ECG signal denoising by fractional wavelet transform thresholding

Abstract

The analysis of electrocardiogram (ECG) signals allows experts to diagnose several cardiac disorders. However, the accuracy of such diagnosis depends heavily on the signal quality. In this paper, an efficient method based on fractional wavelet decomposition coupled with thresholding techniques is proposed for noise removal. The usual low-pass and high-pass filters of the wavelet transform are replaced by fractional-order ones. Thus, fractional wavelets are proposed, simulated, and compared to other wavelets for ECG denoising. The denoising process was made operational by the means of an appropriate choice of the wavelet transform coefficient thresholding and the wavelet decomposition level of the signal. Considering the relative error metrics, the best wavelet function for efficient denoising is the fractional one. In our study, we have used eight real ECG signals from the Physionet MITBIH. In order to prove the effectiveness of our method, we investigated the filtering of two types of noises, namely Gaussian white noise and power-line interference (PLI) noise. The proposed method removed the Gaussian white noise completely and had better performance on the PLI noise. Considering classical metrics of assessment, results show the advantage of the proposed method compared to other types of wavelets. The proposed method is the most suitable one for removing PLI and Gaussian white noise from ECG signals with superior performance than other wavelets. Also, it can be applied for high-frequency denoising even without a priori frequency knowledge.