心臟病一直以來都是人類健康殺手，一旦病發有極高的致命性。而現行心臟病的醫療十分被動，多數情況下於病人發病後，才在緊急狀況下進行治療，容易造成遺憾。因此，近年來心臟病的長期監控逐漸受到重視，藉由攜帶式或居家式心電圖量測儀器，病患可以自行量測心電圖並即時傳送到醫院，使醫生更確實地掌握病患的近況，並在出現異常徵兆時，及時進行治療。
但長期監控同時也會造成資料量上升以及資料品質穩定度下降的問題，本研究針對其提出一套結合深度學習領域中的卷積神經網路架構與k-means演算法的心律不整輔助診斷系統，並在最後對分類結果進行離群資料處理。該系統能對心電圖資料進行初步分析與分類，在診斷時可提供資訊給醫療人員，並標註出危險度較高的區段，減少醫療人員的負擔，並提升診察效率。
本研究提出之系統分為兩個部分，第一部分以k-means演算法對原始資料進行分子類動作，透過此步驟可以使資料特徵更容易被捕捉。第二部分以卷積神經網路為基礎，建構心電圖波形分類器，對分完子類之資料進行分類並將結果進行離群資料處理。有別於傳統類神經網路，卷積神經網路的特點在於該網路能自動提取資料中價值較高的特徵，並運用於資料的辨識，同時卷積神經網路架構還能有效減少網路之參數數量，減少神經網路的訓練時間與消耗資源，提升整體效率。
本研究使用MITBIH心律不整資料庫進行測試，準確度平均可達99.41%，漏診率僅1.23%。整體而言，為一兼具穩定及高效之智慧型診斷輔助系統。

A Heart Attack has always been a deadly killer to humans’ health. It is unpredictable and fatal. However, the current related heart attack therapies are inactive. Most people look for first aid in an emergency, after someone has a heart attack, which cause tragedies easily. Recently, long-term monitoring has gradually been valued. People can take portable ECG machine and use Home ECG machine to record their cardiograms and send them back to the hospital immediately. If the doctors are aware of any unusual signs, patients will be notified that they should go to the hospital as soon as possible.
At the same time, the long-term monitoring will cause the problems of the arising data base and declining in data quality. This study provides an auxiliary diagnostic system of arrhythmias in deep learning with a combination of convolution neural network (CNN), k-means clustering and outlier processing. The system can provide an initial analysis and classification of ECG signals and mark the more dangerous area for the doctors. In this way, we may enlighten the doctors’ burden and increase the efficiency of diagnosis.
This study includes two parts. First, the original data will be classified by k-means clustering. In this way, the features of the data will be extracted more easily. Second, the classifier of ECG signals, which can be used to classify the data of the subgroups, will be built on CNN, and conduct an outlier processing on the results. The differences between CNN and traditional methods are that CNN can get higher-valued features automatically and used in the identification of the data. CNN can also reduce the amount of network parameter, the training time of the neural network and the consumption of resources to increase the overall efficiency.
This study uses MTIBIH arrhythmias database to test the training model. The average accuracy rate of classification can be as high as 99.41%, only with a false negative rate of 1.23%. In conclusion, it is a smart auxiliary diagnostic system featured with stability and efficiency.

[1] 中華民國衛生福利部, "104年死因統計年報," 2015.
[2] S. Kiranyaz, T. Ince, and M. Gabbouj, "Real-time patient-specific ECG classification by 1-D convolutional neural networks," IEEE Transactions on Biomedical Engineering, vol. 63, no. 3, pp. 664-675, 2016.
[3] 張家熏, "基於K-means 演算法、小波轉換及支持向量機之心電訊號辨識系統," 碩士, 機電科技研究所, 國立臺灣師範大學, 台北市, 2011.
[4] M. Abadi et al., "Tensorflow: Large-scale machine learning on heterogeneous distributed systems," arXiv preprint arXiv:1603.04467, 2016.
