Automated ECG heartbeat classification by combining a multilayer perceptron neural network with enhanced particle swarm optimization algorithm

Abstract

The electrocardiograms analysis helps cardiologists to diagnosis different cardiac disorders. In this paper, we have proposed a new training algorithm of a multilayer perceptron (MLP) neural network to improve the classification task of arrhythmias. For this purpose, an enhanced particle swarm optimization algorithm (EPSO) was implemented by proposing a new updating equation of the inertia weight factor through iterations. In this work, five predominant types of ECG beats from MIT-BIH database are taken as desired output classes of the MLP neural network. The considered heartbeats are normal (N), premature ventricular contraction (PVC), atrial premature contraction (APC), right bundle branch block (RBBB), and left bundle branch block (LBBB). The accurate classification and recognition of different classes have been focused on the extraction of the more relevant and appropriate feature parameters. For this purpose, we have implemented the classifier using several kinds of features (morphological, wavelet, or statistical features) and also with different combinations of them in order to select the more pertinent features to the studied classes of ECG beats. The evaluation of the proposed EPSO-MLPNN classification system on ECG data from MIT-BIH database achieves the good classification results in terms of three statistical parameters. We have computed a specificity of 99.81%, a sensitivity of 97.67%, and an accuracy of 98.56%. Furthermore, these results demonstrate a clear improvement in the classification performances compared with those of other methods of classification carried out on MIT-BIH database. The proposed classification system can be used as a powerful tool in the field of arrhythmia diagnostic.