A Fuzzy Inference System and Stationary Wavelet Decomposition for Identification and Removal of ECG Artifact from sEMG Signals

Surface electromyography signal (sEMG) is one of the most widely used biomedical signals in various fields such as rehabilitation, control, medical diagnosis, sport, etc. However, the quality of sEMG signals may be affected by physiological and non-physiological contaminants. ECG (electrocardiogram) is one of the most common artifacts contaminating the sEMG recordings. Therefore, removing the ECG artifact from the sEMG signal is of great interest. This paper proposes an automated approach to detect and remove the ECG artifact from the sEMG signals. Three spectral statistical features: compsite multiscale entropy (CMSE), kurtosis (K), and skewness (S) are extracted from the power spectral density (PSD) of the sEMG signal. Based on these features, a fuzzy inference system (FIS) detects if a window of the sEMG is affected by ECG or not. Then, the artifactual windows are decomposed via stationary wavelet transform (SWT) into SWT coefficients. Since the ECG artifact is located only in the low frequencies, approx. coefficient is selected to be filtered using a thresholding method, whereas the detail coefficients are preserved. This method is evaluated using several metrics and is compared with one of the most used methods for removing artifacts from sEMG. The results show the superiority of our method at all considered SNR (signal to noise ratio) levels.