摘要: | 腦電波(electroencephalography, EEG)量測大腦皮質的電流,其優勢為前置作業簡單、可攜式、低成本和高時間解析度。然而,腦電波很容易受到雜訊的影響,包括自發性腦波律動、生理雜訊和外部干擾。腦電波也會隨著使用者的生理狀態而改變,導致大腦人機介面(brain computer interface, BCI)必須經常校正以達成高準確率。因此,萃取事件相關的訊號特徵與腦波的自適性校正為腦波訊號處理上相當重要的議題。為了解決這些問題,本研究發展整體經驗模態分解(ensemble empirical mode decomposition, EEMD)為基礎的方法提取出腦波訊號,以提高視覺腦波訊號的訊雜比,並研發雙相位刺激與步進延遲閃爍序列實現自動校正穩態視覺誘發電位(steady-state visual evoked potential, SSVEP)為基礎的大腦人機介面。整體經驗模態分解法不需要預設訊號基底,具有分析非平穩時間序列的優點,可以適用於穩態視覺誘發電位的萃取應用。整體經驗模態分解法將視覺腦波訊號分解為許多的內建震盪函數,視覺腦波可以藉由選擇適當的內建震盪函數(intrinsic oscillatory function, IOF)進行重建,濾除不必要的雜訊。在雙相位刺激技術部分,每一段閃光被分為參考相位與偏移相位段落,每一個閃爍光源使用不同的偏移相位進行編碼,因此使用者的腦波可以根據參考相位與偏移相位的相位差進行校正。為了進一步提升自動校正穩態視覺誘發大腦人機介面的效率,我們進一步開發步進延遲閃爍序列(stepping delay flickering sequence, SDFS)技術,藉由給予每一個閃光選項不同的步進時間延遲,產生不同選項閃光序列之間的獨立性,使得受試者所注視的閃光選項可以藉由簡單的平均技術就能辨識出來。本研究成果,在離線與線上實驗都有高的準確率與資訊傳輸率(information transfer rate, ITR),所完成的自動校正穩態視覺誘發大腦人機介面,目前已經可以達成97.38?5.97%的準確率、命令傳輸間隔(command transfer interval, CTI)為3.56?0.68秒的平均指令時間、以及每分鐘42.46?11.47位元的資訊傳輸率。本論文開發全世界第一套具有自動校正功能的穩態視覺誘發大腦人機介面,論文的研究成果將有助於幫助漸凍人、肢體殘障病人、腦創傷病人、癱瘓病人與外界的溝通控制應用。Electroencephalography (EEG) noninvasively measures neural electrophysiological activities in human brain. It has the advantages of easy preparation, portability, low cost, and high temporal resolution. However, EEG is highly susceptible to noise, including spontaneous brain rhythm, electrophysiological artifacts and external interference. The EEG signals also can be affected by the change of subject’s physiological states. Both of these reasons causes a brain computer interface (BCI) usually has to be calibrated after using a certain time period to make sure its high accuracy. Therefore, an effective method for brain signal extraction and a self-calibrated procedure are two crucial issues to develop a practical BCI. In this dissertation, we adopted ensemble empirical mode decomposition (EEMD) to extract steady-state visual evoked potential (SSVEP). The EEMD doesn’t require pre-defined basis or statistical assumption for signal extraction which is beneficial to extract non-stationary signals. EEG signals recorded from Oz position were decomposed by EEMD into a series of intrinsic oscillatory functions (IOF). The SSVEPs were then reconstructed by selecting pertinent IOFs for noise suppression. Regarding the development of self-calibrated SSVEP-based BCI, two techniques, one is biphasic stimulation and the other is stepping delay flickering sequence (SDFS), were developed. The biphasic stimulation technique divides flickering sequences into reference phase segments and shift phase segments. By designating predefined phases to reference phase segments and shift phase segments, the phase difference between reference phase segment and phase shift segment was calculated. Since the designed phase difference between the two segments was known, subject’s SSVEP phase drifts can be corrected by comparing the measured phase difference and the designed phase difference. The second self-calibrated SSVEP-based BCI was implemented by SDFS method. By designing different SDFS with distinct stepping delay, the independency among distinct flickering sequences was achieved which resulted in subject’s gazed target can be recognized through a simple averaging process. The experimental results of this dissertation have shown high accuracy and high information transfer rate (ITR) in both off-line and on-line experiments. The averaged accuracy, command transfer interval (CTI) and ITR are 97.38?5.97%, 3.56?0.68 s and 42.46?11.47 bits/min, respectively. The study result of this dissertation has achieved the first self-calibrated SSVEP-based BCI in the world. The achievement provides an effective way for amyotrophic lateral sclerosis (ALS), brain trauma and paralyzed patients to communicate with external environments. |