A New Artificial Neural Network Method for the Interpretation of ECGs

Abstract

Acute myocardial infarct could be a life-threatening disease. Early treatment could be life saving, but treatment of patients not suffering from infarct may cause serious complications. Therefore a rapid decision regarding diagnosis and treatment is of great importance. The physician at the emergency department has to rely on the electrocardiogram (ECG) for the diagnosis. A reliable computer-aided interpretation would be of great value. The purpose of this study was to develop a decision-support system for the diagnosis of acute myocardial infarct using a method that can estimate the error of an artificial neural network. The material consisted of 3088 ECGs from both patients with and without acute myocardial infarction. The ECG material was randomly divided into three groups. The first group was used to train the neural network for the diagnosis of acute myocardial infarct. The second group was used to calculate the error of the network outputs and the third group was used to test the network performance and to obtain error estimates. The performance of the neural network, measured as the area under the receiver operating characteristic (ROC) curve, was 0.81. The 25% test ECGs with the lowest error estimates had an area under the ROC curve as high as 0.92, i.e. most of these ECGs were correctly classified. The results indicates that neural networks can be trained to diagnose acute myocardial infarct and to signal when the advice is given with high confidence or should be considered more carefully.