ECG Arrhythmia Classification using Discrete Wavelet Transform and Artificial Neural Network

Electrocardiogram (ECG) is used as one of the important diagnostic tool for the detection of the health of a heart. Growing number of heart patients has necessitated development of automatic detection techniques for detecting various abnormalities or arrhythmias of the heart to reduce pressure on physicians and share their load. The present work will help in developing a computer based system that will be able to categorize the ECG signals. In this paper artificial neural network (ANN) based classifier is developed, where discrete wavelet transform (DWT) is used for preprocessing and feature extraction purposes and neural network designed is used to classify five types of arrhythmias namely Left Bundle Branch Block (LBBB), Right Bundle Branch Block (RBBB), Paced Beat (PB), Atrial Premature Beat (APB) and First degree AV Block (AVB) beats apart from normal (N) beats. MIT-BIH ECG arrhythmia database acquired from pysionet.org is used for analysis purpose. The main aim of the present work is to find out optimum feature set and number of hidden layer neurons which increases classification performance of the neural network based classifier. Simulation in MATLAB R2009a showed that the classification accuracy is more if both morphological features and wavelet coefficients are together used for training the neural network than only morphological feature or wavelet coefficients. The proposed neural network (NN) based global classifier provides enhanced performance sensitivity more than 65%, specificity more than 92%, positive predictive value more than 63%, negative predictive value of more than 92% and classification accuracy more than 87%. It is observed that performance of classifier can increased if the number of neurons in the hidden layer is significantly increased.