Feature extraction from a novel ECG model for arrhythmia diagnosis

Abstract

Feature extraction is a crucial aspect of computer-aided arrhythmia diagnosis using an electrocardiogram (ECG). A location, width and magnitude (LWM) model is proposed for extracting each wave's features in the ECG. The model is a stream of Gaussian function in which three parameters (the expected value, variance and amplitude) are applied to approximate the P wave, QRS wave and T wave. Moreover, the features such as the P–Q intervals, S–T intervals, and so on are easily obtained. Then, a mixed approach is presented for estimating the parameters of a real ECG signal. To illustrate this model's associated advantages, the extracted parameters combined with R–R intervals are fed to three classifiers for arrhythmia diagnoses. Two kinds of arrhythmias, including the premature ventricular contraction (PVC) heartbeats and the atrial premature complexes (APC) heartbeats, are diagnosed from normal beats using the data from the MIT–BIH arrhythmia database. The results in this study demonstrate that using these parameters results in more accurate and universal arrhythmia diagnoses.