根據世界衛生組織 (WHO) 2016年的研究，心臟病係高居全球人類十大死因之首。而美國心臟協會的一項調查也指出，45%的心臟病是在無預警的情況下發生。因此，多年來有許多研究致力於提升各類心律不整的偵測準確度。這些方法包含使用傳統的形態學分析、統計分析、小波變換、以及近年蓬勃發展的卷積神經網路等技術。本研究的目的，即在利用主成分分析法及卷積神經網路的特點，來提出一個具高準確度，且可解讀心電圖中八種不同心律的演算法。
本研究之一維心電圖資料與標記係取自MIT-BIH心律不整資料庫，我們首先採行訊號前處理將一維心電圖資料擴增為原來4倍。接下來，將每6.4秒的一維心電圖資料，以取樣時間為橫軸，振幅為縱軸，產生二維128*128的灰階圖像集。再將R波波峰標記對應至128個橫軸的位置，然後進行多標記分類卷積神經網路訓練與預測。實驗結果發現，R波波峰定位之靈敏度可達99.95%以上。
接續則是心律種類的偵測，利用取得之R波波峰位置來分隔心搏區間，將每一心搏區間的一維心電圖訊號視為是一個二維128*128解析度的灰階圖像。然後利用訊號前處理的結果與心搏區間平移的方法，來進行資料擴增。同時也讀取各心搏區間心律種類的標記，與將卷積神經網路結合各心搏區間的五大主成分，來進行八種心律種類的分類模型訓練與預測。實驗結果證明，所提出之模型可以在有限的系統資源與簡單的卷積神經網路環境下，即達成99.83%的準確度與99.33%的靈敏度，證明本研究所提出之演算法，可以達成優質的心電圖心律分類偵測。

According to the WHO reports, heart disease has been consistently ranked among the top ten causes of human death. Therefore, there have been many studies dedicated to exploring accurate detection algorithms for arrhythmia. These algorithms include the use of morphological image analysis, statistical analysis, wavelet transform, and convolutional neural network (CNN). The goal of this study is to integrate features extracted by principal component analysis (PCA) into CNN to develop an algorithm that can accurately predict the type of Electrocardiography (ECG) arrhythmia. The one-dimensional (1D) ECG data and annotations were adopted from the MIT-BIH arrhythmia database. First, data augmentation is performed to obtain 4 times number of data sources. To provide a suitable signal for the input of the CNN network, the 1D ECG input signal is treated as a 2D gray image with signal sampling information as one dimension and the signal amplitude as the other dimension. To keep the important information of the 1D ECG signal, the signal’s principal components are also extracted. The 2D image gives the signal timing information and the principal components shows the importance of the wave. This information is combined to do the type classification of arrhythmia. The experimental results show that the proposed method can perform very well for arrhythmia type classification. The experimental results show the proposed method can achieve 99.83% accuracy and 99.33% sensitivity.

