Deep learning-based system to predict cardiac arrhythmia using hybrid features of transform techniques

Abstract

• Decomposed of ECG signals with discrete wavelet transform (DWT), empirical mode decomposition (EMD) and variational mode decomposition (VMD) methods. • Extraction of informative time-frequency based multi-domain features. • Ranking of features based on weightage in Chi-squared test and PSO search methods. • Classification of ECG beats using deep neural network. • Comparison of results with deep learning and discussion with published works. An early and accurate detection of arrhythmias is essential reduce the mortality rate due to cardiac diseases. Manual screening of the electrocardiogram (ECG) signals are time consuming, strenuous, and liable to human errors. This article proposes a deep learning approach for automated detection of cardiac arrhythmia using RCG signals fro MIT-BIH database. Various decomposition techniques namely: discrete wavelet transform (DWT), empirical mode decomposition (EMD) and variational mode decomposition (VMD) are used to de-noise the ECG signal. The time-frequency based multi-domain features are extracted from the various coefficients of the sub-bands from de-noised signals. These obtained features are ranked based on Chi-squared test and particle swarm optimization (PSO) based methods to select the best informative features for better classification accuracy. The hybrid features was classified with deep neural network (DNN) with ten-fold cross validation strategy in classifying five types of ECG beats. The best results was obtained with an accuracy of 99.75% with less computational complexity of 0.14 s using Chi squared selection approach. Thus the proposed model can be used in the hospitals set-up to automatically screen the abnormal ECG beats.