Table Of ContentLecture Notes in Medical Informatics 44
Editors:
o. Rienhoff, Marburg
D. A. B. Lindberg, Washington
M. Stefanelli A. Hasman M. Fieschi
J. Talmon
(Eds.)
AIME91
Proceedings of the Third Conference
on Artificial Intelligence in Medicine,
Maastricht, June 24-27, 1991
Springer-Verlag
Berlin Heidelberg New York
London Paris Tokyo
Hong Kong Barcelona
Budapest
Editors
Mario Stefanelli
Universita di Pavia
Dipartmento di Informatica e Sistemistica
Via Abbiategrasso 209, 27100 Pavia, Italia
Arie Hasman
University of Limburg
Department of Medical Informatics
PO Box 616, 6200 MD Maastricht, The Netherlands
Marius Fieschi
Departement d'Information Medicale, Centre Hospitalier
Regional et Universitaire de Marseille
Hopital de la Conception
147 Bd Bailie, 13385 Marseille Cedex 5, France
Jan Talmon
University of Limburg
Department of Medical Informatics
PO Box 616, 6200 MD Maastricht, The Netherlands
ISBN-13: 978-3-540-54144-8 e-ISBN-13: 978-3-642-48650-0
DOl: 10.1007/978-3-642-48650-0
This work is subject to copyright. All rights are reserved, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, re-use of
illustrations, recitation, broadcasting, reproduction on microfilms or in other ways, and
storage in data banks. Duplication of this publication or parts thereof is only permitted
under the provisions of the German Copyright Law of September 9,1965, in its current
version, and a copyright fee must always be paid. Violations fall under the prosecution
act of the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1991
2127/3140-543210 -Printed on acid-free paper
Proceedings editors
Mario Stefanelli, Arie Hasman, Marius Fieschi, Jan Talman
International Programme Committee
Chair: Mario Stefanelli, University of Pavia
K. Adlassnig, Vienna J. Mira-Mira, Madrid
S. Andreassen, Aalborg G. Molino, Torino
R. Engelbrecht, Munich E. Oliveira, Porto
M. Fieschi, Marseilles J-L. Renaud-Salis, Bordeaux
T. Groth, Uppsala N. Saranummi, Tampere
A. Hasman, Maastricht J. Talman, Maastricht
J. Hunter, Aberdeen T. Wetter, Heidelberg
J. Wyatt, London
Local Organising Committee
Chair: Arie Hasman, University of Limburg, Maastricht
Hans Blom, Eindhoven Hilde Pinc9, Leuven
Joaquim de Witte, Maastricht Jan Talman, Maastricht
Joachim Hofener, Aachen Johan van der Lei, Rotterdam
Pieter Zanstra, Groningen
Tutorial Chair: Marius Fieschi, Hopital de la Conception, Marseilles
REFEREES OF PAPERS
K. Adlassnig University of Vienna
S. Andreassen Aalborg University
F. Beltrame University of Genova
C. Berzuini University of Pavia
C.Cobelli University of Padova
l. Console University of Torino
R. Engelbrecht GSF, Neuherberg
M. Fieschi H8pital de la Conception, Marseilles
J. Fox ICRF, London
F. Gremy Centre Hospitalier R6gional, Montpellier
T. Groth Uppsala University
A. Hasman University of Limburg, Maastricht
W. Horn University of Vienna
P. Hucklenbroich Medical University, Hannover
J. Hunter Aberdeen University
W.lrler IRST, Trento
G. Molino Torino University
E. Oliveira University of Porto
A. Rector University of Manchester
J.-l. Renaud-Salis Fondation Bergonie, Bordeaux
N. Saranummi VIT, Tampere
D. Spiegelhalter MRC Biostatics Unit, Cambridge
M. Stefanelli University of Pavia
P. Struss Siemens, Munich
J. Talmon University of Umburg, Maastricht
P. Torasso University of Torino
J. Van Bemmel Erasmus University, Rotterdam
T. Wetter IBM, Heidelberg
J. Wyatt Heart and Lung Institute, London
Table of contents
Keynote Addresses
3
Model-based Image Segmentation: Methods and Applications
P. Suetens, R. Verbeeck, D. Delaere, J. Nuyts, B. Bijnens
Real versus Artificial Expertise: The Development of Cognitive Models of Clinical 25
Reasoning
V.L. Patel, G.J. Groen
Methodology
