An ICA-based artifact suppression method for online extraction of TMS-evoked potentials: toward closed-loop TMS-EEG applications beyond the motor cortex.

Objective: Transcranial magnetic stimulation (TMS) combined with electroencephalography (EEG) has become a valuable tool in clinical and cognitive neuroscience. However, TMS-EEG signals often suffer from severe artifacts, particularly in lateral cortical regions where TMS-evoked muscle arti-facts are pronounced, making real-time recovery of TMS-evoked potentials (TEPs) challenging. We developed and validated a real-time, two-step independent component analysis (ICA)-based artifact cleaning method for TMS-EEG signals, facilitating the rapid extraction of clean neural signals for closed-loop neurostimulation applications.

Approach: Our method involves an offline ICA training phase, where ICA weights and artifact topographies are identified using pre-experimental trials, fol-lowed by an online phase in which the precomputed weight matrices are applied in real-time to in-coming data. We conducted simulations on two pre-published TMS-EEG datasets (N = 28, ROIs = 6) to validate the method by identifying the minimum number of trials required to estimate ICA weights. We also assessed the reproducibility of TEPs and the stability of ICA components, taking classical offline TEPs as the relative ground truth.

Main Results: ICA analysis suggests that it can be applied reliably within each region without significant loss of convergence and stability, provided careful consideration is given to the size and composition of the data used for ICA training. Simula-tion results indicated that while central regions could achieve reliable TEPs similar to ground truth with as few as 20-30 trials to train ICA in the pre-experimental phase, frontal and occipital regions required 50-60 trials to reach a comparable level of reliability. Later TEP peaks (>100 ms) in all re-gions achieved high reproducibility when at least 35 training trials were used, whereas earlier peaks (>20 ms) showed moderate reproducibility with the same number of trials.

Significance: These find-ings establish the feasibility and proof-of-concept for real-time ICA-based artifact removal for closed-loop TMS-EEG applications. The method enables rapid extraction of clean neural signals, allowing adaptation of stimulation parameters in real time, thereby facilitating individualized neurostimula-tion paradigms. .