Great Expectations: Expectation Based Reasoning in Medical Diagnosis

Several different approaches to knowledge representation for medical expert systems have been explored. We suggest that a modified version of the script formalism, which we term “expectation-based reasoning”, may offer an additional knowledge representation for medical information, addressing certain shortcomings of previous approaches. This representation can drive expert system analysis for diagnosis and workup advice. The script formalism structures the knowledge base around a set of temporally sequenced event frames, each containing a list of default expectations. This model, we believe, allows straightforward knowledge generation from a domain expert, since it may closely parallel a central aspect of human clinical decision-making: that of projecting assumptions for a “hypothesize-and-test” inference mechanism. A prototype expectation-based expert system, OSCAR, is under development to explore this approach.