A Developmental Perspective on the Role of Biomedical Knowledge in Medical
Problem Solving; Implications for AI 41
H.P.A. Boshuizen, H.G. Schmidt, J.L. Talmon
Reconstructing Medical Problem Solving Competence: MACCORD 51
D. Kraus, B. Petkoff, H. Mannebach
The Role of Domain Models in Maintaining Consistency of Large Medical Knowledge
Bases 72
A Glowinski, E. Coiera, M. O'Neil
Knowledge Representation
A Framework for Causal Reasoning with a Functional Approach 85
P. Barahona, M. Veloso
Modelling and Knowledge (Re)presentation within HIOS+ 95
F.M.H.M. Dupuits, A Hasman, E.M.J.J. Ulrichts
Medical Knowledge Representation and Predictive Data Entry 105
W.A Nowlan, AL. Rector
Clinical Applications
A Connectionist Aid to the Early Diagnosis of Myocardial Infarction 119
R.F. Harrison, S.J. Marshall, R.L. Kennedy
Automation of Medical Audit in General Practice 129
W.P.A. Beckers, P.F. de Vries Robbe, E.J. van der Haring, AM. Zwaard, H.G.A.
Mokkink, R.P. T.M. Grol
An Intelligent System for Monitoring Infections in Heart Transplant Recipients 140
C. Larizza, M. Stefanelli, P. Grossi, L. Minoli, A Pan
Measuring Performance of a Bayesian Decision Support System for the Diagnosis of
Rheumatic Disorders 150
H.J. Bernelot Moens, J.K. van der Korst
VIII
The Application of Distributed Artificial Intelligence to Medical Diagnosis 160
P. Burke, R.D. Appel, M. Funk, R.J. Vargas, D.F. Hochstrasser, J.-R. Scherrer
Knowledge Representation of Discharge Summaries 173
R.H. Baud, A.-M. Rassinoux, J.-R. Scherrer
Modelling
Dual Teleological Perspectives in Qualitative Circulatory Analysis 185
K.L. Downing
Physiological Modelling Using RL 198
F. de Geus, E. Rotterdam, S. van Denneheuve/, P. van Emde Boas
Integrated Use of Causal and Algebraic PhYSiological Models to Support
Anaesthetists in Decision Making 211
E. Rotterdam, P. de Vries Robbe, J.P. Zock
Uncertainty Management
Cytotoxic Chemotherapy Monitoring Using Stochastic Simulation on Graphical Models 227
R. Bel/azzi, C. Berzuini, S. Quaglini, D. Spiegelhalter, M. Leaning
A Model-Based Approach to Insulin Adjustment 239
S. Andreassen, R. Hovorka, J. Benn, K.G. Olesen, E.R. Carson
A Blackboard Control Architecture for Therapy Planning 249
S. Quaglini, R. Bellazzi, C. Berzuini, M. Stefanelli, G. Barosi
Knowledge Acquisition
A Comparative Evaluation of Three Approaches to the Acquisition of Medical
Knowledge 263
W. Post, M. W. van Someren
A Knowledge Acquisition Tool for Medical Diagnostic Knowledge-Based Systems 273
G. Lanzola, M. Stefanelli
Machine Learning in Data Rich Domains: Some Experiences from the KAVAS Project 283
J.L. Talmon, P. Braspenning, J. Brender, P. McNair
The User Perspective
Patient's and Physician's Opinion about Computer Expert Systems 297
S. Schewe, J. MDI/er-Nordhorn, S. Mitterwald, M. Schreiber
Designing an Adaptive Interface for EPIAIM 306
D.C. Berry, F. de Rosis
CAP: A Critiquing Expert System for Medical Education 317
L. Console, R. Conto, G. Molino, V. Ripa di Meana, P. Torasso
Keynote Addresses
MODEL-BASED IMAGE SEGMENTATION: METHODS AND APPLICATIONS
P. Suetens1, R. Verbeeck, D. Delaere
Interdisciplinary Research Unit for Radiological Imaging
(ESAT + Radiology)
K.U.Leuven, Kardinaal Mercierlaan 94, B-3001 Heverlee, Belgium
J. Nuyts
ESAT and Department of Nuclear Medicine
B. Bijnens
ESAT and Department of Cardiology
ABSTRACT
We discuss different methods and applications of model-based segmentation of medical
images. In this paper model-based segmentation is defined as the assignment of labels to
pixels or voxels by matching the a priori known object model to the image data. Labels may
have probabilities expressing their uncertainty. Particularly we compare optimization
methods with the knowledge-based system approach.
