Medical diagnostic expert systems: performance vs. representation

A research effort is described which represents an inquiry into an important problem of automated acquisition, indexing, retrieval, and effective use of knowledge in diagnostic tasks. The principal tool is INC2, an incremental concept formation system which automates both the design and the use of diagnostic decision-support systems by a novice. The system's prediction performance is evaluated in the domains of breast cancer, primary tumor, and audiology cases, relative to the language used for representing concepts. The study includes the whole continuum of concept representations from logical to probabilistic ones. The results demonstrate that the quality of performance indeed depends on the chosen representation language.<>