Adaptive abstraction in expert systems for medical diagnosis

Abstract

Preliminary research on applying a novel control scheme to a medical expert system is presented. The solution of most diagnosis problems requires reasoning at multiple abstraction levels. Adaptive abstraction (AA) is a control scheme that enables an expert system to switch automatically to an appropriate level of abstraction to solve the task at hand. The overall knowledge organization, the control mechanisms, and the knowledge base update method of the AA scheme are described along with examples in automated gait analysis. The advantages of the AA scheme are mentioned. Several theoretical issues concerning the design of expert systems are raised, including: determination of a knowledge base's granularity, the downward inclusion property, which is needed for the AA scheme to work, and the necessity of common-sense knowledge for expert systems to overcome brittleness.<>