Abstract / Summary
Abstract Background Upper-limb impairment after stroke is primarily characterized by the emergence and resolution of spastic synergies. Brunnstrom staging captures these recovery phases and provides a clinically grounded framework for studying cortical reorganization. Applying functional near-infrared spectroscopy (fNIRS) during ecological gross-motor tasks introduces task-related systemic hemodynamic artifacts that obscure focal cortical signals. This study aims to determine whether an EEG-informed multimodal framework can effectively mitigate these artifacts and extract stage-sensitive patterns of cortical reorganization. Methods After multimodal quality control, 32 participants with stroke (Brunnstrom stages III–VI) and 15 healthy controls were enrolled. EEG and fNIRS were recorded simultaneously during an auditory-cued unilateral arm-raising task with concomitant fist clenching, comprising movement initiation, sustained elevation, and rest. Our framework combined Filter Bank Common Spatial Pattern (FBCSP) with an EEG-informed General Linear Model (GLM). FBCSP extracted focal sensorimotor features, while EEG-derived motor-intention regressors improved GLM modeling of task-related hemodynamic responses. Model performance was evaluated using R ² and ΔAIC, and feature discriminability using AUC from a linear discriminant analysis (LDA) classifier. Results The proposed FBCSP-GLM framework demonstrated improved model fit and spatial discriminability compared with conventional analytical approaches. Participant-level FBCSP estimates revealed a distinct non-linear, stage-associated trajectory of cortical reorganization. Specifically, the FBCSP-derived EW-LI significantly decreased from 0.46 at Stage III to − 0.55 at Stage IV, and rebounded to 0.13 at Stage V. The target-M1 response t -score similarly dropped from − 0.06 to − 1.05 before increasing to 0.85 at Stage V (all FDR-corrected q < 0.01). The standard GLM did not capture these stage-associated patterns, as its corresponding comparisons did not survive FDR correction. Conclusions By improving hemodynamic model fit during ecological gross-motor execution, the EEG-informed multimodal framework reveals cortical reorganization patterns associated with post-stroke synergistic recovery stages. Its artifact-resilient EW-LI and target-M1 response t -score provide a robust computational assessment of neural state transitions.