ABSTRACT
Some candidate medical expert system applications have a significant visual component. Knowledge engineers usually dismiss such task domains as potential expert systems applications. Our success in developing ESCA, a system for evaluating serial coronary angiograms, shows that such task domains should not be dismissed so quickly. We used a symbiotic approach between man and machine, where technologists provide the visual skills with an expert system imitating the conceptual skills of the expert, to produce a partially automated system that is more consistent and cost effective than one that is fully manual. The agreement between the system's conclusions and that of a panel of experts is good. The expert system actually has a slightly higher agreement rate with the expert panel than the agreement rate between two expert panel teams evaluating the same film pair.
