ECG beats extraction and classification using radial basis function neural networks

Abstract

This paper aims the design of an ECG diagnosis system that helps physicians in the interpretation of ECG signals. This system preprocesses and extracts the ECG beats of an ECG record and some feature extraction techniques are invoked to get a feature vector that represents the main characteristics of the ECG wave. After that a well trained RBF artificial neural network is used as a classifier for four different ECG heart conditions selected from MIT-BIH arrhythmia database. The ECG samples were processed and normalized to produce a set of reduced feature vectors. The results of sensitivity, specificity accuracy and recognition rate of the system are analyzed to find the best RBF neural network for ECG classification. Different ECG feature vectors composed of averaged amplitude values, DCT coefficients, DFT coefficients, and wavelet coefficients were used as inputs to the neural network. Among different feature sets, it was found that an RBF which has one layer and the feature vector 61 inputs, and 20 neurons possessed the best performance with highest recognition rate of 95 % for four cardiac conditions. © 2013 Springer.