The accuracy of EEG source localization depends on electrode data quality; however, electrode loss is common in brain stimulation paradigms, and its impact on localization accuracy remains unclear. This study evaluated depth-dependent modulation of source localization accuracy and compared electrode processing strategies across multiple inverse methods. The Localize-MI dataset was used, comprising simultaneous stereoelectroencephalography (sEEG) and high-density EEG (256 channels) from seven epilepsy patients, where single-pulse electrical stimulation through implanted electrodes provided ground-truth source locations. Eight inverse methods covering distributed (WMN, dSPM, sLORETA, eLORETA), sparse (MxNE, Champagne), and parametric (dipole fitting (DF), RAP-MUSIC) approaches were compared under five scenarios: baseline, proximal electrode removal or interpolation, and distal electrode removal or interpolation. Peak localization error (PLE) and spatial dispersion (SD) were analyzed using linear mixed-effects models incorporating source depth, electrode-to-source distance, and their interaction. Proximal electrode degradation significantly increased both PLE and SD, whereas distal manipulations had negligible effects. DF and RAP-MUSIC showed significantly lower PLE than WMN and, together with MxNE, demonstrated the strongest resilience to proximal degradation. Among distributed methods, dSPM maintained the lowest and most stable SD across scenarios. Source depth was positively associated with both PLE and SD, and this association was somewhat strengthened when proximal electrode data were compromised. These findings indicate that electrode-to-source proximity is the critical determinant of degradation impact, and that sparse and parametric methods offer superior resilience to localized electrode loss, providing practical implications for electrode management and inverse method selection in NIBS-EEG protocols where near-target sensors are obstructed by stimulation hardware.
Qin et al. (Tue,) studied this question.