Abstract ESAO 2021: H46 - Wavelet Transform and Nonlinear SVM for Cardiac Arrhythmia Classification

Abstract

Copyright © 2021 The Author(s). Objectives: Cardiac arrhythmia is a cardiovascular disease caused by impaired electrical conduction of the heart, resulting in irregular rhythms that increase the risk of stroke or can lead to sudden cardiac deaths. This work aims to develop an intelligent software for cardiac arrhythmia classification able to decrease the misdiagnosis probability and reduce the need for human expertise.

Methods: Open-source ECG signals were utilized to produce an arrhythmia classification model using signal processing and Artificial Intelligence (AI) algorithms. The signals were decomposed into 8 levels using discrete wavelet transform. The R-peaks were detected as the maximum points, while the Q and S were detected as the first local minima points on either side of the R-peaks. A set of nine features were extracted and fed into a support vector machine (SVM), classifying the signals in two classes: normal sinus rhythm and arrhythmia.

Results: The proposed SVM model uses a Gaussian kernel function with a scale of 0.75, and the best total accuracy of 92.2% was achieved. The average area under the receiver operating characteristic curve was 0.82. Also, the confusion matrix of the model indicates a true positive rate (TPR) of 93.3%, for normal sinus rhythm and 80.5% for arrhythmia classification, respectively.

Conclusion: The proposed methods provide a good base towards developing a software for accurate arrhythmia classification. The experimental results demonstrate its efficiency. For the future work, deep learning methods such as Convolutional Neural Networks will be explored to automatically extract the representative features, increasing the accuracy of the system.