4 - Patient-specific ECG beat classification using EMD and deep learning-based technique

Abstract

Electrocardiogram (ECG) is a vital tool to identify cardiac disorders effectively in the medical field. However, manual detection of crucial episodes in ambulatory ECG is very difficult. Therefore, an automatic diagnosis system is needed to detect the beats automatically. High-quality ECG signals are required to classify ECG beats properly but in real time, acquired ECG signals are severely affected by several noises due to wearable sensors. In this chapter, a customized deep learning technique is proposed for ECG beat detection. Two steps have been followed in this work: (i) preprocessing and (ii) classification. In the preprocessing step, the ECG signal is segmented into individual beats based on the location of the R-peak. Then, intrinsic mode functions (IMFs) are decomposed from the individual ECG beats using the empirical mode decomposition (EMD) technique. Significant IMFs are selected to remove the high-frequency noise from the ECG signal. Finally, the resulting ECG beats are utilized for ECG beat detection with the help of a deep learning-based custom model in the classification stage. Six models are utilized in the proposed deep learning architecture with different strides in convolutional layers that extract significant features from the input. Three different inputs are fed to the proposed model: the complete signal (−128 to +128), left (−128 to 0), and right (0 to +128) portions from the R-peak. The proposed model can extract different in-depth features using these three inputs. Standard publicly available Massachusetts Institute of Technology-Beth Israel Hospital cardiac arrhythmia database is used to check the ability of the EMD-based deep learning technique. In this work, five different ECG beats are detected as suggested by the Association for the Advancement of Medical Instrumentation. The experimental results show that the proposed method provides better performance compared to the state-of-the-art techniques.