Differential diagnosis generation from a causal network with probabilities

One of the problems with building expert systems for medical domains is handling the uncertainty that exists in almost every task. The authors solve this problem, in the context of a system for diagnosing and managing patients with cardiovascular disease, by developing a heuristic method for generating likely causal hypotheses for a set of clinical findings. The method uses the potential causal pathways to determine sets of primary causes that could produce the findings. From each set it builds a hypothesis by determining the most probable explanation for each finding and adding that explanation to the hypothesis. Hypotheses are compared by computing the overall probability of each as an explanation for the findings. The most likely hypotheses are presented to the user as a detailed differential list. The method has been tested in a network with 150 physiological nodes and about 280 potential findings. It produces differential diagnoses for cases with 10 to 15 abnormal findings in a few minutes. Each hypothesis contains 20 to 30 physiological nodes which can be displayed with their interconnections, showing the user how the findings can be accounted for by the possible causes