A Novel WTS-EDC Network for Multi-Source ECG Signal Classification

Electrocardiography (ECG) is a commonly used diagnostic tool in the clinical setting for detecting cardiovascular diseases (CVDs). However, its manual interpretation can be time-consuming and prone to human error. With the recent advancements in deep learning techniques, there has been an increasing interest in using more robust and generalized deep learning algorithms for automated analysis and interpretation of ECG signals. This article provides a novel model, namely Wavelet Time Scattering-Ensembled Dilated Convolution Network (WTS-EDCN), for ECG signal classification. The classifier is an ensemble of dilated convolutional networks, with each network utilizing wavelet time scattering features extracted from multi-source (MS) ECG signal data. The WTS features efficiently constitute the time–frequency characterization of an ECG signal. WTS features are extracted from different ECG signals containing normal and abnormal rhythms. These features are exploited with a deep neural network (DNN) composed of an ensemble of convolutional nets (CNN) with varied dilation rates. Each varied dilation-rated CNN captures the time–frequency information at different scales, and the ensemble ensures the combined use of the information from individual CNNs. The WTS-EDCN model is experimented on MS ECG signals, which have been diagnosed as having either a normal sinus rhythm, Arrhythmia, or congestive heart failure. The model is successful in classifying the diagnosis with high accuracy. These WTS features have proved significant for multi-source ECG signal classification. The EDC classifier reflects a robust and well-generalized classifier since it constitutes an ensembled approach that is a proven strategy for creating better generalized classification models, especially with multi-source datasets.