摘要
近年來，機器學習已在各個工程領域上大放異彩。有鑑於此，本論文使用機器學習方法來偵測心衰竭。第一種方法使用支援向量機(SVM)配合心律變異度分析來偵測心衰竭。利用心律變異分析計算出時間、頻率、非線性等特徵。接著使用統計分析來從上述24個特徵挑選出具有顯著差異的特徵。最後採用循序向前選取法(SFS)來找出最適合支援向量機的前五個特徵。經過以上種種步驟後，支援向量機在台大醫院臨床資料庫的訓練集和測試集上的準確率分別是78.37%、72.91%。第二種方法是以一維卷積神經網路處理，可以分成三個模塊:小波濾波、建立模型、分類。我採用symlet5作為母波濾除雜訊，模型方面參考卷積神經網路來建立，最後使用台大醫院臨床資料庫在模型訓練時可達99.56%準確率，在測試時可達99.59%準確率。
本篇論文也探討此兩種分類方法的差異，最後利用開源的MIT資料集測試模型的遷移學習能力。以臺大醫院臨床資料庫上做訓練，MIT的資料做為測試資料的情境下訓練集和測試集的準確率分別是82.4%和63.4%。

Recently, machine learning approaches generally work well in a variety of engineering areas. In this thesis, I proposed classification systems for congestive heart failure (CHF). First, I applied support vector machine (SVM) to detect CHF via heart rate variability (HRV) analysis. Using HRV to derive time, frequency, non-linear features. Then a statistical method is done to verify whether there exist significant difference in 24 features. The last step is sequential forward selection (SFS) that searches the top 5 features for SVM model. After doing these steps, SVM model could attain 78.37\% accuracy on NTUH training set, 72.91\% accuracy NTUH testing set. The second approach is 1D convolutional neural network (CNN) model, it could be composed of three stages: wavelet filtering, model instantiation, classification. I chose symlet5 as mother wave to filter noisy signal. I adopted convolution neural network as model backbone. Afterward, the classification result of convolutional neural network (CNN) model were 99.56\% training accuracy and 99.59\% testing accuracy in NTUH database.

The comparison of these two approaches were discussed in this thesis. Using public MIT-BIH dataset to examine transferability of model. I chose NTUH dataset as training set, MIT-BIH dataset as testing set. The accuracy of training set and testing set are 82.4\%, 63.4\% respectively.

[1] “Cardiovascular diseases (CVDs),” Sept. 2020. [Online]. Available: https://www.who.int/health-topics/cardiovascular-diseases/#tab=tab1
[2] “Taiwan statistical result of cause of death in 2019,” June. 2019. [Online]. Available:https://www.mohw.gov.tw/cp-16-54482-1.html
[3] “Lateststatisticsshowheartfailureontherise;cardiovascu-lardiseasesremainleadingkiller,”January2017.[Online].Avail-able:http://newsroom.heart.org/news/latest-statistics-show-heart-failure-on-the-rise;-cardiovascular-diseases-remain-leading-killer
[4] M. Y. Lee and S. N. Yu, “Selection of heart rate variability features for congestive heartfailure recognition using support vector machine-based criteria,” inSpringer 5th Euro-pean Conference of the International Federation for Medical and Biological Engineer-ing, 2011, pp. 400–403.
[5] L. Pecchia, P. Melillo, M. Sansone, and M. Bracale, “Discrimination power of short-termheart rate variability measures for CHF assessment,”IEEE Transactions on InformationTechnology in Biomedicine, vol. 15, no. 1, pp. 40–46, 2011.
[6] K. H. Yoon, T. Thap, C. W. Jeong, N. H. Kim, S. Noh, Y. Nam, and J. Lee, “Analy-sis of statistical methods for automatic detection of congestive heart failure and atrialfibrillation with short RR interval time series,” inIEEE 2015 9th International Confer-ence Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS), 2015,pp. 452–457.
[7] C. G. C. U. National, “Chronic heart failure: National clinical guideline for diagnosisand management in primary and secondary care: Partial update,” 2010.
