Recovering Resting-State EEG During AM-tACS Using Dynamic Adaptive Stimulation Artifact Source Separation

Abstract Objective. Transcranial alternating current stimulation (tACS) can modulate brain oscillations in a frequency-specific manner, but strong stimulation artifacts severely contaminate concurrent electroencephalography (EEG), limiting direct assessment of ongoing neural activity. Conventional artifact suppression methods often lack adaptability to nonstationary and frequency-dependent artifacts, which may cause incomplete artifact removal and neural signal distortion. This study aimed to develop a dynamic adaptive stimulation artifact source separation framework to improve EEG recovery during amplitude-modulated tACS (AM-tACS). Approach. We proposed a dynamic adaptive stimulation artifact source separation framework, termed DASASS, integrating recursive covariance updating, adaptive component selection, and wavelet-based multiband processing. This framework was designed to suppress time-varying and frequency-dependent stimulation artifacts while preserving the spectrotemporal and spatial characteristics of neural signals. The method was evaluated on resting-state EEG recorded during AM-tACS and compared with stimulation artifact source separation (SASS), principal component analysis (PCA), independent component analysis (ICA), and adaptive noise cancellation (ANC). Performance was assessed using normalized root mean square error, band-power deviation, and demodulated residual modulation. Microstate analysis examined whether recovered EEG preserved canonical spatial topographies and temporal dynamics. Main results. DASASS achieved the lowest NRMSE, shift-averaged NRMSE, and demodulated residual modulation among the tested methods, indicating improved waveform reconstruction and effective suppression of residual modulation artifacts, while its small negative band-power deviation suggested mild conservative attenuation rather than superior spectral preservation. Further microstate analysis showed that EEG recovered by DASASS preserved the four canonical microstate topographies, with spatial correlation coefficients ranging from 0.8643 to 0.9412 relative to the artifact-free reference EEG, while maintaining overall microstate temporal dynamics. Significance. These findings indicate that DASASS provides an effective solution for EEG recovery during AM-tACS. The proposed framework may support stimulation-state brain activity analysis and future closed-loop neuromodulation applications.