5 - Empirical wavelet transform and deep learning-based technique for ECG beat classification

Abstract

This chapter proposes a novel empirical wavelet transform (EWT)-based deep learning technique for automatic electrocardiogram (ECG) beat classification in a patient-specific way. Preprocessing and classification are the important steps in the proposed method. In preprocessing step, the ECG signal is processed for R-peak detection, beat segmentation, and denoising. Further, automatic feature extraction and beat detection are carried out in the classification stage. The major limitation of the automatic classification systems based on deep learning is that they require noise-free signals. However, several noises contaminate the ECG signal during the acquisition time, which will affect the signal's quality. To overcome this issue, EWT is used in this work to decompose the signal into different modes. After successful decomposition, low-frequency modes are added to generate a high-quality signal. These high-quality ECG beats are used to train and test the customized deep learning model. As per the Association for the Advancement of Medical Instrumentation (AAMI), five ECG beats are classified in this work. The proposed classification system is tested on the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) publicly available database, and it performs better than the existing state-of-the-art techniques.