Efficient methodology for electrocardiogram beat classification

Abstract

Electrocardiogram (ECG) beat behaves as a non-linear and non-stationary signal. Since most of the existing data processing tools are poor alternatives for processing such signals, Hilbert–Huang transform (HHT) proves to be an efficient method as it deals with a time-varying frequency spectrum. In this study, a new and efficient methodology is proposed using HHT for feature selection which includes a set of essential features such as weighted mean frequency, Kolmogorov complexity and other statistical features (median, standard deviation, kurtosis, skewness and central moment) computed from the intrinsic mode functions extracted using the empirical mode decomposition (EMD) algorithm. Further, one-against-one multi-class support vector machine is employed for the classification of six generic ECG beats, namely: normal, left bundle branch block, right bundle branch block, premature ventricular contraction, paced beat and atrial premature beat. The classification process in this study yields better results than existing methodologies in terms of classification accuracy equal to 99.51% along with sensitivity, specificity and positive predictivity of 98.64, 99.77 and 98.17%, respectively.