An intelligent classifier for cardiac arrhythmias recognition

The aim of this article is to propose an intelligent electrocardiogram classifier. The classifier is similar to probabilistic neural networks. In these networks, a user needs to set some parameters optionally. Improper selections may decrease the performance drastically. The proposed method needs no optional parameter settings and all required parameters are extracted from the statistics of the input signals. The proposed classifier has two layers and a database of known signals that has been categorized and labeled to M classes based on their similarities. The first layer calculates the similarities of the input unknown signal to the known signals of each class using Basis Radial functions and outputs Bayesian variables equal to the number of classes. The second layer is just a maximum selector of these Bayesian variables as the winner. In fact, it indicates that the input signal most probably belongs to the class in which Bayesian variable is maximum. Five classes of ECG signals from MIT-BIH arrhythmia database are selected to illustrate the good performance of the non-invasive proposed classifier compared to the previous methods. Moreover, acceptable low computational complexity and robustness against high noise are significant features of the proposed classifier.