應用於神經復健最佳動作想像通道之腦機介面開發

Abstract

儘管在醫藥發達的現代，腦中風與其它腦傷造成的癱瘓與不便仍須依賴復健來治療，結合腦電波的自發性神經復健為一個主動而有效率的復健方式。然而此種自發性神經復健的關鍵在於準確的判斷使用者的動作想像(motor imagery, MI)，傳統上要提高判斷的準確率必須配帶多通道腦波電極帽，而其麻煩的事前準備工作卻大大降低了復健系統的實用性。本論文透過一系列機器學習演算法，包括頻率分解、共同空間模式(Common Spatial Pattern, CSP)、分類器來提高判斷準確率。並透過預先收取使用者整頭腦波、分析找出最適合的電極擺放位置，將電極通道數目減少到兩個，在實際操作時即可達到只用兩通道卻有高判斷準確率效果。有此操作流程後，本論文更將上述之神經復健系統之軟硬體實現出來並實際操作驗證之，結果不管是事前分析還是實際的即時操作，所有受測者平均皆有七成以上的判斷準確率，不僅提供使用者良好的控制感，又省去了使用前煩瑣的準備工作。透過此神經復健系統，期能大大改善腦傷患者的生活。

Although the well developed in medical area, rehabilitation is still the main therapy for the patient who suffered from paralysis caused by stroke and other brain diseases. The spontaneous neuro-rehabilitation which combines with electroencephalogram (EEG) is an active and efficient way to rehabilitation. However judging the subject’s brain state of motor imagery precisely is the key factor of this spontaneous neuro-rehabilitation. To judge the brain state more accurately, we need to take multi-channel EEG cap in tradition. But the time consuming and troublesome channel preparation makes the rehabilitation system unpractical. This research enhances the judgment accuracy by a series of machine learning algorithm like frequency decomposition, Common Spatial Pattern (CSP) and classifier. We reduce the channel number to two and obtain the optimal channel position by pre-collecting and analyzing the subjects’ whole scalp EEG. So we can have a high performance only with two channel when practical operating. After establishing the protocol, this research implements the hardware and software and test the system. The average accuracy of the judgments is over 70% no matter in the off-line analysis or the on-line practical operation. Through this rehabilitation system, we hope the patients’ life can be changed heavily.