ECG Beat Classification Using Feature Extraction from Wavelet Packets of R Wave Window

Abstract

AbstractIn this paper we describe proposal of a feature extraction method for classification of different beat types based on wavelet packets decomposition, nodes selection and template matching. In this work we investigate the use of wavelet packets transform as a mean to extract features capable of providing the information needed by a classifier for discrimination between normal beats (N), premature ventricular (V), left bundle branch block beat ‘L’ and right bundle branch block beat ‘R’. We have used our approach of feature extraction which is based on wavelet packets decomposition of window with QRS complex and on the method of wavelet packet tree node selection. This selection has been performed as finding of the path with maximum of relative entropy between current beat and template. In comparison with previous work we have used template from each type of beat and the final features have been given as differencies of current beat and all the templates.We have used the MIT-BIH arrhythmia database for testing this approach. The database was divided into two subsets for training and testing the classifier. Results of the classification are evaluated by the sensitivity (Se), specificity (Sp) and overall accuracy. We obtained sensitivity 81,1%, specificity 98,5% and 97,8% overall accuracy for ventricular beat.KeywordsFeature ExtractionRelative EntropyOrthogonal BasisWavelet PacketRight Bundle Branch BlockThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.