Efficient wavelet families for ECG classification using neural classifiers

Abstract

The emerging technologies in science and engineering helps in developing predictive models for accurate diagnosis (wellness) of diseases such as electrocardiogram (ECG) and electroencephalogram (EEG) related to heart and brain respectively. In this paper, a comparative study of neural classifiers such as back propagation neural network (BPN), feed forward network (FFN) and radial basis function neural network (RBFNN) on ECG signals using classical wavelet transform is presented. The ECG files of normal and abnormal classes depending upon the beats present in the signal are taken from the standard arrhythmia database from MIT-BIH. After removal of noise from the original signal, the features selected using discrete wavelet transform with different wavelet families like daubechies, symlet, coiflet, biorthogonal and reverse biorthogonal are supplied as input to different neural classifiers. The performance of different neural classifiers are figured out by the sensitivity (Se), the positive predictivity (PP), specificity (Sp) and accuracy (Acc). From the experimental results, it is inferred that RBFNN gives Se 100, PP 100, Sp 100, and Acc 100% with the feature extraction using daubechies wavelet as compared to other classifiers. This gives an intellectual diagnosis approach to health care system using signal processing and neural network.