Action-Perception-Driven Dual Closed-Loop Stroke Neurorehabilitation with Immersive Intervention and Pharmacotherapy Awareness

Stroke remains a leading cause of death and long-term disability worldwide, with a high incidence in East Asia. Rehabilitation is often challenged by substantial inter-individual variability, weak coordination between behavioral training and pharmacotherapy, and a lack of quantifiable neurofeedback signals for session-by-session adjustment. We propose APART (AI-powered Personalized Action Regulation Therapy), a dual closed-loop framework that integrates an Action Regulation Loop (ARL) and a Pharmacotherapy Regulation Loop (PRL). The ARL orchestrates multi-agent session configuration, immersive presentation, action-observation training, behavioral logging, and EEG-based neurofeedback. The PRL encodes pharmacotherapy timing variables—such as time since oral administration (TSOA) and postprandial state—as explicit scheduling constraints and statistical covariates, providing pharmacotherapy-aware training decision support under a clinician-defined human-in-the-loop boundary. In a two-month feasibility pilot (14 participants; 7 stroke patients, 7 healthy controls; 13 passing EEG QC), immersive presentation increased median μ-suppression in both patients (18%→24%) and controls (23%→30%). Pharmacotherapy-aware linear mixed-effects modeling further improved explanatory power and cross-subject predictive performance relative to experimental factors alone. These findings support the implementability and short-term safety of APART as a QC-gated, provenance-traceable neurorehabilitation framework. This pilot study does not claim long-term clinical efficacy but establishes a pharmacotherapy-aware, dual-loop foundation for future longitudinal and mechanistically informed stroke rehabilitation studies.