1970年以後由於睡眠呼吸中止症(sleep apnea syndrome，SAS)漸漸的受到重視，因為睡眠呼吸中止症會造成人體精神及心臟血管系統病變，其中又有高達90%以上為阻塞型呼吸中止症(Obstructive sleep apnea, OSA)，目前診斷睡眠呼吸中止症所用的儀器主要多為睡眠呼吸多項生理監測儀（Polysomnography, PSG），然而利用PSG診斷，卻有代價昂貴、過程煩複、睡眠檢查室不足以及耗費醫護人力資源等不利的因素，使得睡眠呼吸中止症的診斷不易普及。
本研究將以患者整夜的血氧飽和濃度，利用此生理訊號運用訊號處理技巧，並經由特徵萃取與人工智慧技術估測呼吸障礙指數RDI。

SAS has became an increasingly important public-health problem since 1970. It can adversely affect neurocognitive, cardiovascular, respiratory diseases and can also cause behavior disorder. Moreover, up to 90% of these cases are obstructive sleep apnea (OSA). Presently, Polysomnography is considered as the gold standard for diagnosing sleep apnea syndrome (SAS). However, Polysomnography-based sleep studies are expensive and time-consuming because they require overnight evaluation in sleep laboratories with dedicated systems and attending personnel.

In this study, based on the nocturnal oxygen saturation (SpO2) signals, this work develops a method to classify patients with different levels of respiratory disturbance index (RDI) values. To achieve this goal, this study uses neural network in conjunction with different sets of feature variables to perform classification.

