時序特徵分類與投票之三選項腦機界面系統

Abstract

腦機界面系統(BCI)提供一個單靠腦部活動來取代平常使用的肌肉做為和外界溝通的管道。而腦電波圖(EEG)廣泛的應用在腦機界面系統中。透過記錄和分析使用者執行特定任務時的腦部活動，可以將分析過後的結果轉換成對應的指令而達到像是控制義肢或指標、打字或回答問題。時至今日，像是運動想像腦機界面或事件相關腦機界面等許多腦機界面系統蓬勃發展。在現存的腦機界面研究中，以P300為基礎的腦機界面因穩定且不需預先訓練受試者的優點，因而被廣泛研究。 為了達到降低訓練時間和降低使用者的負擔，我們使用一個三選項的界面實作以P300為基礎腦機系統。我們並使用一個經典的P300分析方法步進線性鑑別分析(Stepwise linear discriminant analysis)做為特徵選取和分類。並提出一投票策略來達到系統可自動且即時的依使用者做調整，以增進線上系統的效能。更具體來說，我們結合步進線性鑑別分析和活動窗口(moving window)來產生時序特徵做投票。透過這些時序特徵我們可以決定門檻值使得線上系統可以在維持一定的分辨率的條件下動態決定結果。因此，我們可以增進線上系統溝通的效能和效率。 三個健康的受試者被邀請參與離線和線上的實驗。研究結果顯示，我們的系統比Sellers和Donchin在2006所發表的四選項腦機系統有更好的效能。在離線分析我們在83.4%的分辨率下達到7.7 bits/min的轉移率，而當線上系統達到5.23 bit/min 的轉移率時則有100%的分辨率。這些結果都證明了比四選項系統在97%的分辨率下達到1.8 bit/min有更好的效能，也顯示了適應性在腦機界面系統上的優點。

Brain-computer interface (BCI) provides a channel to communicate with external world only through cerebral activity, thus replacing the normal pathway of communication by using muscles. The electroencephalography (EEG) is commonly used in the BCI system. When a subject is performing specific tasks, the EEG signals induced by the subject’s neuronal activities are recorded and analyzed. Then, the analyzed EEG signals will be translated to the corresponding commands to control prosthesis or cursor, spell words, answer questions. Nowadays, there are many development of BCI systems such as Motor-imagery based BCI systems, ERP-based BCI systems. In the existing BCI studies, P300-based BCI systems are commonly conducted, because P300 ERP can be reliably measured without initial user training. To reduce training time and subjects’ burden, our P300-based BCI system is implemented by using a three-choice paradigm. We use a typical P300 analysis method, stepwise linear discriminant analysis (SWDA) for feautre selection and classification. To improve the system, we propose a ”voting strategy” to automatically make the online system adaptable to users. More specifically, we combine SWDA with moving window to produce the temporal features for voting. Through the temporal features which can decide the threshold, the online system can dynamically make decision while maintaining the accuracy of classification. In this way, we improve the performance and efficiency of communication in online BCI system. Three able-body subjects are recruited to participate in the offline experiments, and seven able-body subjects are recruited to participate in online experiments. The results of this study present better performances than those in the four-choice offline system provided by Sellers and Donchin (2006). In offline analysis, the transfer rates can be achieved up to 7.7 bits/min with an accuracy of 83.4%, while the transfer rates of online testing can be achieved up to 5.28 bits/min with an accuracy of 100%. These results suggest that the performances of our system are better than four-choice system in which transfer rate is 1.8 bits/min with accuracy of 97%, thus indicating the advantage of the adaptability in BCI system.