An enhanced model-based EEG source imaging method captures dynamic changes during motor skill learning
Abstract
Understanding brain reorganization during motor learning requires methods with both high temporal and spatial resolution. We developed an enhanced model-based approach to solve the EEG inverse problem that combines a multi-layer perceptron neural network with finite element modeling for subject-specific head conductivity estimation. The framework resolves up to six simultaneously active sources across 116 cortical and subcortical regions and achieves 85.98% source localization accuracy, outperforming existing methods (67.1% accuracy,1 even under varying noise conditions. A dedicated leakage analysis using single- and multi-region simulations confirmed minimal cross-talk between cortical motor and deep subcortical/cerebellar sources, supporting reliable separation of activity across brain depths. As a proof-of-concept application, we used this framework to analyze EEG data from five healthy adults performing a novel foot-drawing motor learning task across multiple sessions, with behavioral performance improving after the fourth session. Model-based source estimates revealed a broad reorganization of cortical and subcortical activity across sessions, consistent with functional refinement of motor networks with practice; given the small cohort (n = 5), these findings are preliminary and offered as suggestive of learning-related dynamics rather than definitive evidence. This enhanced model-based EEG source localization approach offers a practical, non-invasive method for combining computational modeling with neuroimaging of motor behavior.