A FEED FORWARD NEURAL-NETWORK FOR CLASSIFICATION OF BULLS-EYE MYOCARDIAL PERFUSION IMAGES

Identification of hypoperfused areas in myocardial perfusion single-ph oton emission tomography studies can be aided by bull's-eye representa tion of raw counts, lesion extent and lesion severity, the latter two being produced by comparison of the raw bull's-eye data with a normal data base. An artificial intelligence technique which is presently bec oming widely popular and which is particularly suitable for pattern re cognition is that of artificial neural network. We have studied the ab ility of feed forward patterns from bull's-eye capability to predict l esion presence comparison with a normal data base. Studies were undert aken on both simulation data and on real stress-rest data obtained fro m 410 male patients undergoing routine thallium-201 myocardial perfusi on scintigraphy. The ability of trained neural networks to predict les ion presence was quantified by calculating the areas under receiver op erating characteristic curves. Figures as high as 0.96 for non-preclas sified patient data were obtained, corresponding to an accuracy of 92% . The results demonstrate that neural networks can accurately classify patterns from bull's-eye myocardial perfusion images and detect the p resence of hypoperfused areas without the need for comparison with a n ormal data base. Preliminary work suggests that this technique could b e used to study perfusion patterns in the myocardium and their correla tion with clinical parameters.