論文目錄
論文目錄................................................Ⅰ
圖表目錄................................................Ⅲ
論文摘要（中文）........................................Ⅷ
論文摘要（英文）........................................Ⅸ
第一章 緒論............................................1
1.1 前言............................................1
1.2 研究動機........................................2
1.3 論文架構........................................3
第二章 相關背景與研究..................................5
2.1 障礙型睡眠呼吸中止症概述........................5
2.1.1 睡眠呼吸中止症之診斷與治療......................7
2.1.2 血氧飽和度簡介.................................10
2.1.3 血氧飽和度的量測方法...........................11
2.2 相關研究.......................................13
2.2.1 Patrick Lévy等人之研究.............. ..........13
2.2.2 Magalang等人之研究.............................14
2.2.3 Carlos Zamarrón等人之研究........... ...........15
第三章 研究資料與方法.................................17
3.1 研究資料.......................................17
3.2 資料前處理.....................................18
3.3 mRDI...........................................19
3.4 特徵介紹.......................................20
3.4.1 ODI............................................20
3.4.2 △Index........................................23
3.4.3 時域特徵變數...................................24
3.4.4 頻域特徵變數...................................25
3.5 分類器.........................................27
3.5.1 類神經網路.....................................27
3.5.2 MLP網路架構....................................29
3.5.3 MLP網路在型態鑑別上的相關設定..................30
3.6 委員會機器.....................................31
3.6.1 Averaging Committee............................32
第四章 實驗與結果.....................................34
4.1 實驗設定.......................................34
4.2 實驗結果.......................................35
4.3 實驗結果分析...................................41
第五章 結論...........................................52
參考文獻................................................53
圖目錄
圖2.1阻塞型睡眠呼吸中止症示意圖..........................5
圖 2.2 阻塞型睡眠呼吸中止症睡眠週期流程圖................6
圖2.3 睡眠檢查 (睡眠多項生理檢查Polysomnography) ........7
圖2.4 戴上睡眠呼吸輔助器的情形...........................9
圖2.5 各種口內裝置......................................10
圖2.6氧化血色素和還原血色素之不同波長吸收度.............12
圖2.7 Sensitivity與Specificity之ROC曲線[Patrick Lévy et al.,1999] ..............................................14
圖2.8 估測AHI與真實AHI對應圖[Magalang et al.,2003] .....15
圖2.9 血氧飽和濃度之頻譜圖..............................16
圖3.1 資料庫，RDI分佈情況...............................17
圖3.2 hypopnea count對應RDI.............................19
圖3.3 apnea count對應RDI................................19
圖3.4 計算血氧飽和濃度之示意圖..........................21
圖3.5 ODI3對應RDI.......................................21
圖3.6 ODI4對應RDI.......................................22
圖3.7 ODI5對應RDI.......................................22
圖3.8 △Index對應RDI....................................23
圖3.9 計算血氧飽和濃度變異數之mean之示意圖..............24
圖3.10 血氧飽和濃度變異數之均值(V)對應RDI...............24
圖3.11 計算頻寬之示意圖.................................25
圖3.12 血氧飽和濃度頻寬均值(B)對應RDI...................26
圖3.13 生物神經元.......................................27
圖3.14 一MLP網路架構....................................29
圖3.15 MLP網路的訊號傳遞方向，以及誤差訊號的傳遞方向....30
圖3.16 Averaging Committee的組合方式....................32
圖3.17 mRDI流程示意圖...................................33
圖4.1 資料分配圖........................................34
圖4.2 單一維度之testing acuracy.........................36
圖4.3 單一維度之training acuracy........................36
圖4.4 單一維度之validation accuracy.....................37
圖 4.5 特徵變數與RDI之correlation.......................39
圖4.6 雙特徵維度之testing acuracy.......................39
圖4.7雙特徵維度之Training Accuracy......................40
圖4.8雙特徵維度之Validation Accuracy....................40
圖4.9 百分之百分對之RDI統計圖...........................42
圖4.10 百分之百分錯之RDI統計圖..........................42
圖4.11 不含百分百分對及百分百分錯之RDI統計圖............42
圖4.12　c=0.6相對c=1時，apnea與hypopnea之關係...........48
圖4.13 分段之Testing Accuracy...........................50
圖4.13 分段之Training Accuracy..........................50
圖4.14 分段之Vaildation Accuracy........................51
表目錄
表2.1 各△Index 門檻之Sensitivity與Specificity[Patrick Lévy et al.,1999] ........................................14
表3.1　資料庫資料內容...................................17
表4.1 選用之特徵變數....................................35
表4.2 類神經網路的參數設定..............................35
表4.3 單一維度中，各維度Testing最佳之mRDI coefficient...37
表4.4 單一維度中，各維度Training最佳之mRDI coefficient..37
表4.5 單一維度中，各維度Validation最佳之mRDI coefficient.............................................38
表4.6 特徵變數相互之correlation表.......................38
表4.7 雙特徵維度中，各維度Testing最佳之mRDI coefficient.41
表4.8 雙特徵維度中，各維度Training最佳之mRDI coefficient.............................................41
表4.9 雙特徵維度中，各維度Validation最佳之mRDI coefficient ............................................41
表4.10 mRDI coefficient=0.6時，分類精度變差之資料整理...43
表4.11 mRDI coefficient=0.6時，分類精度變差之特徵資料整理 .....................................................44
表4.12 mRDI coefficient=0.6時，分類精度變好之資料整理...45
表4.13 mRDI coefficient=0.6時，分類精度變好之特徵資料整理......................................................46
表4.14 mRDI coefficient=0.6及mRDI coefficient＝1全部分錯之資料整理................................................47
表4.15 mRDI coefficient=0.6及mRDI coefficient＝1全部分錯之特徵資料整理............................................47
表4.16特徵資料之mean值..................................48
表4.17 男女差異分析資料表...............................49

參考文獻
Atiya AF, El-Shoura SM, Shaheen SI and El-Sherif MS, 1999, “A comparison between neural-nework forecasting techniques-case study: river flow forecasting, “ IEEE Transactions on Neural Networks, vol 10(2), pp402-409.

Atul Malhotra and David P white, 2002, “Obstructive sleep apnoea,” Lancet, vol. 360, pp. 237-245.

Behbehani K, Yen FC, 1995, “Automatic control of airway pressure for treatment of obstructive sleep apnea,” IEEE Transactions on Bio medical Engineering, vol. 42 , NO. 10, pp. 1007-1016.

Behbehani K, Yen FC, Axe J, Burk J, and Lucas E, 1993, “Adaptive positive airway pressure therapy for obstructive sleep apnea,” IEEE Transactions on Biomedical Engineering, pp. 970-971.

Cao L, Tay FEH, 2001, “Financial forecasting using support vector machines,” Neural Comput & Applic, vol. 10, pp. 184-192.

Charylonink W and Chen MS, 2000, “Very short-term load forecasting using artificial neural networks, “IEEE Transactions on Power Systems, vol. 15(1), pp. 263-268

Haykin S, 1999, Neural Networks: A Comprehensive foundation, 2nd ed. Englewood cliffs, NJ: Pretice-Hall.
Juan-Carlos V, Willis HT, W Ward F, Akira M, Rollin B, Eric H, William AW,
　John ER, 1999, “Automated analysis of digital oximetry in the diagnosis of obstructive sleep apnea,” Thra 2000, vol. 55, pp.302-307.

