A Probabilistic Causal Model for Diagnosis of Liver Disorders

Directed probabilistic graphs, such as Bayesian networks, are useful tools for coherent representation of and reasoning with uncertain knowledge. They are based on the sound foundations of probability theory and they readily combine available statistics with expert judgment. When extended with decision options and measures of desirability of outcomes (utilities), they support decision making. This paper describes our work in progress on a probabilistic causal model for diagnosis of liver disorders that we plan to apply in both clinical practice and medical training. The model, and especially its numerical parameters, is based on patient records at the Gastroenterological Clinic of the Institute of Food and Feeding in Warsaw, collected over the period of several years. We present the model and report initial results of our diagnostic performance tests.