Detection and Classification of Heart Arrhythmias by Convolutional Neural Network

The arrythmias are found frequently in patients and it could be an indicator of cardiovascular diseases or an immediate health danger. This is why it is important to identify and classify correctly these irregularities in the heartbeat. The present paper proposes the architecture and evaluates the performance of a 1D convolutional neural network capable of classifying ECG signals in five classes, four types of arrythmias and one considered as normal. The five classes were first balanced and then separeted in 70% and 30% for training and testing, respectively. Metrics were calculated for each of the five classes: precision, recall, specificity and F1 score. The mininum value obtained was 0.97 in recall for the normal class, and the maximum was 0.99 for the specificity of all the classes and most of the other metrics.