摘要 I
誌謝 XVIII
目錄 XIX
表目錄 XXIII
圖目錄 XXIV
符號列表 XXVII
第一章 緒論 1
1.1 研究動機 1
1.2 研究目的 2
1.3 論文架構 2
第二章 相關研究探討 3
2.1 心電圖 3
2.1.1 心電圖原理與組成 3
2.1.2 心電圖量測法 5
2.1.3 心電圖的波形 6
2.2 形態學影像處理 (Morphological Image Processing) 8
2.2.1 結構元素(Structure Element) 8
2.2.2 侵蝕(Erosion) 9
2.2.3 膨脹(Dilation) 9
2.2.4 斷開(Open) 10
2.2.5 閉合(Close) 10
2.2.6 中心點移動平均法(Centered Moving Average, CMA) 10
2.3 主成分分析法(Principal Component Analysis, PCA) 11
2.3.1 原理說明 11
2.3.2 主成分分析法的統計學觀點[1] 12
2.4 卷積神經網路(Convolution Neural Network, CNN) 14
2.4.1 半精度浮點數(Half Precision Floating Point) 14
2.4.2 權重值初始化(Weight Initialization) 15
2.4.3 輸入層正規化(Normalization又稱Regularization) 18
2.4.4 隱藏層正規化(Hidden Layer Normalization) 20
2.4.5 一位有效編碼(One Hot Encoding) 22
2.4.6 資料擴增(Data Augmentation) 22
2.4.7 激勵函數(Activation Function) 23
2.4.8 損失函數(Loss Function) 27
2.4.9 單標籤分類(Single-label Classification) 31
2.4.10 多標籤分類(Multi-label Classification) 31
2.5 相關研究 33
2.5.1 QRS波偵測 33
2.5.2 心律不整偵測 35
第三章 研究方法 37
3.1 整體架構 37
3.1.1 R波波峰定位研究主要架構 38
3.1.2 心律不整分類研究主要架構 39
3.2 資料取得 41
3.3 訊號前處理 44
3.4 資料擴增(Data Augmentation) 46
3.5 主成分分析計算(Principle Component Analysis) 48
3.6 卷積神經網路架構 48
3.6.1 R波波峰定位卷積神經網路 48
3.6.2 R波波峰定位卷積網路整體模型 52
3.6.3 定位結果還原為原始ECG位置 53
3.6.4 心律不整分類 54
3.6.5 心律不整分類卷積網路整體模型 56
第四章 實驗設置 58
4.1 實驗環境 58
4.2 實驗資料 58
4.2.1 一維心電圖訊號資料 58
4.2.2 原始訊號與訊號前處理後之訊號源 59
4.2.3 二維心電圖圖像資料 61
4.3 實驗驗證 64
4.3.1 K折交叉驗證(K-fold Cross Validation) 64
4.3.2 評估準則 65
4.4 實驗結果 68
4.4.1 R波波峰定位結果 68
4.4.2 R波波峰定位模型成果比較 71
4.4.3 心律分類結果 72
4.4.4 心律分類模型成果比較 77
第五章 結論與未來展望 78
5.1 結論 78
5.2 未來展望 78
參考文獻 80

[1]T. Huang, 機器/統計學習:主成分分析(Principal Component Analysis, PCA), Medium, 2018, Retrieved from https://medium.com/@chih.sheng.huang821/%E6%A9%9F%E5%99%A8-%E7%B5%B1%E8%A8%88%E5%AD%B8%E7%BF%92-%E4%B8%BB%E6%88%90%E5%88%86%E5%88%86%E6%9E%90-principle-component-analysis-pca-58229cd26e71.
[2]U. R. Acharya, H. Fujita, O. S. Lih, Y. Hagiwara, J. H. Tan, and M. Adam, Automated detection of arrhythmias using different intervals of tachycardia ECG segments with convolutional neural network, Information Sciences, Vol. 405, 2017, pp. 81-90.
[3]E. Alickovic, and A. Subasi, Medical Decision Support System for Diagnosis of Heart Arrhythmia using DWT and Random Forests Classifier, Journal of Medical Systems, Vol. 40, 2016, p. 108.
[4]N. M. Arzeno, Z.-D. Deng, and C.-S. Poon, Analysis of First-Derivative Based QRS Detection Algorithms, IEEE Transactions on Biomedical Engineering, Vol. 55, 2008, pp. 478-484.
[5]J. L. Ba, J. R. Kiros, and G. E. Hinton, Layer Normalization, stat, Vol. 1050, 2016, p. 21.
[6]L. Boltzmann, Weitere studien über das wärmegleichgewicht unter gasmolekülen, Springer, 1970.
[7]L. Cuiwei, Z. Chongxun, and T. Changfeng, Detection of ECG characteristic points using wavelet transforms, IEEE Transactions on Biomedical Engineering, Vol. 42, 1995, pp. 21-28.
[8]The danger of “silent heart attacks, Harvard Health Publishing, 2019, Retrieved from https://www.health.harvard.edu/heart-health/the-danger-of-silent-heart-attacks.
