基於獨立成分分析自動去除腦波內眨眼干擾波及其應用

Automatic Removal of Eye-blink Artifacts in EEG Based on ICA and Its Applications

王璟萍(Ching-Ping Wang)

指導教授 ： 黃漢邦

國立臺灣大學/工學院/機械工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.01910

摘要

腦波訊號在量測時往往伴隨許多干擾波混雜其中，例如眼電、肌電、心電訊號等。這將嚴重影響腦機介面的性能及臨床診斷的準確率，而這些干擾波中，又以眼電訊號對腦波的影響最為嚴重，而獨立成分分析已經被證明能夠有效的將腦波及干擾波分離。本篇論文提出一個自動去除眼電訊號的方法，透過獨立成分分析將腦波及眼動分離，接著以樣本熵、碎型維度、峰度作為特徵抽取。另一方面，本研究以實際的眼動訊號做單類別分類器的訓練，並利用此訓練後的單類別分類器作為眼動成分的自動選擇器。在移除被選出的眼動成分後，將剩下的腦波成分以獨立成分分析所得到的去混合矩陣，重建去眼動的腦波訊號。根據實驗的結果，證明此方法能夠在較少的資訊損失下獲得去除眼動的腦波訊號。接著以兩個實際應用驗證其性能。第一個應用為P300 拼字機，透過本研究提出的方法將可提升拼字成功率3~9%，證實此方法能有效去除眼動訊號而保留P300訊號。第二為睡眠腦波上的應用，使用台中澄清醫院睡眠中心資料庫做睡眠腦波的分析，以獨立成分分析前處理完畢後的腦波訊號進行睡眠階段的分類，可提升睡眠階段的R1分類率平均1.56%，提升整體睡眠階段分類率0.44%。

關鍵字

腦電圖(Electroencephalograph) ； 獨立成分分析(Independent Component Analysis) ； 眨眼干擾波(eye-blink artifact) ； 單類別分類器(One-Class Classification)

並列摘要

Electroencephalogram (EEG) recordings are often contaminated by, for example, ocular artifacts, muscle artifacts, heart signals, and line noise. The influence of such contaminated EEG signals on the performance of a brain–computer interface and on clinical examination is serious. Among the types of noise, the influence of eyeblink artifacts is the most adverse. However, Independent Component Analysis (ICA) has been proven as an effective tool to separate artifacts from EEG signals. In the thesis, an approach is presented to remove eyeblink artifacts automatically. Artifacts from brain waves signals are separated by ICA, based on features of with sample entropy, fractal dimension, and kurtosis. This study sets actual eyeblinks as the training data for one-class classification, and the trained one-class classifier is used as an automatic selector for eye blink artifacts. After removing the eye-blink artifacts, the remaining brain waves use the demixing matrixes from ICA to reconstruct eyeblink-removed brain waves. According to the results of experiments, the proposed method confirms that eyeblink artifacts can be removed with less loss of information.
Two applications were implemented to verify the performance of the proposed approach. One application is the P300 speller. Through the approach of this research,
iii
the classification rate is improved by 3~9%, thus confirming that this method can remove eyeblink artifacts and retain P300 signals. The second application used to verify the results is one done on sleeping brain waves: the analysis was done with the database of the sleep center of Taichung Cheng Ching Hospital. Stages of sleep were classified by pre-procedure brain waves. Through the method, the R1 classification rate of stages of sleep is increased by 1.56%, and the classification rate of the full stage is increased by 0.44%.

並列關鍵字

Electroencephalograph ； Independent Component Analysis ； eye-blink artifact ； One-Class Classification

參考文獻

[1] L. F. Haas, "Hans Berger (1873–1941), Richard Caton (1842–1926), and electroencephalography," Journal of Neurology, Neurosurgery & Psychiatry, vol. 74, pp. 9, 2003.

[2] H. H. Jasper, "The ten-twenty electrode system of the International Federation," Electroencephalography and Clinical Neurophysiology, pp. 371-375, 1958.

[4] G. Schalk, D. J. McFarland, T. Hinterberger, N. Birbaumer, and J. R. Wolpaw, "BCI2000: a general-purpose brain-computer interface (BCI) system," IEEE Transactions on Biomedical Engineering, vol. 51, pp. 1034-1043, 2004.

[5] C. J. James and O. J. Gibson, "Temporally constrained ICA: an application to artifact rejection in electromagnetic brain signal analysis," IEEE Transactions on Biomedical Engineering, vol. 50, no.9, pp. 1108-1116, 2003.

[6] Z. Weidong, Z. Jin, Z. Hao, and J. Liu, "Removing Eye Movement and Power Line Artifacts from the EEG based on ICA," in Engineering in Medicine and Biology Society, 27th Annual International Conference of the, pp. 6017-6020, 2005.