利用腦電波訊號之循環交替模式偵測

Abstract

本論文之目的在發展睡眠腦電波之循環交替模式偵測演算法，並利用分類器進行改善A-phase之子模式分類準確度。而現代人生活中，睡眠品質的重要性節節提高，但睡眠品質主要利用宏觀結構之睡眠階段來作為判斷依據。微觀結構部分，醫學文獻指出睡眠中循環交替模式與A-phase之子模式可用於判斷睡眠品質。但由於人工判讀循環交替模式費時費力且較主觀，所以尚未大量應用在臨床上。本研究結合過去方法之優點，利用相對性的功率譜數值作為腦波睡眠正規化特徵，並計算最適當循環交替模式長度，以發展改良型長度變化模板式演算法來進行睡眠循環交替模式之偵測。在本演算法中，比較相對性的功率譜數值與閾值來判斷循環交替模式中的A-phase 之存在與否和長度，而在判斷出A-phase之後，利用循環交替模式之標準評比方式決定出單一睡眠循環與睡眠循環序列。在A-phase之子模式分類中則結合相關Carlo Navona等學者在2002年所提出之方法、碎型維度與k最近鄰居法分類器以提升分類準確度。 首先我們利用二位有失眠疾病的受測者的睡眠腦波作為腦電波特徵的測試資料，此兩位受試者循環交替模式之偵測準確度為66%，A-phase之子模式分類準確度為67%。並藉由和Barcaro演算法做比較，驗證本演算法結合改良型長度變化模板式演算法與閾值之優越性。接著我們利用147位呼吸中止指數大於五的呼吸中止症受試者與41位呼吸中止指數小於五的無呼吸中止症受試者睡眠腦波作為腦電波特徵的測試資料。在不同的年齡層中，計算循環交替模式率的結果，呼吸中止症的受測者顯著高於無呼吸中止症的受測者。且在重度呼吸中止症受試者之循環交替模式率會有顯著提升。可驗證我們所設計的睡眠循環交替模式演算法能夠偵測到可信賴的循環交替模式率。

The purpose of this study is to develop a cyclic alternating pattern (CAP) detection algorithm based on EEG signal and to improve the classification accuracy of the A-phase subtypes. In the past study, sleep-related researches have been a lot, in which there is a large part of the analysis on the quality of sleep. But most of researches are applied for macrostructure analysis of sleep stage. Sleep medical literature has demonstrated that microstructure such as CAP and A-phase subtypes can be used in the analysis of quality of sleep. However, because of CAP has been noticed for sleep quality analysis in recent years and its intensive labor requirement, there are very few studies for the microstructure of CAP. Our study has inherited advantages of previous methods. It employed the descriptor of power spectrum, because the descriptor of power spectrum can provide a normalized measure. It also made use of the improved variable length template to detect the most suitable A-phase length of CAP detection algorithm. In order to decide A-phases’ length and existence, our method made use of descriptor to compare with threshold. After the detection of A-phase, it is used for determining CAP cycle and CAP sequence by CAP scoring rule. In the classification of A-phase subtypes, our method has attempted to enhance classification accuracy by using Navona classification method, fractal dimension, and k-NN classifier. In our experiment, we applied sleep EEG features of two insomnia subject's brain waves as the testing data. The accuracy of the proposed method is 65% in the CAP detection and 70.30% in A1, 24.4% in A2, and 65.4% in A3 in the classification of A-phase of subtypes. The proposed method combines variable length template method and EXIST/LENGTH thresholds method. Compared to Barcaro method, the proposed method is superior because of its usage of variable length template. We also used sleep EEG of 147 obstructive sleep apnea syndrome (OSAS) subjects and 41 non-OSAS subjects as the testing data. The CAP rates of 188 OSAS patients obtained by the proposed method are higher than those of non-OSAS subjects, which is consistent to the literature. For the severe OSAS subjects, the CAP detection results show increased values of CAP rates.