A hybrid framework for muscle artifact removal in EEG: combining variational mode decomposition, stationary wavelet transform, and canonical correlation analysis

Electroencephalography (EEG) analysis is critical for diagnosing various neurological disorders and for other brain-related control applications. However, the presence of artifacts, including muscle artifacts, significantly alter signal power and lead to misinterpretation of neural information. Hence, it is essential to identify and remove these signals to enable a reliable EEG interpretation. By addressing the limitations of existing signal decomposition methods, this study develops a hybrid approach combining variational mode decomposition (VMD), stationary wavelet transform (SWT), and canonical correlation analysis (CCA) to enhance EEG signal quality by precisely aiming to remove muscle artifacts. The study utilized EEGdenoiseNet, an open-source dataset, to test the efficacy of the proposed method. The performance metrics employed in this study include: signal-to-noise ratio (SNR), correlation coefficient (CC), root mean square error (RMSE), and power spectral density (PSD). The findings indicate that the proposed approach yields superior results, with an average SNR improvement of 4.1360 dB and an average CC of 0.8171 compared with the combination of the VMD-CCA, EMD-CCA, EEMD-CCA, and EWT methods. The lower RMSE values and PSD plots further demonstrate the effectiveness of the method in muscle artifact suppression while retaining the relevant EEG characteristics.