ABSTRACT
A neural network approach to computer-aided diagnostic systems for coronary artery diseases is described as one of the case studies in cardiac nuclear medicine. Recently, we have been developing a computerized system by using artificial neural networks, called ‘BULLsNET’, which can aid the physician in the detection and classification of coronary artery diseases in 201 Tl myocardial SPECT bull’s-eye images. Three-layer feedforward neural networks with a backpropagation algorithm were employed, in which whole or partial images were fed into the input layer. The BULLsNET system, which includes two major neural-network-based elements for the analysis of ‘EXTENT’ and ‘SEVERITY’ bull’s-eye images, was trained using pairs of training input images and the desired output data (‘correct’ diagnosis). The system classified the input image data into eight cases, that is, one normal case and seven different types of abnormal cases. The results showed that the recognition performance of the system was comparable to that of a two-year RI-experienced physician. Our study suggests that the neural network approach is useful for developing a computer-aided diagnostic system for coronary artery diseases in myocardial SPECT bull’s-eye images.
