Classification of heart rhythm disorders using instructive features and artificial neural networks

Abstract

Accurate detection of the heart rhythm disorders at an early stage is helpful for improving survival rate. This paper presents an automated detection and classification methods of cardiac arrhythmia by time-frequency analysis of the recorded ECG signals from MIT-BIH database. The discrete wavelet transform has been used to eliminate noises in order to enhance the quality of signals and adaptive thresholding-based Hilbert transform has been used to find precise R-peaks. Temporal, morphological and statistical features were extracted from each heartbeat and has been used as input to the classifier to detect five cardiac arrhythmia beats. The results show less detection error rate of 0.17% in detecting QRS complex. The MLP-BP, RBF-NN, and the PNN classifiers provide an average accuracy of 98.72%, 99.77% and 99.16% respectively. The result indicates the efficiency of the proposed method in classifying ECG beats which is useful in diagnosis of cardiac arrhythmias.