Using a qualitative probabilistic network to explain diagnostic reasoning in an expert system for chest pain diagnosis

Abstract

A chest pain expert system, which diagnoses the cause of chest pain in patients admitted to the Emergency Department, was developed. The system relies on a Bayesian belief network (BBN) to combine evidence in a cumulative manner and provide a quantitative measure of certainty in the final diagnoses. Probabilistic schemes support reasoning at levels ranging from purely quantitative to purely qualitative. Qualitative probabilistic networks (QPNs) are qualitative abstractions of BBNs replacing numerical probabilities with qualitative influences. QPNs support explanations about the structure and reasoning of probabilistic models. We show that a QPN effectively satisfies the explanation requirements of our expert system. By combining a BBN and the corresponding QPN in the expert system, robustness of performance and understandability of reasoning are achieved. The system produces results which are compatible with the diagnoses of doctors.