用於中風復健之即時視覺回饋腦機介面開發

Abstract

基於腦電波(EEG)訊號的腦機介面(BCI)系統可以透過運動想像的方式，幫助患者訓練受損的功能，並恢復受損的運動腦區。但對於中風病人而言，不容易做到運動想像，如何讓病人在訓練階段與試驗階段都能提升想像的特徵更加關鍵。本論文建立一個即時腦機介面系統，透過遊戲提供即時視覺回饋的方式在病人跟腦機介面之間形成封閉的迴路，以增進復健的效果。 本論文採用透過多頻帶濾波共同空間模式（Filter-Bank Common Spatial Pattern, FBCSP）來做特徵擷取方法，以及線性判別分析（Linear Discriminate Analysis, LDA）做為分類器。為了應用在即時分析，我們利用離線資料來驗證即時分析模型。並改善了濾波器組，用無限脈衝響應濾波器(IIR filter)取代原本離線分析使用的快速傅立葉轉換濾波器(FFT filter)，以降低運算複雜度。最後利用RISC-V提供的仿真器計算出每筆分類結果只需要11ms就能輸出。 即時視覺回饋腦機介面實驗共收錄6名不同程度的中風病患，實驗結果顯示在扣除一名無法完成指定動作想像指令的病患，在實際動作平均準確度達84.93%，在想像動作平均準確度達80.01%。實驗驗證出有視覺回饋的遊戲介面可提升約6.56%的分類準確度。

The electroencephalographic (EEG) signals based brain-computer interface (BCI) system help stroke rehabilitation on the motor function of the brain by performing Motor imagery (MI) or related tasks. But for stroke patient who isn’t easy to image movement, how to highline imagery feature in training and testing is more important. For enhancing performance in rehabilitation, this thesis proposes a real time BCI system with visual feedback which forms a close loop between BCI and subject. The proposed approach adopts the filter bank common spatial pattern (FBCSP) method for feature extraction and linear discriminate analysis for classification. To reduce complexity, we replace the Fast Fourier Transform filter (FFT filter) to Infinite Impulse Response filter (IIR filter). Finally, the whole algorithm outputs one classification under RISC-V emulator only needs 11 ms. The experiment invites the stroke patients who are in different stage of stroke rehabilitation. The results show 84.93% accuracy for motor execution and 80.01% accuracy for motor imagery. It is verified that our BCI system enhances about 6.56% classification accuracy by playing visual-feedback game.