Artificial neural network for the exercise electrocardiographic detection of coronary artery disease

The objective of this study was to apply an artificial neural network in computerized exercise ECG analysis for detection of coronary artery disease (CAD). A multilayer perceptron neural network with backpropagation weight updating was developed and validated with a study population of 347 patients. The neural network was fed with 27 input nodes, two hidden nodes and one output node. The input nodes consisted of three exercise ECG variables, the end-exercise ST-segment depression, ST depression/heart rate (ST/HR) index and ST/HR hysteresis, each of which determined from leads I, II, III, aVF, V2, V3, V4, V5, V6. The desired output was O for patients without CAD and I for patients with CAD which was verified by coronary angiography. The performance of the neural network measured as the area under the receiver operating characteristic (ROC) curve was 91.5% The discriminative capacity of the network was greater than provided by any of the 27 single inputs alone. In conclusion, the results suggest that the diagnostic accuracy of the exercise electrocardiographic ST/HR analysis in detection of CAD can be further improved by using neural networks.