GVICA: A Multi-Channel EEG Hierarchical Noise Reduction Framework Based on GWO Dynamically Optimized VMD-ICA Fusion

Electroencephalogram (EEG) signal acquisition is prone to artifact interference, which affects the interpretation of neural activity. This paper aims to develop a new method that can automatically and effectively remove ocular and electromyographic artifacts from multi-channel EEG, while maximally preserving neurophysiological information.

We propose a multi-level denoising method called GVICA. This method is implemented by integrating optimized Variational Mode Decomposition (VMD) and Independent Component Analysis (ICA). The signals were first grouped by noise level. The Grey Wolf Optimizer (GWO) was then employed to determine the optimal VMD parameters adaptively. Anomaly detection and correction were performed on the obtained intrinsic mode functions (IMFs). Finally, an ICA-based entropy denoising module was combined to objectively select sub-signals in the frequency domain for accurate artifact removal.

Evaluations on both simulated and real EEG data show that GVICA can effectively remove artifacts, achieving lower root mean square error (RMSE) and higher Pearson correlation coefficient (PCC) under different signal-to-noise ratios (SNR). Compared with existing methods, GVICA can simultaneously and automatically remove EMG and EOG artifacts and more accurately restore brain responses.

The GVICA method significantly improves the quality of multi-channel EEG signals through its group-based multi-level and adaptive denoising framework, achieving artifact removal with minimal loss of neurophysiological information.

This study provides a reliable tool for cleaning electroencephalogram (EEG) data in clinical settings and demonstrates significant potential for application in the analysis of motor execution EEG during stroke rehabilitation.