Performance Analysis of Advanced Filtering Techniques for Removal of EEG Muscle Artifacts

The ‘electroencephalogram” (EEG) has become an essential technique for recording cerebral activity. EEG is extensively used in the field of clinical applications and BCI systems. However, these recorded brain signals are vulnerable to potential artifacts due to internal (physiological) and external factors. Though it is possible to limit the external factors affecting the actual brain signal, the internal factors are unavoidable. In the proposed work, two filtering models based on the normalized least mean square (NLMS) algorithm and Kalman filtering (KF) are applied to effectively remove the identified muscle artifact segments. The efficient stationary wavelet transform (SWT) was employed to identify the muscle artifact segments in the contaminated EEG signals. To test the models, EEGdenoiseNet, an open source dataset, was utilized. By varying the signal to noise ratio within the range of [−72] dB, ten contaminated EEG signals were generated. The results obtained show that the Kalman filter was more efficient in removing the high-frequency muscle artifact compared to the normalized LMS-based adaptive filter. The average correlation coefficient (CC) computed for Kalman filter was 0.82786, 2.7 % better than the normalized LMS-based filter. Results show that the output of the Kalman filter was more suitable to clean EEG signal with a 10.20 dB improvement in average SNR.