Artificial neural network nuclear cardiac stress test interpretation

A neural network can be used to provide a preliminary interpretation of nuclear cardiac stress tests, identifying areas of ischemia and infarction. A three-layer neural network was trained with bulls-eye plot images of simulated data and actual patient data using a standard backpropagation algorithm. Testing was then performed using 100 male patients who had undergone cardiac catheterization within two months of their stress thallium study. Using cardiac catheterization as the gold standard, the sensitivity and positive predictive values were calculated for studies read by a trained physician utilizing the tomographic images as well as bulls-eye plots, and for studies read by the neural network utilizing the bulls-eye plots alone. The neural network´s sensitivity (84%) approached that of the experienced physician (95%) and the positive predictive value (90%) was similar to that of the physician (91%)