasons for the application of computers in medicine for computer-
aided diagnosis.
2. List the relative advantages and disadvantages of X-ray, ultrasound, CT, MR,
and nuclear medicine imaging for two clinical applications. Indicate where
each method is inappropriate or inapplicable.
3. Discuss the factors aecting the choice of X-ray, ultrasound, CT, MR, and
nuclear medicine imaging procedures for clinical applications.
4. Describe a few sources of artifact in X-ray, ultrasound, CT, MR, and nuclear
medicine imaging.
5. Discuss the sources and the nature of random, periodic, structured, and phys-
iological artifacts in medical images.
6. Describe the dierence between anatomical (physical) and functional (physi-
ological) imaging. Give examples.
7. Distinguish between active and passive medical imaging procedures give ex-
amples.
8. Distinguish between invasive and noninvasive medical imaging procedures
give examples.
9. Discuss factors that aect the resolution in various medical imaging modali-
ties, including X-ray, ultrasound, CT, MR, and nuclear medicine imaging.
1.14 Laboratory Exercises and Projects

1. Visit a few medical imaging facilities in your local hospital or health sciences
center. View the procedures related to the acquisition of a few medical images,
including X-ray, ultrasound, MR, CT, and SPECT images.
Respect the priority, privacy, and condentiality of patients.
Discuss the imaging protocols and parameters with a medical physicist and
the technologists. Develop an understanding of the relationship between the
imaging system parameters, image quality, and radiation exposure to the pa-
tient.
Request a radiologist to explain how he or she interprets the images. Ob-
tain information on the dierences between normal and abnormal (disease)
patterns in each mode of imaging.
Collect a few sample images for use in image processing experiments, after
obtaining the necessary permissions and ensuring that you carry no patient
identication out of the facility.
2. Most medical imaging facilities use phantoms or test objects for quality control
of imaging systems. If a phantom is not available, prepare one by attaching
strips of dierent thickness and dierent metals to a plastic or plexiglass sheet.
With the help of a qualied technologist, obtain X-ray images of a phantom at
widely dierent kV p and mAs settings. Study the contrast, noise, and detail
visibility in the resulting images.
Digitize the images for use in image processing experiments.
3. Visit a medical (clinical or pathology) laboratory. View several samples and
specimens through microscopes.
Respect the priority, privacy, and condentiality of patients.
Request a technologist or pathologist to explain how he or she interprets the
images. Obtain information on the dierences between normal and abnormal
(disease) patterns in dierent types of samples and tests.
Collect a few sample images for use in image processing experiments, after
obtaining the necessary permissions and ensuring that you carry no patient
identication out of the laboratory.
4. Interview physicians, radiologists, pathologists, medical physicists, and medi-
cal technologists to nd areas where they need and would like to use comput-
ing technology, digital image processing, computer vision, pattern recognition,
and pattern classication methods to help in their work. Volunteer to assist
them in their work! Develop projects for your course of study or research
in biomedical image analysis. Request a specialist in the relevant area to
collaborate with you in the project.
5. Prepare a set of test images by collecting at least ten images that contain the
following features:
a collection of small objects,
a collection of large objects,
directional (oriented) features,
ne texture,
coarse texture,
geometrical shapes,
human faces,
smooth features,
sharp edges.
Scan a few photos from your family photo album. Limit synthesized images
to one or two in the collection. Limit images borrowed from Web sites to two
in the collection. Use the images in the exercises provided at the end of each
chapter.
For a selection of test images from those that have been used in this book,
visit www.enel.ucalgary.ca/People/Ranga/enel697
2
Image Quality and Information Content
Several factors a ect the quality and information content of biomedical images
acquired with the modalities described in Chapter 1. A few considerations
in biomedical image acquisition and analysis that could have a bearing on
image quality are described in Section 2.1. A good understanding of such
factors, as well as appropriate characterization of the concomitant loss in
image quality, are essential in order to design image processing techniques to
remove the degradation and/or improve the quality of biomedical images. The
characterization of information content is important for the same purposes as
above, as well as in the analysis and design of image transmission and archival
systems.
An inherent problem in characterizing quality lies in the fact that image
quality is typically judged by human observers in a subjective manner. To
quantify the notion of image quality is a dicult proposition. Similarly, the
nature of the information conveyed by an image is dicult to quantify due
to its multifaceted characteristics in terms of statistical, structural, percep-
tual, semantic, and diagnostic connotations. However, several measures have
been designed to characterize or quantify a few specic attributes of images,
which may in turn be associated with various notions of quality as well as
information content. The numerical values of such measures of a given image
before and after certain processes, or the changes in the attributes due to cer-
tain phenomena, could then be used to assess variations in image quality and
information content. We shall explore several such measures in this chapter.
2.1 Diculties in Image Acquisition and Analysis

