Removal of High-Voltage Brain Stimulation Artifacts From Simultaneous EEG Recordings

Objective: noninvasive electrical brain stimulation (NEBS) can modulate brain dynamics and be used to induce changes in target brain areas. Because high-voltage stimulation artifacts tend to completely dominate the simultaneously recorded EEG and vary slightly trial by trial in both amplitude and phase, the majority of studies to date have simply compared poststimulation and prestimulation periods. Here, we propose the use of a modified joint blind source separation (JBSS) approach for removing stimulation artifacts when the same stimulus is applied in multiple epochs. Methods: quadrature regression and subsequent independent vector analysis (q-IVA) was applied to simulated and real EEG datasets recorded from ten subjects who received theta-band galvanic vestibular stimulation. Results: in simulations, q-IVA and JBSS approaches significantly improved the relative root-mean-squared error, correlation coefficient, and power deviation between the original and recovered EEG compared to conventional methods. In the real EEG data, after removing the artifacts with q-IVA, the power spectra during stimulation demonstrated significantly enhanced beta and gamma power compared to prestimulation, becoming similar to that seen in immediate poststimulation periods. In addition, we demonstrate that increased alpha power (8-13 Hz) in occipital regions with eye closure could be reliably detected in the cleaned EEG data after applying q-IVA. Conclusion: q-IVA and JBSS approaches outperform conventional artifact removal methods in both time and frequency domains. Significance: our results provide a promising way to effectively isolate stimulation artifacts in EEG, paving the way for future studies attempting to uncover ongoing modulation of brain activity during NEBS.