[9]U. Desai, R. J. Martis, C. G. Nayak, K. Sarika, and G. Seshikala, Machine intelligent diagnosis of ECG for arrhythmia classification using DWT, ICA and SVM techniques, Annual IEEE India Conference (INDICON), India, 2015.
[10]H. Feldman, A Guide to Reading and Understanding the EKG, Columbia University, 1999, Retrieved from http://www.columbia.edu/~ss45/EKG-2.PDF.
[11]X. Glorot, and Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, Proceedings of the thirteenth international conference on artificial intelligence and statistics, 2010, pp. 249-256.
[12]A. L. Goldberger, L. A. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov, R. G. Mark, et al., PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals, Circulation, Vol. 101, 2000, pp. e215-e220.
[13]A. Habib, C. Karmakar, and J. Yearwood, Impact of ECG Dataset Diversity on Generalization of CNN Model for Detecting QRS Complex, IEEE Access, Vol. 7, 2019, pp. 93275-93285.
[14]P. S. Hamilton, and W. J. Tompkins, Quantitative Investigation of QRS Detection Rules Using the MIT/BIH Arrhythmia Database, IEEE Transactions on Biomedical Engineering, Vol. BME-33, 1986, pp. 1157-1165.
[15]K. He, X. Zhang, S. Ren, and J. Sun, Delving deep into rectifiers: Surpassing human-level performance on imagenet classification, Proceedings of the IEEE international conference on computer vision, 2015, pp. 1026-1034.
[16]G. E. Hinton, N. Srivastava, A. Krizhevsky, I. Sutskever, and R. R. Salakhutdinov, Improving neural networks by preventing co-adaptation of feature detectors, arXiv preprint arXiv:1207.0580, 2012.
[17]H. Hotelling, Analysis of a complex of statistical variables into principal components, Journal of educational psychology, Vol. 24, 1933, pp. 498-520.
[18]J. S. Huang, B. Q. Chen, B. Yao, and W. P. He, ECG Arrhythmia Classification Using STFT-Based Spectrogram and Convolutional Neural Network, IEEE Access, Vol. 7, 2019, pp. 92871-92880.
[19]IEEE Standard for Floating-Point Arithmetic, IEEE Std 754-2008, 2008, pp. 1-70.
[20]S. Ioffe, and C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift, arXiv preprint arXiv:1502.03167, 2015.
[21]S. Jayaraman, V. Sangareddi, R. Periyasamy, J. Joseph, and R. M. Shanmugam, Modified limb lead ECG system effects on electrocardiographic wave amplitudes and frontal plane axis in sinus rhythm subjects, Anatolian journal of cardiology, 2016.
[22]T. J. Jun, H. M. Nguyen, D. Kang, D. Kim, D. Kim, and Y.-H. Kim, ECG arrhythmia classification using a 2-D convolutional neural network, ArXiv, Vol. abs/1804.06812, 2018.
[23]V. Kalidas, and L. Tamil, Real-time QRS detector using Stationary Wavelet Transform for Automated ECG Analysis, IEEE 17th International Conference on Bioinformatics and Bioengineering (BIBE), 2017, pp. 457-461.
[24]R. E. Klabunde, Cardiovascular Physiology Concepts, Cvphysiology, 2016, Retrieved from https://www.cvphysiology.com/Arrhythmias/A006.
[25]J. L. Lagrange, Mécanique analytique, Mallet-Bachelier, 1853.
[26]J. Malmivuo, and R. Plonsey, Bioelectromagnetism. 15. 12-Lead ECG System, Oxford University Press, 1975.
[27]J. P. Martínez, R. Almeida, S. Olmos, A. P. Rocha, and P. Laguna, A wavelet-based ECG delineator: evaluation on standard databases, IEEE Transactions on Biomedical Engineering, Vol. 51, 2004, pp. 570-581.