INTRODUCTION
In this paper we define model-based segmentation as the assignment of labels to pixels by
matching the a priori known object model to the image data. Labels may have probabilities
expressing their uncertainty. While model-based segmentation is a generalization of
traditional segmentation, wh:ch assigns deterministic labels to pixels by using only low
level features such as discontinuity and homogeneity, it is also a special case of object
recognition. Consequently, computational strategies for object recognition, reviewed in
[Suetens, et al., 1991], can also be applied to segmentation problems.
Medical images and/or medical object models are typically complex and require adapted
strategies.
1 P. Suetens is also a senior research associate of the National Fund for Scientific Research,
Belgium.
4
The simplest and traditional object recognition approach relies completely on the local
photometric properties of the image. Hence, the problem is transformed into a problem of
symbolic reasoning by means of some low-level feature extraction. Because objects are not
unambiguously defined by their local photometry, but also by their global geometry and
semantic characteristics, this initial transformation from pixels to symbols fails for complex
images. Complex images are images that do not unambiguously and completely encode the
modeled object characteristics due to poor resolution, noise and/or occlusions. In this case,
it is important to use the complete model, i.e. the a priori knowledge of the object and its
context, early in the procedure at the pixel processing level. The best model instance can
then be found in the image data by means of an optimization approach. In other words, the
problem is represented as one of finding the best description of the image data in terms of
the model descriptive vocabulary. Using this strategy, we have solved three applications,
which are described below: delineation and quantification of the left ventricular heart wall
in ECT images, delineation and quantification of the endocardium in ultrasound image
sequences, and the enhancement of MRA images.
If the knowledge about the object and its context is extensive and uncertain, heuristic
procedures may be unavoidable. The knowledge about such a complex scene often changes
during system development. The interpretation system is therefore expected to be flexible
and understandable. Consequently, for complex scenes the knowledge-based system
approach is an obvious strategy. The strategy to build a knowledge-based system is basically
different from optimization. Unlike optimization, expert sytems reason about symbols
extracted from the image data. To solve photometric ambiguity problems both strategies
can be integrated into a hierarchical strategy. Using this hybrid approach we have
developed an automatic interpretation system for the coronary blood vessels. More about
this below.
DELINEATION OF ECf IMAGES USING GLOBAL CONSTRAINTS AND DYNAMIC
PROGRAMMING
1. OBJECTIVE
The quantification of myocardial perfusion is of great importance in the evaluation of new
thrombolitic agents and in patient follow up. With single photon emission computer
tomography (SPECf) or positron emission tomography (PET) a three-dimensional image of
the perfusion of the heart is obtained. The value in each voxel is approximately
proportional to the blood flow at the corresponding position in the patient body. The
interpretation and quantitative analysis of such images is hampered by the lack of
photometric information in an infarcted region, by the low resolution and the statistical