In Chapter 1, we studied several imaging systems and procedures for the
acquisition of many di erent types of biomedical images. The practical appli-
cation of these techniques may pose certain diculties: the investigator often
faces conditions that may impose limitations on the quality and information
content of the images acquired. The following paragraphs illustrate a few
practical diculties that one might encounter in biomedical image acquisition
and analysis.
61

Biomedical 3 100 PDF

Uploaded by

Copyright:

Available Formats

Biomedical 3 100 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Biomedical 3 100 PDF

Uploaded by

Copyright:

Available Formats

Biomedical Engineering

Library of Congress Cataloging-in-Publication Data

Catalog record is available from the Library of Congress

Visit the CRC Press Web site at www.crcpress.com

© 2005 by CRC Press LLC

No claim to original U.S. Government works

Background and Motivation

The Problem-solving Approach

The Intended Audience

Teaching and Learning Plans

Rangaraj (Raj) Mandayam Rangayyan was born in Mysore, Karnataka, In-

Photo by Trudie Lee.

Writing this book on the multifaceted subject of biomedical image analysis

the mean (average) of a random variable

X-ray attenuation coecient

The human body is composed of many systems, such as the cardiovascular

1.1 Body Temperature as an Image

1.3 Light Microscopy

Astigmatism: Due to nonuniformities in lenses, a point may appear

Chromatic aberration: Electromagnetic (EM) waves of di erent wave-

Spherical aberration: The rays of light arriving at the periphery

Geometric distortion: Poorly crafted lenses may cause geometric

1.4 Electron Microscopy

Example: Figure 1.8 illustrates SEM images of collagen bers in rabbit

1.5 X-ray Imaging

we see that the radiographic image so produced is a 2D planar image of the

A uoroscopy system uses an image intensi er and a video camera in place

1.5.1 Breast cancer and mammography

X-ray source (target)

Path of film movement

Computed tomography: The technique of CT imaging was developed

ing. Details of the mathematical principles related to reconstruction from

1.7 Nuclear Medicine Imaging

NaI(Tl) crystal detector

Positron emission tomography (PET): Certain isotopes of carbon

Ultrasonography is also useful in tomographic imaging 83, 93], discrimi-

1.9 Magnetic Resonance Imaging

Various pulse sequences may be used to measure di erent parameters of

Pattern recognition, Analysis of regions Detection Filtering

Biomedical Image Analysis

1.10 Objectives of Biomedical Image Analysis

Therapy and control | modi cation of the behavior of a system based

Evaluation | objective analysis to determine the ability to meet func-

Invasive versus noninvasive procedures: Image acquisition procedures

1.11 Computer-aided Diagnosis

1.13 Study Questions and Problems

1.14 Laboratory Exercises and Projects

2.1 Diculties in Image Acquisition and Analysis

You might also like

X-ray attenuation coecient

A uoroscopy system uses an image intensi er and a video camera in place

Ultrasonography is also useful in tomographic imaging 83, 93], discrimi-