RETRACTED ARTICLE: Composite feature vector based cardiac arrhythmia classification using convolutional neural networks

Abstract

Electrocardiogram analysis for the classification of several cardiac arrhythmias has gained a significant research importance in the medical field. Towards such objective, this paper proposed a novel approach based on the fusion of multiple features extracted from a signal through various methods and Convolution Neural Networks. The multiple features are precisely consisting of morphological features, temporal features, and statistical features. Every electrocardiogram signal is initially pre-processed to remove the base line and then processed for segmentation through a simple strategy. Further, for every heart beat segment, three different set of features are extracted. Among them, morphological features are obtained through Dual Tree Complex Wavelet Transform and remaining features are extracted through statistical measures. Further, Principal Component Analysis is applied over the morphological feature set to reduce the dimensionality. Finally, a composite and final feature vector is formulated and then fed to Convolutional Neural Networks classifier to predict the label for a given input heartbeat. Simulation experiments conducted through the MIT-BIH benchmark database exhibited that the proposed system achieves better classification accuracy and on an average, it of 98%. Compared with state-of-art methods, the improvement is approximately 5%.