Abstract
We have been developing a computerized system (“BULLsNET”) that can help the radiologist's diagnosis in the detection and classification of coronary artery diseases from myocardial SPECT bull's-eye stress images. We newly construct a system by adding a rest (redistribution) image analysis to the stress image analysis. The system consists of 2 major neural networks (NN) for recognition of 2 types of bull's-eye images. “Extent” and “Severity”, acquired in both stress and redistribution examinations. When the confidence level of the “Extent NN” is lower than 0.9, the “Severity NN”, in which each part of the coronary vessel regions in the bull's-eye plot is examined by three sub-neural networks, is applied. Recovery score is determined based on the area and intensity differences between SEVERITY images. Our new BULLsNET with the redistribution image analysis can classify three interpretations, ischemia and partial or total infarction. The performance of the redistribution image analysis was comparable to that of experienced radiologists. Our studies suggest that the neural network approach is useful for the computer-assisted interpretation of bull's-eye images. For further investigation to improve the BULLsNET, the stress image analysis will have to be upgraded.
