A software engineering approach to the design of a medical expert system

Abstract

We present a case study of the use of formal specification techniques in the design of a knowledge-based system. The Z specification notation [ll] is used in the development of a medical flow chart program, beginning with a highly abstract notion ofinference and proceeding by refinement toward a particular implementation for flow charts. Diagnostic inference is modelled b y idempotent, decreasing functions. The flow chart was found to be significantly more accurate than a statistical classifier (Independence Bayes).Formal specification provides several benefits. Design decisions are made explicit and are documented concisely, and the results of making alternative choices can be explored within the same framework. The formal specification also suggests approaches that probably would not otherwise have been consadered.

The Relative Accuracy of a Variety of Medical Diagnostic Programs

“…Independence Bayes is significantly more accurate than the flowchart (p = 0.017). This reverses an earlier conclusion based on a much smaller sample of 202 training cases [55]. The inferential rule-based system made fewer errors than the flowchart, but the difference does not reach statistical significance (p = 0.079).…”

“…Our solution [55] is a ternary decision tree. Each non-terminal node is a Boolean expression over atomic propositions, like an antecedent of one of the probabilistic rules shown above.…”

“…Diagnosis entails iterative traversal of the flowchart, each pass augmenting the knowledge about the patient, until a fixed point is eventually reached. It is easy to show that this procedure has polynomial time complexity in the number of nodes in the tree [55].…”