Artificial Neural Networks in Cancer Diagnosis, Prognosis, and Patient Management (Biomedical Engineering)

212 Pages·2001·3.42 MB·English

by  R. N. G. Naguib

Artificial Neural Networks in
Cancer Diagnosis, Prognosis,
and Patient Management
Biomedical Engineering
Series
Edited by Michael R. Neuman
Published Titles
Electromagnetic Analysis and Design in Magnetic Resonancer
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
Forthcoming Titles
Noninvasive Instrumentation and Measurement in Medical
Diagnosis, Robert B. Northrop
Handbook of Neuroprosthetic Methods, Warren E. Finn
and Peter G. LoPresti
The BIOMEDICAL ENGINEERING Series
Series Editor Michael Neuman
Artificial Neural Networks in
Cancer Diagnosis, Prognosis,
and Patient Management
Edited by
Raouf N.G. Naguib
Gajanan V. Sherbet
CRC Press
Boca Raton London New York Washington, D.C.
disclaimer Page 1 Thursday, May 3, 2001 2:40 AM
Library of Congress Cataloging-in-Publication Data
Artificial neural networks in cancer diagnosis, prognosis, and patient management /
[edited by] Raouf N.G. Naguib and Gajanan V. Sherbet.
p. ; cm.— (Biomedical engineering series)
Includes bibliographical references and index.
ISBN 0-8493-9692-1 (alk. paper)
1. Cancer—Computer simulation. 2. Neural networks (Computer science) 3. Artificial
intelligence—Medical applications. I. Naguib, Raouf N.G. II. Sherbet, G.V. (Gajanan
V.) III. Biomedical engineering series (Boca Raton, Fla.)
[DNLM: 1. Neoplasms—diagnosis. 2. Diagnosis, Computer-Assisted. 3.
Neoplasms—therapy. 4. Neural Networks (Computer). 5. Prognosis. QZ 241 A791 2001]
RC254.5 .A66 2001
616.99′4′00113—dc21 2001025509

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.
All rights reserved. Authorization to photocopy items for internal or personal use, or the personal or internal use of specific
clients, may be granted by CRC Press LLC, provided that $1.50 per page photocopied is paid directly to Copyright clearance
Center, 222 Rosewood Drive, Danvers, MA 01923 USA. The fee code for users of the Transactional Reporting Service is
ISBN 0-8493-9692-1/01/$0.00+$1.50. The fee is subject to change without notice. For organizations that have been granted
a photocopy license by the CCC, a separate system of payment has been arranged.
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
© 2001 by CRC Press LLC
No claim to original U.S. Government works
International Standard Book Number 0-8493-9692-1
Library of Congress Card Number 2001025509
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
“Weigh and closely scan both favourable and unfavourable features and the output will
offer an option for action”
Thiru Valluvar
(Tamil Poet, Second Century, India)
Thirukkural, Chapter 51, Verse 504

