Alternatives to Bayes?

Abstract

Recent proposals have suggested that rule-based systems of diagnostic inference are an attractive medium for computer-aided diagnosis, in part because clinicians find their behaviour easy to understand. Bayesian systems have been more prominent in this field to date, but no direct comparison of their clinical abilities has been reported. A rule-based system that was closely modelled on clinical thinking is described and a quantitative comparison with a successful Bayesian system for the diagnosis of »dyspepsia« is presented. The results suggest that the rule-based approach may have considerable potential as an efficient alternative to Bayesian inference.