ECG Signal Denoising and its Significance in Heartbeat Classification for Arrhythmia Detection

Abstract

Electrocardiogram (ECG) signal represents the electrical picture of newlinethe heart and it is very significant to the heartbeat classification and newlineArrhythmia detection task. The non-invasive nature of ECG makes it newlinea preferable choice for cardiac diagnosis. But, the major issue with the newlineECG is the addition of various noises and artifacts during its recording newlineand acquisition process. This thesis proposes signal processing as newlinewell as deep learning based approaches for filtering noises from the newlineECG signal. The significance of ECG denoising is then studied for newlinethe heartbeat classification system for Arrhythmia detection. Apart newlinefrom this, the noise robustness of various morphological and temporal newlinefeatures for the development of heartbeat classification system is also newlinestudied. newlineAt first, the efficacy of empirical mode decomposition (EMD) is utilized to overcome the rare-patch effect of the non-local means (NLM) newlinefilter. In this case, the computationally efficient modified EMD (MEMD) technique is combined with the NLM filter for properly removing both the high- and low-frequency noises from the ECG signal. The newlinehybrid NLM and M-EMD technique provides improved ECG denoising performance compared to the existing ECG denoising methods. newlineTo further enhance the ECG denoising performance, another technique based on the knowledge of non-local information present in the newlinediscrete wavelet transform (DWT) approximation coefficients is proposed. The efficacy of NLM and DWT complements each other and overcomes their individual drawbacks. This work achieves improved newlinedenoising performance over the existing and previously proposed MEMD based approach. newlineThe major issue with EMD based ECG denoising techniques is their newlinenoise sensitivity. To overcome this, a variational mode decomposition newline(VMD) based ECG denoising technique is also proposed. The significance of the aforementioned NLM and DWT based ECG denoising newlineframework is also utilized in the proposed VMD based approach.