用於中風復健之線上適應性腦機介面開發

Online Adaptive Brain–Computer Interface for Stroke Rehabilitation

王韻綺(Yun-Chi Wang)

指導教授 ： 張添烜

國立交通大學/電機學院/電子工程系所/碩士(2015年)

https://doi.org/10.6842/NCTU.2015.00907

摘要

基於腦電波(EEG)訊號的腦機介面(BCI)系統可用來幫助中風病人復健， 但腦電波訊號本身有不穩定的問題，會降低其應用在腦機介面上的效果。因此，本論文提出了適用於中風復健的線上適應性腦機介面，去解決這樣的問題。
本論文所提出的適應性方法，是用輸入訊號去對特徵擷取器與分類器做全部的更新，而不像在傳統上是用舊的結果和少量的新進的資料只針對特徵擷取器或分類器其中的一種做部分的更新。與以往的方法相比，這樣的方法讓準確度最多可以達到13%的提升。除此之外，我們的方法只需要1.5到4分鐘的時間去訓練初始模型，比起以前需要20分鐘左右的方法，明顯地降低了線上適應性腦機介面訓練模型的初始時間。
本論文最後以24個試驗的窗口大小和每20秒更新一次的速度，讓在線適應性腦機介面對中風病人的資料能達到81.77%的準確度，而且這樣的準確度比非適應性腦機介面更高出了6.7%。

關鍵字

腦機介面 ； 腦電波 ； 線上 ； 適應性 ； 中風復健

並列摘要

The electroencephalographic (EEG) signals based brain-computer interface (BCI) system help stroke rehabilitation but face the signal nonstationary problem and results in lower effectiveness. To solve this problem, this thesis proposes an online adaptive BCI interface for stroke rehabilitation.
The proposed approach adopts the full update of the feature extraction and classification from input data instead of the previous leaky update of either feature extraction or classification with old results and small amount of new input data. Our approach can improve accuracy up to 13% when compared to the previous method. This method enables significantly lower initial training time to 1.5 to 4 minutes for online adaptive BCI instead of 20 minutes in the previous approach. The final online adaptive BCI simulation can attain 81.77% accuracy in average for stroke patients with 24 trials window size and 20 second update rate, which is 6.7% better than that in non-adaptive online BCI.

並列關鍵字

BCI ； EEG ； online ； adaptive ； stroke rehabilitation

參考文獻

[1] K. K. Ang and C. Guan, “Brain-computer interface in stroke rehabilitation,” J. Computing Science and Engineering, vol. 7, no. 2, pp. 139–146, 2013.

[2] S. R. Soekadar, “Brain computer interfaces in the rehabilitation of stroke and neurotrauma,” Syst. Neurosci. Rehabil., pp.3 -18, 2011.

[3] P. Shenoy, et al. “Towards adaptive classification for BCI,” J. Neural Eng., vol. 3, no. 1, pp.R13 -R23, 2006.

[5] K. K. Ang, et al. “Filter Bank Common Spatial Pattern (FBCSP) algorithm using online adaptive and semi-supervised learning,” in Proceeding of International Joint Conference on Neural Networks (IJCNN), pp. 392-396, 2011.

[6] S. Darvishi, et al. “Investigation of the trade-off between time window length, classifier update rate and classification accuracy for restorative brain-computer interfaces,” in Engineering in Medicine and Biology Society (EMBC), 2013 35th Annual International Conference of the IEEE. pp. 1567-1570, Jul. 2013.