Integrating probabilistic and rule-based systems for clinical differential diagnosis

Abstract

CLAUDE is a hybrid expert system for differential diagnosis that combines a rule-based and probabilistic system, integrating their independent opinions with a neural network. CLAUDE's application was to classify 17 central auditory diagnostic areas by analyzing a client's case history. Two network structures were examined, one connecting only those nodes applying to the same class within a diagnostic area, and the other fully connected. CLAUDE performed much better than either a rule-based system or a probabilistic system and made correct judgments when neither system could provide a diagnosis. Furthermore, it was able to classify 75% correctly, even in the presence of incomplete data. Thus, two individual systems using different criteria and balanced by a neural network provided more accurate results than either a probabilistic or a rule-based system alone.<>