基於小波轉換之腦電訊號分析與長期多項生理訊號自動分類系統

Wavelet-Based EEG Analysis and Automatic Classification System of Long-Term Polysomnography

趙志峰(Chih-Feng Chao)

指導教授 ： 江昭皚

國立臺灣大學/生物資源暨農學院/生物產業機電工程學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.00028

摘要

本研究針對腦電訊號與多項長期生理訊號資料，發展有效率且精確的分析方法，可提供臨床診斷與相關研究之輔助與應用。研究方法以小波轉換(wavelet transform)理論為基礎，結合數位訊號處理方法、非線性能量運算(non-linear energy operator)訊號分段法、Fuzzy C-Means相關分群法以及貝氏分類器(Bayesian classifier)等，提出了新發展之小波基底選用準則以及相關應用演算法。其中包括了小波轉換應用於「腦電訊號動態節律之擷取」：藉由擷取出腦電訊號中α、β與慢波(slow wave)節律，排除多餘的雜訊或生理訊號干擾，讓醫療人員能直接觀察出腦電訊號動態之變化；「偵測與消除腦電訊號內之心電訊號干擾演算法」：針對麻省理工學院與貝瑟醫院所提供之腦電訊號資料庫，實驗結果偵測率達93%以上，並能從腦電訊號中消除心電訊號干擾；以及新發展之「結合小波轉換與貝氏分類器之睡眠紡錘波自動偵測與辨識演算法」，藉由自行設計「Spindlet」小波基底作為訊號偵測與特徵擷取工具，結合貝氏分類器之辨識，實驗結果偵測率達87.97%。除了改善過去文獻所提出之方法外，透過這些演算法之設計與應用，亦簡化原本大量且複雜的腦電訊號並達到自動化分析之目的。
此外，本研究利用上述小波轉換等相關理論與研究基礎，所發展之長期多項生理訊號自動分類系統(Automatic Classification System of long-term Polysomnography, ACSP)，將臨床個案所量測長達數個小時之大量腦電訊號、眼電訊號以及肌電訊號經訊號分段、特徵值擷取後，以自組氏分群演算法進行整合與分類，將具有相似特徵與型態之生理訊號分類成群，並且提供分類後各個群組之代表樣本訊號與生理訊號分類時間序列圖表，將長達7 ~ 9小時之多項生理訊號以精簡之壓縮型式呈現，藉此提供醫療人員快速、精確且完整的各項生理特徵與生理狀態變化等重要資訊，降低在判斷與評估個案之睡眠生理狀態所需的大量時間與人力資源。
目前研究成果已經過台大醫院神經生理診斷科邱銘章醫師之認可，且達到臨床參考價值的標準，希望藉由此系統的開發與應用，能夠提供醫療人員與相關研究專家精確且有效率的輔助工具，並可應用於睡眠生理相關之研究，進而提升整體醫療品質。

關鍵字

小波轉換 ； 小波基底選用準則 ； 模糊C-means分群法 ； 貝氏分類器 ； 腦電訊號 ； 心電訊號干擾 ； 睡眠紡錘波

並列摘要

An automatic analysis method for 「extracting primary rhythms」, 「detecting and eliminating electrocardiograph (ECG) artifacts of electroencephalography (EEG)」, and「automatic detection and recognition of sleep spindles」are proposed in this paper. The idea is to decompose the EEG into quasi-stationary states based on wavelet Multi-Resolution Analysis. Considering the properties of wavelet filters and the relationship between wavelet basis and EEG characteristics, this paper presents a wavelet basis selection criterion to choose suitable basis and optimal scale for decomposition and detection without time-shift by wavelet transform. Unlike previously investigations, the proposed method separates pure rhythms (alpha, beta and slow wave) to reduce the complexity of reviewing EEG and also conforms to medical requirements. Furthermore, an automatic and adaptive method with high reliability to detect and eliminate ECG artifacts from EEGs is also developed without an additional synchronous ECG channel. By using the method, the total detection rate is above 93% for MIT/ BIH database. To the aspect of 「automatic detection and recognition of sleep spindles」, the new wavelet basis 「Spindlet」is created to extract sleep spindles, the total detection rate is above 87.97% for Sleep-EDF database of Physiobank.
Based on the above-mentioned research and methodology, the proposed approach of Automatic Classification System of long-term Polysomnography, ACSP, contains 4 primary procedures: (1) segmentation; (2) feature extraction; (3) classification; (4) presentation. Nonlinear energy operator method is used to divide the prolonged polysomnography into moderate segments, which are utilized to extract the features. All the segment features are used to classify the segments into groups of like patterns by self organization strategy. The analyzed data of the long-term polysomnography are presented in a compressed form in final procedure. This is completed by providing a representative sample from each group and a compressed time profile of the whole polysomnography. The performance evaluations indicate that the approaches are feasible and can be used as a new way for automatic biomedical signal analysis.

並列關鍵字

wavelet transform ； wavelet basis selection criterion ； Fuzzy C-Means clustering ； Bayesian classifier ； EEG ； ECG artifact ； sleep spindle

參考文獻

Acir, N. and C. Guzelis, 2004. Automatic recognition of sleep spindles in EEG by using artificial neural networks. Expert System with Applications, 27, 451-458.

Agarwal, R. and J. Gotman. 1998. Automatic EEG analysis during long-term monitoring in ICU. Electroencephalography and clinical Neurophysiology 107: 44-58.

Agarwal, R. and J. Gotman, 1999. Adaptive Segmentation of Electroencephalographic Data Using A Nonlinear Energy Operator. Circuits and Systems, 1999. ISCAS '99. Proceedings of the 1999 IEEE International Symposium on. 4: 199-202.

Agarwal, R. and J. Gotman, 2001. Computer-Assisted Sleep Staging. IEEE Transactions on Biomedical Engineering 48(12): 1412-1423.

Blanco, S., S. Kochen, O. A. Rosso and P. Salgado, 1997. Applying time-frequency analysis to seizure EEG activity, Engineering in Medicine and Biology Magazine IEEE 16(1): 64-71.