Source-space EEG functional connectivity and prediction of cognition in Parkinson’s disease: No added benefit of individualized head models over standard templates

Cognitive decline is a major non-motor feature of Parkinson’s disease (PD), but reliable and accessible biomarkers remain limited. Resting-state electroencephalography (EEG) is a promising candidate because it is low-cost, portable, and well suited to repeated assessment. Recent work has increasingly focused on source-space functional connectivity (FC) for the prediction of cognition. However, the influence of source-modelling based on an individualized MRI-based head model relative to that based on standard template model is unknown. To compare these two source-space EEG FC methods, we analysed EEG data from the New Zealand Parkinson’s Progression Programme, including 136 people with PD and 51 age-similar controls. Source reconstructed resting-state EEG was parcellated with the HCP-MMP1 atlas, and used to derive amplitude envelope correlation (AEC) and debiased weighted phase lag index (dwPLI) across six canonical frequency bands. The twenty-four FC modalities were evaluated using six machine-learning regression algorithms within a nested cross-validation framework. Theta-, alpha-, and beta-band FC showed the most consistent prediction of global cognition, with the strongest performance observed for theta- and alpha-band AEC and dwPLI features (maximum R² = 0.170, r = 0.439). Standard and individualized head models showed comparable predictive performance across nearly all modalities. Feature-importance patterns for Cole-Anticevic networks were also highly similar between the two head-model options. These findings show that source-space resting-state EEG FC can predict cognitive performance in PD. The comparability of the two head models suggests that the more user-friendly and less resource intense standard head model template is satisfactory. This supports feasible, scalable, and clinically accessible EEG-based biomarkers of cognition in PD.