CAPTURING LONG-TERM DEPENDENCIES IN ECG SIGNALS USING DEEP CONVOLUTIONAL NEURAL NETWORKS FOR CARDIAC ARRHYTHMIA DETECTION

Abstract

The analysis of electrocardiograms (ECGs) is a major supporting factor in the detection of cardiac arrhythmias, which are basically electrical activity disorders in the heart. But, on the other hand, the conventional means of diagnosis are slow and susceptible to mistakes made by humans. This paper suggests an automatic deep learning framework that fully analyzes the ECG signal through the use of Deep Convolutional Neural Networks (CNNs). The model utilizes a Continuous Wavelet Transform (CWT) method to change the one-dimensional ECG signals into two-dimensional time-frequency scalograms, which are then classified by transfer-learningbased architectures such as AlexNet and Inception-V3. A series of tests on the MIT-BIH Arrhythmia Database yield an accuracy of 99.79%, which is a sign of the performance going beyond the detection of heartbeat abnormalities like atrial fibrillation and premature contractions. The method suggested is a major reduction of preprocessing complexity and at the same time keeping the high level of precision and generalization, thus offering a solid base for real-time clinical diagnosis and decision support.