[8] H.-H. Osterhues, G. Großmann, M. Kochs, and V. Hombach, “Heart-rate variabilityfor discrimination of different types of neuropathy in patients with insulin-dependentdiabetes mellitus,”Journal of Endocrinological Investigation, vol. 21, no. 1, pp. 24–30,1998. [Online]. Available: http://dx.doi.org/10.1007/BF03347282
[9] R. Virtanen, A. Jula, T. Kuusela, H. Helenius, and L. M. Voipio-Pulkki, “Reduced heartrate variability in hypertension: associations with lifestyle factors and plasma renin ac-tivity,”Journal of Human Hypertension, vol. 17, no. 3, pp. 171–179, 2003.
[10] M. G. Kienzle, D. W. Ferguson, C. L. Birkett, G. A. Myers, W. J. Berg,and D. Mariano, “Clinical, hemodynamic and sympathetic neural correlatesof heart rate variability in congestive heart failure,”The American Journalof Cardiology, vol. 69, no. 8, pp. 761 – 767, 1992. [Online]. Available:http://www.sciencedirect.com/science/article/pii/000291499290502P
[11] U. Rajendra Acharya, K. Paul Joseph, N. Kannathal, C. M. Lim, and J. S.Suri, “Heart rate variability:a review,”Medical and Biological Engineeringand Computing, vol. 44, no. 12, pp. 1031–1051, 2006. [Online]. Available:http://dx.doi.org/10.1007/s11517-006-0119-0
[12] “Heart rate variability,” 2016. [Online]. Available:https://www.firstbeat.com/en/science-and-physiology/heart-rate-variability/.
[13] E. von Borell, J. Langbein, G. Despr ́es, S. Hansen, C. Leterrier, J. Marchant-Forde,R. Marchant-Forde, M. Minero, E. Mohr, A. Prunier, D. Valance, and I. Veissier, “Heartrate variability as a measure of autonomic regulation of cardiac activity for assessingstress and welfare in farm animals — a review,”Physiology & Behavior, vol. 92,no. 3, pp. 293 – 316, 2007, stress and Welfare in Farm Animals. [Online]. Available:http://www.sciencedirect.com/science/article/pii/S0031938407000157

[14] R. Sassi, S. Cerutti, F. Lombardi, M. Malik, H. V. Huikuri, C.-K. Peng, G. Schmidt, andY. Yamamoto, “Advances in heart rate variability signal analysis: joint position statementby the e-cardiology ESC working group and the european heart rhythm association co-endorsed by the asia pacific heart rhythm society,”Ep Europace, vol. 17, no. 9, pp.1341–1353, 2015.
[15] M. Pagani, F. Lombardi, S. Guzzetti, O. Rimoldi, R. Furlan, P. Pizzinelli, G. Sandrone,G. Malfatto, S. Dell’Orto, and E. Piccaluga, “Power spectral analysis of heart rate andarterial pressure variabilities as a marker of sympatho-vagal interaction in man and con-scious dog.”Circulation Research, vol. 59, no. 2, pp. 178–193, 1986.
[16] A. Malliani, M. Pagani, F. Lombardi, and S. Cerutti, “Cardiovascular neural regulationexplored in the frequency domain.”Circulation, vol. 84, no. 2, pp. 482–492, 1991.
[17] J. S. Richman and J. R. Moorman, “Physiological time-series analysis using approximateentropy and sample entropy,”American Journal of Physiology-Heart and CirculatoryPhysiology, vol. 278, no. 6, pp. H2039–H2049, 2000.
[18] M. Costa, A. L. Goldberger, and C.-K. Peng, “Multiscale entropy analysis of biologicalsignals,”Physical Review, vol. 71, no. 2, p. 021906, 2005.
[19] C.-K. Peng, S. V. Buldyrev, S. Havlin, M. Simons, H. E. Stanley, and A. L. Goldberger,“Mosaic organization of dna nucleotides,”Physical review e, vol. 49, no. 2, p. 1685,1994.