[28]G. B. Moody, and R. G. Mark, The impact of the MIT-BIH Arrhythmia Database, IEEE Engineering in Medicine and Biology Magazine, Vol. 20, 2001, pp. 45-50.
[29]S. Nattel, and L. Carlsson, Innovative approaches to anti-arrhythmic drug therapy, Nature Reviews Drug Discovery, Vol. 5, 2006, pp. 1034-1049.
[30]T. Nguyen, X. Qin, A. Dinh, and F. Bui, Low Resource Complexity R-peak Detection Based on Triangle Template Matching and Moving Average Filter, Sensors, Vol. 19, 2019, p. 3997.
[31]J. Pan, and W. J. Tompkins, A real-time QRS detection algorithm, IEEE Trans. Biomed. Eng, Vol. 32, 1985, pp. 230-236.
[32]N. Patchett, File: Limb leads of EKG.png, Wikimedia Commons, the free media repository, 2016, Retrieved from https://commons.wikimedia.org/w/index.php?title=File:Limb_leads_of_EKG.png＆oldid=218319987.
[33]K. Pearson, LIII. On lines and planes of closest fit to systems of points in space, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, Vol. 2, 1901, pp. 559-572.
[34]E. Roth, and B. Wu, 24-Hour Holter Monitoring, Healthline, 2017, Retrieved from https://www.healthline.com/health/holter-monitor-24h.
[35]P. Sasikala, and R. Wahidabanu, Robust r peak and qrs detection in electrocardiogram using wavelet transform, International Journal of Advanced Computer Science and Applications-IJACSA, Vol. 1, 2010, pp. 48-53.
[36]S. Shadmand, and B. Mashoufi, A new personalized ECG signal classification algorithm using Block-based Neural Network and Particle Swarm Optimization, Biomedical Signal Processing and Control, Vol. 25, 2016, pp. 12-23.
[37]C. E. Shannon, A Mathematical Theory of Communication, Bell System Technical Journal, Vol. 27, 1948, pp. 379-423.
[38]M. E. Silverman, and J. Willis Hurst, Willem einthoven-the father of electrocardiography, Clinical Cardiology, Vol. 15, 1992, pp. 785-787.
[39]K. Simonyan, and A. Zisserman, Very deep convolutional networks for large-scale image recognition, arXiv preprint arXiv:1409.1556, 2014.
[40]J. Smith, 30-Year-Old Man Training for First Marathon Says Apple Watch Saved His Life, Runner's World, 2019, Retrieved from https://www.runnersworld.com/runners-stories/a28337843/apple-watch-ecg-saves-mans-life/.
[41]WHO, The top 10 causes of death, World Health Organization, 2018, Retrieved from https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death.
[42]S. C. Wong, A. Gatt, V. Stamatescu, and M. D. McDonnell, Understanding data augmentation for classification: when to warp?, International conference on digital image computing: techniques and applications (DICTA), 2016, pp. 1-6.
[43]Y. Xiang, Z. Lin, and J. Meng, Automatic QRS complex detection using two-level convolutional neural network, BioMedical Engineering OnLine, Vol. 17, 2018.
[44]H. Yang, M. Huang, Z. Cai, Y. Yao, and C. Liu, A Faster R CNN-based Real-time QRS Detector, COMPUTING IN CARDIOLOGY, 2019.
[45]Ö. Yıldırım, P. Pławiak, R.-S. Tan, and U. R. Acharya, Arrhythmia detection using deep convolutional neural network with long duration ECG signals, Computers in Biology and Medicine, Vol. 102, 2018, pp. 411-420.
[46]W. Zhu, X. Chen, Y. Wang, and L. Wang, Arrhythmia Recognition and Classification Using ECG Morphology and Segment Feature Analysis, IEEE/ACM Transactions on Computational Biology and Bioinformatics, Vol. 16, 2019, pp. 131-138.
[47]M. Zubair, J. Kim, and C. Yoon, An Automated ECG Beat Classification System Using Convolutional Neural Networks, 6th International Conference on IT Convergence and Security (ICITCS), 2016.