ECG Beat Classification Based on Stationary Wavelet Transform

Abstract

ECG processing is a non-invasive technique that is frequently used for diagnosis of various cardiac diseases. One of the crucial steps of an ECG diagnosis system is the heartbeat classification. In this work, we propose a new method for QRS complex classification based on Stationary Wavelet Transform (SWT), and two classifiers, which are Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). In our scheme, SWT was used to extract the discriminatory features from the useful frequency sub-bands for each QRS complex class. The extracted features were used as inputs of SVM and KNN in order to classify five types of heartbeats, which are Normal (N), Premature Ventricular Contraction (PVC), Atrium Premature Contraction (APC), Left Bundle Branch Block (LBBB), and Right Bundle Branch Block (RBBB). The experimental results obtained on MIT-BIH Arrythmia database (MITDB), show that the proposed system yields acceptable performances with an overall classification accuracy of 98.56% and 98.74% for KNN and SVM classifiers respectively, using the 10-cross validation technique.