PREFACE
The potential value of artificial neural networks (ANNs) as a predictor of malignancy has
now been widely recognised. The concept of ANNs dates back to the early part of the 20th
century; however, their latest resurrection started in earnest in the 1980s when they were
applied to many problems in the areas of pattern recognition, control, and optimisation. Here
we present a series of articles that emphasise the keen interest displayed by the scientific
community in the application of neural networks in the management of human cancers, and
also reflect the recent intense activity in this field. The neural systems use prognostic cancer
markers as input neurons and provide an ideal means of combining the input signals so that
the output neurons can provide a predictive basis on which to determine the course of patient
management. In essence, the two characteristics that determine the reliability of neural
networks are the discriminant analysis of the input variables and the correct classification of
tumours with regard to the accuracy of the predictive output, whether in the form of diagnosis,
its spread to regional lymph nodes, or patient survival. This versatility of the artificial neural
networks is aptly encapsulated by the Tamil quotation.
One of the main objectives of this book is to ensure that the material reported in it
emanates from a variety of institutions, in order that an objective and unbiased view of the
different approaches undertaken by researchers in those institutions can be presented. Thus,
the book contains chapters relating to different types of cancer. These have been written by
leading international researchers in the field of artificial neural networks, as well as by leading
experts in oncology, physicians, pathologists, and surgeons from Europe and North and South
America.
The subject of artificial neural networks has been extensively treated in the literature and
therefore the object of this book is not to present yet another monograph in this area. The aim,
however, is to focus solely on their applicability in the field of oncology which, as witnessed
by the growing number of publications, is rapidly developing into a major area of research in
its own right. Chapter 1, therefore, presents a very brief introduction to ANNs whilst
attempting to emphasise their direct relationship to the inherent problems in cancer diagnosis,
prognosis, and patient management.
Breast cancer has been the focus of much research lately, especially with regard to the
identification of molecular cancer markers. In Chapter 2, Angus et al. discuss some issues
relating to prognosis in breast cancer patients. They highlight the clinical importance of
molecular, cytological, and histological prognostic markers, including the proteins associated
with metastatic potential. They also present an ANN analysis of the significance of the
expression of the metastasis-associated h-mts1 and nm23 genes and their individual
relationship to nodal spread of breast cancer. This is then followed in Chapter 3 by a
discussion of the use of image cytometric measurements of DNA ploidy, the S-phase cell
fraction and nuclear pleomorphism as prognostic aids. The practicalities and putative benefits
of analysing these data using neural systems are then scrutinised.
The application of ANNs in lung cancer is the subject of Chapters 4 and 5. Chapter 4 is a
contribution by Esteva and colleagues. The chapter presents the general issues involved in the
prognostic analysis of patients with carcinoma of the lung, and assesses the accuracy of
applying artificial neural networks to the prognosis of post-surgical outcome of lung cancer
patients. Chapter 5 by Jefferson et al., on the other hand, focuses on the use of a genetic
algorithm neural network for prognosis in surgically treated nonsmall cell lung cancer.
In Chapter 6, Speight and Hammond deal with the use of machine learning in screening for
oral cancer. The chapter summarises the investigations into the application of ANNs in the
selection of high-risk groups of patients, comparing their predictive performance with other
machine learning techniques and evaluating the potential performance of machine learning for
the detection of high-risk individuals. This chapter also discusses the merits of utilising
machine learning for the prediction of risk of oral, and other cancers, as an adjunct to
population screening.
The contribution by Wayman and Griffin in Chapter 7 describes the application of ANNs
in the prediction of outcome for cancer of the oesophago-gastric junction. As the decision on
whether a surgical procedure is necessary or not is made on the interpretation of pre-operative
assessments of tumour stage and patient fitness, the correlation between pre- and post-
operative findings is frequently poor. Thus, results relating to the application of ANNs to
patients undergoing potentially curative resection of adenocarcinoma of the oesophago-gastric
junction, both pre- and post-operatively, are reported in this chapter.
It is inevitable, as with any novel and fast growing area of research within a relatively
tightly knit and closely collaborating community of researchers, that a degree of overlap will
occur. This is particularly evident in the chapters on urological oncology, an area which has
perhaps attracted a more concerted effort on the part of clinicians, engineers, and biomedical
computing scientists. This is reflected in the number of chapters in this book that deal with
this specific area of cancer from both a diagnosis and prognosis viewpoint.
Chapters 8 to 12 therefore represent a comprehensive treatment of this area of machine
learning in urological oncology. Chapters 8 (Douglas and Moul) and 9 (Niederberger and
Ridout) review the basic concepts of artificial neural networks and summarise their
application in renal cell carcinoma (RCC), prostate, bladder, and testicular cancers. These
reviews provide an insight into the difficult task of preoperative diagnosis of RCC, issues in
prostate cancer diagnosis, outcome prediction, and patients' quality of life. These parameters
are increasingly being regarded as important elements in the choice of treatment. The
combined application of image analysis procedures and ANNs to identify bladder cells
expressing the tumour antigen p300, as well as automated cytology-based interpretation in
bladder cancer, are emphasised. Testicular cancer is reviewed, especially from the angle of
pathologic stage I vs. stage II prediction in patients with clinical stage I nonseminomatous
testicular cancer.
In Chapter 10, Stamey et al. give a detailed description of ProstAsure™, a neural-based
technique to predict the risk of prostate cancer in men with a PSA ≤ 4.0 ng/ml, and predict
tumour recurrence following radical prostatectomy. Hamdy reports, in Chapter 11, on the use
of a neural network to predict prognosis and outcome in prostate cancer. Conventional input
variables (age, stage, bone scan findings, grade, PSA, and type of treatment) are used in
addition to data derived from the immunohistochemical staining performed on tissue
specimens for the proto-oncogene bcl-2 and tumour suppressor gene p53. The presence of
increased abnormal expression of these genes has been associated with disease progression in
prostate cancer. Hamdy shows that ANNs can be used to assess the benefits, or otherwise, of
such newer experimental tumour markers.
Finally, in the field of urological oncology, Chapter 12’s authors Qureshi and Mellon
present a study relating to the prediction of clinical outcome for patients with bladder cancer
utilising prognostic markers identified at initial presentation. In this study, they seek to assess
the ability of an ANN to predict the recurrence and stage progression in bladder cancer. This
is carried out in a group of patients with newly diagnosed Ta/T1 bladder cancer, and 12-month
cancer-specific survival in a group of patients with primary T2-T4 bladder cancer by using
clinical, pathological and molecular prognostic indicators. In addition, the networks'
predictions are compared with those of four consultant urologists supplied with the same data.
The final chapter is a contribution on skin cancer by Hintz-Madsen and his colleagues,
where the authors describe a comprehensive study of the applications of ANNs in the
diagnosis of melanoma. The accurate detection of this malignancy is heavily dependent upon
the precise analysis of skin pigmentation, tumour shape, and colour.
There are several areas of neoplasia that have not been investigated by using the ANN
tool. The editors hope this volume will not only emphasise the value of this tool in the study
of this disease process but will provide a nucleus around which future work might be planned
and executed.
Apart from minor cosmetic changes we have restrained from attempting to change either
the structure or contents of each chapter. We sincerely felt that the views and technical
contents presented by the authors should give an honest insight into the modes and practices
of their efforts in this particularly sensitive, and perhaps contentious, area of research.
The experts who have contributed to this work have certainly made this book move from
the possible to the achievable. We wish to thank them most sincerely. Without the time and
effort they spent on their respective chapters, the objectives that we set for ourselves would
not have been accomplished. Our research students have, over the years, been a tremendous
source of inspiration. They have been instrumental in developing ideas and algorithms which,
in many instances, seemed initially to be impractical. We are indebted to them for their efforts,
original ideas, and many stimulating discussions. Our thanks also go to Sonia Clarke whose
editing skills and careful attention to detail have made our task much easier than originally
expected. We wish to thank Dr. M.S. Lakshmi for providing the Tamil quotation and its
translation. Finally, we also wish to acknowledge the Cancer Research Campaign of the
United Kingdom for supporting many aspects of this work.
R.N.G. Naguib, Ph.D.
G.V. Sherbet, DSc., FRCPath.
Editors