[5] P. Davey, 圖解心電圖快速導讀 (ECG at a Glance). 合計圖書出版社, 2011.
[6] MIT-BIH Arrhythmia Database. Available: https://physionet.org/physiobank/database/mitdb/#additional-references
[7] J. Malmivuo and R. Plonsey, Bioelectromagnetism: principles and applications of bioelectric and biomagnetic fields. Oxford University Press, USA, 1995.
[8] 賀立婷. (2016). 心律不整的臨床治療-你的心臟沒電了嗎? Available: http://www.ntuh.gov.tw/cvc/knowledge/%E4%BD%A0%E5%BF%83%E8%87%9F%E6%B2%92%E9%9B%BB%E4%BA%86%E5%97%8E.aspx
[9] M. Lewell and M. Davis, "12 Lead ECG," 2011.
[10] R. MacLeod and B. Birchler, "ECG measurement and analysis," ed: March, 2005.
[11] 林昆宏, "針對用於遠距醫療的量測裝置之心電圖特徵描繪演算法開發," 碩士, 機電工程學系, 國立臺灣師範大學, 台北市, 2016.
[12] G. B. Moody and R. G. Mark, "The impact of the MIT-BIH arrhythmia database," IEEE Engineering in Medicine and Biology Magazine, vol. 20, no. 3, pp. 45-50, 2001.
[13] R. Y. Rubinstein, A. Ridder, and R. Vaisman, "Cross‐Entropy Method," Fast Sequential Monte Carlo Methods for Counting and Optimization, pp. 6-36.
[14] D. Kingma and J. Ba, "Adam: A method for stochastic optimization," arXiv preprint arXiv:1412.6980, 2014.
[15] J. A. Hartigan and M. A. Wong, "Algorithm AS 136: A k-means clustering algorithm," Journal of the Royal Statistical Society. Series C (Applied Statistics), vol. 28, no. 1, pp. 100-108, 1979.
[16] D. MacKay, "An example inference task: clustering," Information theory, inference and learning algorithms, vol. 20, pp. 284-292, 2003.
[17] G. Hamerly and C. Elkan, "Alternatives to the k-means algorithm that find better clusterings," in Proceedings of the eleventh international conference on Information and knowledge management, 2002, pp. 600-607: ACM.
[18] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," Nature, vol. 521, no. 7553, pp. 436-444, 2015.
[19] W. S. McCulloch and W. Pitts, "A logical calculus of the ideas immanent in nervous activity," The bulletin of mathematical biophysics, vol. 5, no. 4, pp. 115-133, 1943.
[20] P. J. Werbos, "Beyond regression: new tools for prediction and analysis in the behavioral science," Ph. D. Thesis, Harvard University, 1974.
[21] K. Fukushima and S. Miyake, "Neocognitron: A self-organizing neural network model for a mechanism of visual pattern recognition," in Competition and cooperation in neural nets: Springer, 1982, pp. 267-285.
[22] Y. LeCun et al., "Backpropagation applied to handwritten zip code recognition," Neural computation, vol. 1, no. 4, pp. 541-551, 1989.
[23] S. Hochreiter, Y. Bengio, P. Frasconi, and J. Schmidhuber, "Gradient flow in recurrent nets: the difficulty of learning long-term dependencies," ed: A field guide to dynamical recurrent neural networks. IEEE Press, 2001.
[24] G. E. Hinton, "Learning multiple layers of representation," Trends in cognitive sciences, vol. 11, no. 10, pp. 428-434, 2007.
[25] R. Raina, A. Madhavan, and A. Y. Ng, "Large-scale deep unsupervised learning using graphics processors," in Proceedings of the 26th annual international conference on machine learning, 2009, pp. 873-880: ACM.
[26] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," in Advances in neural information processing systems, 2012, pp. 1097-1105.
[27] R. J. Martis, U. R. Acharya, and L. C. Min, "ECG beat classification using PCA, LDA, ICA and discrete wavelet transform," Biomedical Signal Processing and Control, vol. 8, no. 5, pp. 437-448, 2013.