Modeling diagnostic problem-solving at multiple levels of abstraction

Abstract

A precise domain-independent computational model for diagnostic problem-solving at multiple levels of abstraction is presented. The knowledge representation framework allows causal knowledge to be represented in a precise yet natural way that reflects a human diagnostician's experience in guiding diagnostic reasoning at multiple levels of abstraction. The inference mechanism clearly defines how to form plausible diagnostic hypotheses guided by explicit causal links and the principle of parsimonious covering. It permits the efficient formation of high-level diagnostic hypotheses while at the same time ensuring that all plausible diagnostic alternatives will be considered if one wishes to reason at sufficiently detailed levels. A prototype implementation is described to demonstrate that the proposed inference model is natural for capturing a small but representative fragment of medical causal knowledge, and, with the addition of reasonable heuristics, that the inference mechanism exhibits desirable problem-solving behavior. >