lhohner@watdcsu.waterloo.edu (L.Hohner - Statistics) (03/15/89)
S E M I N A R
DEPARTMENT OF STATISTICS AND ACTUARIAL SCIENCE
UNIVERSITY OF WATERLOO
SPEAKER: Dr. David Spiegelhalter
Medical Research Council
Cambridge University
TITLE: Combining Expert Judgement and Statistical Analysis
in Medical Expert Systems
ABSTRACT
Artificial intelligence approaches to medical classification and prediction
systems are polarised between expert-derived networks of propositions and
data-derived 'neural networks'. A middle path is suggested, using a
causal network of variables whose connections are qualified by conditional
probabilities. A full probability model is thus obtained, and propagation
of evidence from fragmentary information takes the form of calculating
exact conditional probabilities on variables (e.g. diseases) of interest.
Fast algorithms for this propagation will be discussed, which extend the
work of Pearl. Expert judgement is used both for initial qualitative structuring
of the network and for assessing imprecise conditional probabilities.
Accumulated data then updates the probability assessments using adapted
Bayesian methods. Applications in paediatric cardiology, electromyography and
adverse drug reactions will be described.
DATE: Thursday, March 16, 1989
TIME: 3:30 p.m.
PLACE: MC 5158