Evidential inference with embedded pattern classifiers: towards a medical expert system for diagnosis

In this paper, we describe a novel quantitative approach to medical diagnosis. Drawing on sound mathematical theories as well as the promising results of previous experiments, the proposed approach provides a computational solution to the modeling and aggregating of partial evidential observations to assure an accurate diagnosis. This approach is particularly useful for diagnosing cases in which a complete set of symptoms is too difficult to observe and the diagnostic judgments are subject to human errors. This paper presents several experiments in which real-world diagnostic problems were investigated. In particular, it attempts to show that (1) with a limited number of case samples, our implication-induction algorithm is capable of inducing implication networks useful for making evidential inferences based on partial observations, (2) observation driven by a network entropy optimization mechanism is effective in reducing the uncertainty of predicted events, and (3) the network-based evidentially predicted events or attributes can provide sufficient information for pattern classification.