Quantitative assessment of motor function is critical for guiding rehabilitation in stroke patients. This study presents a framework that integrates cortico-muscular coupling (CMC) and muscle synergy analysis to evaluate and predict motor impairment levels during a standardized exoskeleton-assisted knee flexion-extension task. Muscle synergies were extracted from surface EMG signals and further characterized by cosine similarity and scalar descriptors. CMC was estimated using frequency domain time evolution (FDTE), distinguishing afferent and efferent cortical-muscular pathways. Group-level FDTE analysis revealed that patients exhibited elevated bidirectional CMC compared to healthy controls, suggesting higher cortical afferent demands and enhanced feedback information caused by compensatory mechanisms in stroke patients. By integrating these features, a support vector regression (SVR) model was trained to predict individual FMA scores, achieving high accuracy (R2 = 0.9589, MSE = 1.91). These results highlight the utility of combining neural coupling and synergy structure analysis during robotic-assisted movement to support objective, data-driven functional assessment in stroke rehabilitation.