Mapping Resting-State Electroencephalography (EEG) Onto Motor Imagery for Cross-Subject Brain-Computer Interface (BCI) Applications

Brain–Computer Interfaces (BCIs), particularly those based on Motor Imagery (MI), still require extensive subject-specific training, mainly because EEG signal characteristics vary widely across individuals. This dissertation investigates whether resting-state EEG contains subject-dependent information that can reduce, or even replace, the need for per-user training for building MI decoding models. The central premise throughout this work is that the brain’s resting activity can serve as a neural signature that influences how MI activity will later manifest. The dissertation first establishes that resting-state spectral power, specifically in the Alpha band, relates systematically to how strongly the hemisphere experiencing this power activates during motor imagery. Based on this relationship, subjects were grouped into three rest-based neurophysiological profiles rather than treated as a single heterogeneous population. Across multiple datasets and model configurations, the clustering strategy consistently improved cross-subject decoding, as evidenced by better performance within clusters than in mixed clusters. This demonstrated clear potential for plug-and-play MI-BCIs. Overall, this dissertation reframes resting-state EEG as a functional and predictive signal source, showing that “rest” contains actionable structure for decoding, model transfer, and subject grouping. The work moves toward the long-standing goal of reducing training overhead in MI-BCIs and introduces a biologically grounded alternative to purely mathematical transfer-learning approaches.