ECG beat classification based on discrete wavelet transformation and nearest neighbour classifier

Abstract

Myocardial infarction (MI) is a coronary artery disease acquired due to the lack of blood supply in one or more sections of the myocardium, resulting in necrosis in that region. It has different types based on the region of necrosis. In this paper, a statistical approach for classification of Anteroseptal MI (ASMI) is proposed. The first step of the method involves noise elimination and feature extraction from the Electrocardiogram (ECG) signals, using multi-resolution wavelet analysis and thresholding-based techniques. In the next step a classification scheme is developed using the nearest neighbour classification rule (NN rule). Both temporal and amplitude features relevant for automatic ASMI diagnosis are extracted from four chest leads v1–v4. The distance metric for NN classifier is calculated using both Euclidian distance and Mahalanobis distance. A relative comparison between these two techniques reveals that the later is superior to the former, as evident from the classification accuracy. The proposed method is tested and validated using the PTB diagnostic database. Classification accuracy for Mahalanobis distance and Euclidean distance-based NN rule are 95.14% and 81.83%, respectively.