Qualitative models in medical diagnosis

Diagnostic systems based solely on associative knowledge are able to draw accurate conclusions in acceptable times but they do not capture all the available medical knowledge. Some of this knowledge, even if incomplete, is sufficiently precise to allow the formulation of qualitative models. The aim of this paper is to show how qualitative models can be exploited in a medical diagnostic system. We present a system, NEOANEMIA, that integrates first generation knowledge representation formalisms (frames and production rules) with qualitative pathophysiological models to diagnose hematologic disorders causing anemia. To this end, qualitative models of iron metabolism, erythropoietin metabolism, and red cell production and destruction have been formulated. The key ideas of our work are: abducing diagnostic hypotheses from observed problem features, modeling pathophysiological systems with dynamic qualitative models, predicting pathophysiological behaviours by qualitative model simulation, comparing clinical observations against simulation results, and when necessary, incrementally creating and testing multiple diagnostic hypotheses. In this way the performance of a diagnostic expert system can be highly enhanced.