傳統中風復健因缺乏患者主動意念參與而常遇瓶頸，為此，本研究旨在開發並驗證一套創新的主動式閉迴路復健系統，其整合鏡像神經元誘發機制、腦機介面（BCI）與重複性經顱磁刺激（rTMS），核心理念在於透過BCI即時解碼患者因動作觀察而產生的運動意圖，並以此驅動rTMS進行精準神經調控，以強化神經可塑性。技術上，本研究提出一創新的深度學習模型（EEG-Spectroformer），整合時頻卷積層與多頭自注意力機制，並透過ERD量化篩選與持續學習策略優化，最終在運動意圖二分類任務上達到87.04%的高準確率，顯著優於經典模型。臨床驗證結果顯示，接受本系統治療的實驗組，其上肢運動功能（FMA-UE）平均提升4.6分（進步幅度29.11%），遠勝於僅接受傳統rTMS刺激對照組的1.66分，腦波分析亦證實本系統能有效誘發患側大腦皮質產生目標性的神經可塑性變化。本研究成功證實，此一整合患者主動意圖的主動式BCI-rTMS閉迴路系統，不僅在技術上能精準解碼腦波訊號，在臨床上更能有效地促進中風患者的上肢運動功能恢復，為未來個人化精準復健的發展提供了有力的實證基礎。;Conventional stroke rehabilitation often encounters bottlenecks due to a lack of active patient participation. To address this, this study developed and validated an innovative active closed-loop rehabilitation system that integrates a mirror neuron-induced mechanism, a brain-computer interface (BCI), and repetitive transcranial magnetic stimulation (rTMS). The core concept is to use the BCI to decode, in real-time, the motor intention generated by the patient during action observation, which in turn drives rTMS for precise neural modulation to enhance neuroplasticity. Technically, this research proposes a novel deep learning model (EEG-Spectroformer) that integrates spectro-temporal convolutional layers with a multi-head self-attention mechanism. Optimized through ERD-based quantitative selection and a continual learning strategy, the model achieved a high accuracy of 87.04% in the binary motor intention classification task, significantly outperforming classic models. Clinical validation results showed that the experimental group treated with this system had an average improvement of 4.6 points in upper extremity motor function (FMA-UE) (a 29.11% increase), far surpassing the 1.66-point improvement of the control group, which received only conventional rTMS. Brainwave analysis also confirmed that the system can effectively induce targeted neuroplastic changes in the cortex of the affected cerebral hemisphere. This study successfully demonstrates that this active BCI-rTMS closed-loop system, which integrates the patient′s active intention, not only accurately decodes brain signals technically but also effectively promotes the recovery of upper limb motor function in stroke patients clinically, providing a strong empirical basis for the future development of personalized precision rehabilitation.
