Wavelet Analysis-Based Reconstruction for sEMG Signal Denoising

Abstract

Surface electromyography (sEMG) recordings provide a safe, easy, and non-invasive method, allowing objective quantification of the electric activity of muscles. Analysis of sEMG plays an important diagnostic role in assessing muscle disorders. Typically, sEMG is a non-stationary signal contaminated by various noises or artifacts that originate at the skin-electrode interface, in the electronics, and in external sources. Thus, appropriate filtering procedures have to be applied to make sEMG clinically usable, in order to extract the main sEMG features. In the recent literatures, among the best performing denoising methods, Wavelet transformation (WT) denoising has been proposed. In particular, aim of this study is to propose a new denoising method based on WT multi-level decomposition analysis. To this aim, Daubechies mother wavelet (4^th order, 9 levels of decomposition) was applied to 5 real sEMG tracings. Tibialis anterior (TA) and gastrocnemius lateralis (GL) signals are considered. This method focusses on the choice of a new thresholding rule for sEMG reconstruction and denoising. Performances of this method are computed against soft-thresholding denoising technique (ST) in terms of Root Mean Square Error (RMSE). After application of WT multi-level denoising technique, signal-to-noise ratio (SNR) increased significantly (TA: 14.5 ± 6.9 vs. 19.5 ± 7.1; GL: 14.0 ± 5.4 vs. 18.7 ± 6.3). Moreover, WT multi-level denoising technique showed a lower dispersion than ST (RMSE for TA: 0.8 vs. 1.2; RMSE for GL: 0.9 vs. 1.1.), introduced no sEMG signal delay. Thus, this method is a novel and efficient tool for sEMG denoising, that could be used to make easier the detection of sEMG activation onset-offset.