經驗模態分解法之清醒與麻醉情形下的腦波特徵判別

賴穎賢(Ying-sian Lai)

指導教授 ： 黃衍任 ； 陳世叡

國立中央大學/工學院/機械工程學系/碩士(2014年)

摘要

本論文使用經驗模態分解法(EMD)與Matlab軟體進行病人腦電圖(EEG)分析，並且運用總體模態經驗分解法(EEMD)消除腦波量測時所受到的雜訊干擾，再搭配快速傅立葉轉換(FFT)，探討各個本質模態函數(IMF)的傅立葉頻譜圖頻率差異，找出清醒與麻醉病人的腦波特徵。
傅立葉頻譜圖的特徵頻率擷取可分為「最大振幅之頻率」和「期望值」兩大部分。尋找頻譜圖中最大振幅所對應的頻率，即為「最大振幅之頻率」，此外，本研究也嘗試利用移動平均來消除雜訊干擾，並且根據分析數值比較清醒與麻醉的腦波特性和EEMD濾波效果。「期望值」的部分則藉由計算傅立葉頻譜圖的收斂期望值作為腦波特徵，再繪製每個IMF的頻率分布機率圖與接收者操作特徵曲線，透過研究曲線圖特性找出能判斷清醒與麻醉特徵的IMF，其中IMF1的辨別準確性達到99%。
最後，分別介紹兩種特徵頻率的實驗結果，並比較兩種特徵頻率的優缺點和討論EMD與EEMD的濾波效果。

關鍵字

經驗模態分解法 ； 總體經驗模態分解法 ； 接收者操作特徵曲線 ； 快速傅立葉轉換

並列摘要

In this research, empirical mode decomposition (EMD) and software, such as Matlab, is used for the analysis of the patients’ electroencephalograms. In order to wipe out the disturbance caused by noise, ensemble empirical mode decomposition (EEMD) is also manipulated in the investigation. Then with the assistance of Fast Fourier Transform, the characteristics of the patients’brain wave in consciousness and anesthesia can be discovered by discussing the difference of the Fourier spectrum of each intrinsic mode function.
The extraction of the characteristic frequency of the Fourier spectrum is divided into two sections, including the frequency of the maximum amplitude and the expected value. The frequency corresponding with the maximum amplitude of the spectrum is the frequency of the maximum amplitude. Besides, moving average is also tried to delete the disturbance caused by noise in the research. According to the analyzed data, the characteristics of the patients’brain wave in consciousness and anesthesia and the filtering effect of EEMD can be compared. In the other section, the convergence of the expected value is calculated and regarded as the characteristics of the brain waves. Subsequently, the possibility graph of the frequency distribution and receiver operating characteristic curve of each IMF are plotted. The IMFs used to identify the characteristics of consciousness and anesthesia can be revealed by exploring features of these graphs. And the identify accuracy of IMF1 is 99 percent.
Finally, the results of these two kinds of characteristic frequency are stated and compared. The filtering effects of EMD and EEMD are also discussed.

參考文獻

[1] 范守仁，「人工智慧與自動化工程在臨床醫學的應用-麻醉期間病人的

pp.903-995, The Royal Society, 1998.

[2] Ernst Niedermeyer, M.D., & Fernando Lopes Da Silva, M.D.,

Electroencephalography: Basic principles, Clinical Application, and Related Fields, Fourth Edition, Lippincott Williams & Wilkins, 1999.

[3] Mark H. Libenson,MD, Practical Approach to Electroencegraphalography,