Application of expert systems to mammographic image analysis.

Abstract

A prototype expert system was designed for differentiation of 16 circumscribed breast abnormalities utilizing findings identified by a human observer from X-ray mammograms, clinical data, and patient history. An off-the-shelf expert system shell served as the foundation for the rule base. The system runs on standard microcomputer hardware. Preliminary results suggest the expert system may be valuable in improving the ability of relatively unskilled observers to screen X-ray mammograms. Overall, radiology residents with limited experience in mammographic image analysis obtained a diagnostic accuracy level of 40% on their own, whereas they attained 73% accuracy level when using the expert system. A group of biomedical engineering students with no prior experience in mammographic image analysis obtained on the average an 80% diagnostic accuracy level utilizing the expert system in comparison with the 70% average accuracy level obtained by practicing radiologists without use of the system.