Lavie P, Here P, Hoffstein V, 2000, “Obstructive sleep apnea syndrome as a risk factor for hypertension, “ BMJ, vol. 320, pp. 479-482.

Leo, B., 1996, “Bagging Predictors, “ Machine Learning, 24, pp.123-140.

Marcelo CM and Àlvaro V, 2000, “A hybrid linear-neural model for time series forecastiong, “IEEE Trans. Neural Networks, vol. 11(6), pp. 1402-1412.

Montserrat JM, Rarre R, Navajas D, 2001, “Automatic continuous positive airway pressure devices for the treatment of sleep apnea hypopnea syndrome, “Sleep Medicine, vol 2, pp. 95-98.

Paola Lanfranchi and Virend K Somers, 2001, “Obstructive sleep apnea and vascular disease, “ Respir Res, vol. 2, pp. 315-319.

Pattie DC and Snyder J, 1996, “Using a neural network to forecast visitor behavior,” Annals of Tourism Research, vol. 23(1), pp. 151-164.

Ranta SOV, Hynynen Mand Räsänen J, 2002, “Application of artificial neural networks as an indicator of awareness with recall during general anaesthesia,” Journal of Clinical Monitoring and Computing, vol. 17, pp.53-60.

Robert C, Karasinski P, Arreto CD and Gaudy JF, 2002, “Monitoring anesthesia using neural networks: a survery,” Journal of Clinical Monitoring and Computing, vol. 17, pp. 259-267.

Rosenthal L, Gerhardstein R, Lumley A, Guido P, Day R, Syron ML, Roth T, 2000, “CPAP therapy in patients with mild OSA:implementation and treatment outcome,” Sleep Medicine, vol. 1, pp. 215-220.

Rossiev DA, Golovenkin SE, Shulman VA and Matjushin GV, 1995, “Neural networks for forecasting of myocardial infarction complications, “2nd International Symposium on Neuroinformatics and Neurocomputers , pp. 292-298.

Teran-Santos J, Jimenez-Gomez A, Cordero-Guevara J, 1999, “The association between sleep apnea and risk of traffic accidents,” N Engl J Med,vol. 340, pp. 847-851.

Ulysses JM, Jacek D, Sateesh V, Azmi D, M. Jeffery M, Ali E, and Brydon J. B. G, 2003, “Prediction of the Apnea-Hypopnea Index From Overnight Pulse Oximetry, “ Chest, vol. 124, pp. 1694-1701.

Wilson H and Recknagel F, 2001, “Towards a generic artificial neural network model for dynamic predictions of algal abundance in freshwater lakes,” Ecological Modelling, vol. 146, pp. 69-84.


Yen FC, Khosrow MS, John R, Burk MD, 1996, “Long trem performance evaluation of an automatic airway positive pressure device,” IEEE Engineering in Medicine and Biomedical Society, pp. 2113-2114.

Yoav, F., and Robert, E. S., 1997, “A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting,” Journal of Computer and System Sciences, 55, pp. 119-139.

Young T, Palta M, Dempsey J, Skatrud J, Weber S, Badr S, 1993, “The occurrence of sleep-disordered breathing among middle-aged adults,” N Engl J Med, vol. 328, pp. 1230-1235.

Young T, Peppard P, Palta M, Hla KM, Finn L, Margan B, and Skatrud J, 1997, “Population-based study of sleep-disordered breathing as a risk factor for hypertension,” Arch. Intern. Med., vol. 157, pp. 1746-1752.

劉勝義, 2004, 臨床睡眠檢查學，合記圖書出版社.

王人鋒, 2003, “兩階段決策之委員會機器” , 碩士論文, 中山大學機械與機電工程研究所.

王元宏, 2004, “以心電圖訊號檢測睡眠呼吸中止症的類神經網路“, 碩士論文,　中山大學機械與機電工程研究所.

張書豪, 2005, “使用血氧飽和度估測呼吸障礙指數” , 碩士論文,　中山大學機械與機電工程研究所.

陳志金,2005 , “睡眠呼吸中止症候群的診斷與治療”, http://www.tssm.org.tw/doc/Education/osas_ECKH.htm

http://www.tssm.org.tw/sleepforum/index.php?showtopic=22

http://www.tssm.org.tw