心血管疾病為全世界頭號死因，造成中等收入的開發中國家極大負擔，在行動電話上技術日益進步且每一人擁有兩支手機，故較可行的解決方案是利用手機進行記錄，再傳送並提供具有專業知識者進行診斷，不過又遇到了採集人員專業或經驗不夠之瓶頸，造成未按照標準流程所採集到的數據不堪使用。
麻省理工PhysioNet 團體每年提供了心電圖資料，並且舉辦挑戰以促進各領域學者們一同解決心電圖機器判讀的難題。以往採取頻域分析以及多種資料處理，難保信息失真以及效率不佳，故採取時間序列模型HMM，並且以極少加工之資料來進行分析。
本論文使用心電圖訓練HMM 模型，針對原始心電圖資料序列進行機器學習，希望能判讀一段10 秒鐘所測得的12 導程5000 個數值心電圖，是否按照標準流程所量得。由於單獨使用時準確率著實有限（80.7%），本論文又再結合Adaboost 決策理論合併HMM 分類器之間的意見，得到最佳答案，順利將判讀成功率頓時提高不少（88.1%）。
而改進的著手點可從判讀成功率跟判讀效率兩個方面進行，以求往後面對其它議題也能獲得絕佳組合方案。

Cardiovascular disease is the No. 1 cause of death worldwide, resulting in middle-income developing countries a great burden. The mobile phone technology advances recently, it is more feasible solution is to use their phones to record, re-transmission and provide who have expertise in diagnosis. But collection staff encountered a bottleneck in lack of experience of, resulting not in accordance with standard procedures of the collected data will be unusable.
PhysioNet provides ECG data every year, and contested to promote together scholars in various fields to solve the problem of interpretation
of ECG machines. Taking variety of analyzed in frequency domain, and data processing, there is no guarantee of information distortion and inefficient. So we take the time series model - HMM, and with minimal processing of data for analysis.
This paper training the HMM using ECG, ECG data for the original sequence of machine learning, hoping to interpret the measured period of 10 seconds of 12 lead ECG 5000 value, whether the amount was in accordance with standard procedures. Accuracy when used alone as really limited (80.7%), this paper again merge HMM combining Adaboost classifier decision theory between the views, the best answer, successful interpretation of the success rate will suddenly increase a lot (88.1%).
To improve the detection accuracy and efficiency from two-pronged approach. We wish to face other issues and get a proper combination.

中文摘要 ………………………………………………………………………………………………I
Abstract………………………………………………………………………………………………II
誌謝 ……………………………………………………………………………………………………III
目錄 ……………………………………………………………………………………………………IV
圖目錄 ………………………………………………………………………………………………VI
表目錄 ……………………………………………………………………………………………VIII
1. 緒論 ………………………………………………………………………………………………1
1.1 前言……………………………………………………………………………………………1
1.2 研究動機……………………………………………………………………………………5
1.3 研究議題……………………………………………………………………………………7
1.4 論文架構……………………………………………………………………………………9
2. 理論背景………………………………………………………………………………………10
2.1 隱藏式馬可夫模型……………………………………………………………………10
2.2 Adaptive Boosting 演算法…………………………………………………………15
2.2.1 Adaboost 原理…………………………………………………………………14
2.2.2 Cascade Adaboost …………………………………………………………18
3. Adaboost-HMM…………………………………………………………………………… 22
3.1 將HMM 導入Adaboost ……………………………………………………………22
3.1.1 將HMM 導入 Scouting …………………………………………………24
3.1.2 將HMM 導入Drafting…………………………………………………… 25
3.1.3 將HMM 導入Weighting ………………………………………………29
3.2 改善Cascade 型Adaboost-HMM ……………………………………………33
4. 實驗結果暨討論……………………………………………………………………………37
4.1 實驗環境…………………………………………………………………………………37
4.2 實現Adaboost-HMM…………………………………………………………………40
4.2.1 訓練HMM 模型………………………………………………………………40
4.2.2 測試Adaboost-HMM 模型………………………………………………43
4.3 心電圖偵測………………………………………………………………………………45
4.4 研究成果…………………………………………………………………………………48
4.4.1 ROC 曲線…………………………………………………………………………48
4.4.2 挑戰成果…………………………………………………………………………50
5. 未來方向與結論……………………………………………………………………………53
參考文獻 …………………………………………………………………………………………55

[1] Raymond B., Cayton R.M., Bates R.A., Chappel M.J., “Screening for
Obstructive Sleep Apnoea Based on the Electrocardiogram”, IEEE
conferences, 2000.
[2] Schrader M., Zywietz C., Einem V., Widiger B., Joseph G.,
“etection of sleep apnea in single channel ECGsfrom the PhysioNet
data base”, IEEE conferences, 2000.
[3] Chazal P., Heneghan C., Sheridan E., Reilly R., Nolan P., O''Malley
M., ”Automated processing of the single-leadelectrocardiogram for
the detection of obstructive sleep apnoea”, IEEE Journals,
Biomedical Engineering, Issue:6 , pp. 686 - 696, June 2003.
[4] Maier C., Bauch M., Dickhaus H., “Recognition and quantification
of sleep apnea byanalysis of heart rate variability parameters”, IEEE
conferences, August 2000.
[5] Mietus J.E., Peng C.K., Ivanov P.C., Goldberger A.L., “Detection Of
Obstructive Sleep Apnea From Cardiac Interbeat Interval Time
Series” , IEEE conferences, August 2002.
[6] Rabiner L.R., “A Tutorial on Hidden Markov Models and Selected
Applications in Speech Recognition”, IEEE Journals, Volume 77, pp.
257 – 286, 1989.
[7] McNames J.N., Fraser A.M., “Obstructive sleep apnea classification
based on spectrogram patterns in the electrocardiogram”, IEEE
conferences, 2000.
[8] Lawrence R., Fundamentals of Speech Recognition., Prentice-Hall
International Inc., 1993.
[9] Andrew M.F., 1 edition, Hidden Markov Models and Dynamic
Systems., Society for Industrial Mathematics, March 2009.
[10] Quinlan J.R., Induction of Decision Trees., Kluwer Academic
Publishers, 1985.
[11] Yong Z., Wenxue H., Yonghong X., Jianxin C., “ECG Beats
Classification Based on Ensemble Feature Composed of
Independent Component and QRS Complex Width”, IEEE
conferences, December 2008.
56
[12] Rojas R., “AdaBoost and the Super Bowl of Classifiers - A Tutorial
Introduction to Adaptive Boosting”, Christmas 2009.
[13] Viola P., Jones M.J., “Robust real-time face detection", IEEE
conferences, July 2003.
[14] Wei-Foo S., Lian Y., Dong L.,"Recognition of visual speech elements
using adaptively boosted hidden Markov models", IEEE journals,
Issue 5, Volume 14, pp. 693 - 705, 2004.
[15] Robert E.S., Yoav F., Peter B., Wee S.L., "Boosting the Margin: A
New Explanation for the Effectiveness of Voting Methods", The
Annals of Statistics, May 1998.
[16] Tieu K., Viola P., "Boosting image retrieval", IEEE conferences on
Computer Vision and Pattern Recognition, 2000.