Patient Adaptable ECG Beat Classifier Using LDC and EMC Algorithms

Abstract: Electrocardiogram (ECG) waveforms are used to find both regular and irregular patterns in cardiac cycles of patients. In this work, the beats like normal, supraventricular and ventricular beats are identified and classified to the corresponding classes based on combining the decisions of both linear discriminant classifier (LDC) algorithm and Clustering algorithm. Classifications of beats are done by considering both R-R interval features and morphological descriptor features. First the Linear Discriminant Classifier is trained for the three different class and the training beats are taken from three different databases namely, MIT-BIH Arrhythmia, MIT-BIH Supraventricular and MIT-BIH ST Change. Expectation Maximization clustering algorithm (EM) forms patient specific clusters which is based on MOG (mixture of expert) model. Finally the beats in the cluster output is labeled & verified to corresponding classes with the aid of linear discriminant analysis function.

Keywords: LDC,EM, Feature selection,Wavelet transform, patient adaptable.