[20] B. B. Mandelbrot and B. B. Mandelbrot,The fractal geometry of nature. WH freemanNew York, 1982, vol. 1.
[21] R. Hardstone, S.-S. Poil, G. Schiavone, R. Jansen, V. V. Nikulin, H. D. Mansvelder, andK. Linkenkaer-Hansen, “Detrended fluctuation analysis: a scale-free view on neuronaloscillations,”Frontiers in Physiology, vol. 3, p. 450, 2012.
[22] J. Bulao,How Much Data Is Created Every Day in 2020, 2020. [Online]. Available:https://techjury.net/blog/how-much-data-is-created-every-day/#gref
82BIBLIOGRAPHY
[23] M. P. Fay and M. A. Proschan, “Wilcoxon-mann-whitney or t-test? on assumptions forhypothesis tests and multiple interpretations of decision rules,”Statistics Surveys, vol. 4,p. 1, 2010.
[24] I. Guyon, S. Gunn, M. Nikravesh, and L. A. Zadeh,Feature extraction: foundations andapplications. Springer, 2008, vol. 207.
[25] P. Pudil, J. Novoviˇcov ́a, and J. Kittler, “Floating search methods in feature selection,”Pattern Recognition Letters, vol. 15, no. 11, pp. 1119–1125, 1994.
[26] M. T. Ribeiro, S. Singh, and C. Guestrin, “Why should I trust you?” explaining thepredictions of any classifier,” inProceedings of the 22nd ACM SIGKDD internationalconference on knowledge discovery and data mining, 2016, pp. 1135–1144.
[27] M. Kachuee, S. Fazeli, and M. Sarrafzadeh, “ECG heartbeat classification: A deep trans-ferable representation,” in2018 IEEE International Conference on Healthcare Informat-ics (ICHI), 2018, pp. 443–444.
[28] S. Ross-Howe and H. R. Tizhoosh, “Atrial fibrillation detection using deep features andconvolutional networks,” in2019 IEEE EMBS International Conference on Biomedical& Health Informatics (BHI), 2019, pp. 1–4.
[29] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convo-lutional neural networks,”Advances in neural information processing systems, vol. 25,pp. 1097–1105, 2012.
[30] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke,and A. Rabinovich, “Going deeper with convolutions,” inProceedings of the IEEE con-ference on computer vision and pattern recognition, 2015, pp. 1–9.
[31] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016,pp. 770–778.
BIBLIOGRAPHY83
[32] S. A. Mahmoud, A. Bamakhramah, and S. A. Al-Tunaiji, “Six order cascaded powerline notch filter for ECG detection systems with noise shaping,”Circuits, Systems, andSignal Processing, vol. 33, no. 8, pp. 2385–2400, 2014.
[33] Z. Zidelmal, A. Amirou, M. Adnane, and A. Belouchrani, “QRS detection based onwavelet coefficients,”Computer Methods and Programs in Biomedicine, vol. 107, no. 3,pp. 490–496, 2012.
[34] M. A. Kabir and C. Shahnaz, “Denoising of ECG signals based on noise reduction algo-rithms in emd and wavelet domains,”Biomedical Signal Processing and Control, vol. 7,no. 5, pp. 481–489, 2012.
[35] W. Chen, G. Liu, S. Su, Q. Jiang, and H. Nguyen, “A CHF detection method based ondeep learning with rr intervals,” in2017 39th Annual International Conference of theIEEE, IEEE Engineering in Medicine and Biology Society (EMBC), 2017, pp. 3369–3372.
[36] H. Zhang, W. Zhao, and S. Liu, “SE-ECGNet: A multi-scale deep residual net-work with squeeze-and-excitation module for ECG signal classification,”arXiv preprintarXiv:2012.05510, 2020.
[37] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” inProceedings of theIEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141.
[38] A. Ullah, S. M. Anwar, M. Bilal, and R. M. Mehmood, “Classification of arrhythmiaby using deep learning with 2-d ECG spectral image representation,”Remote Sensing,vol. 12, no. 10, p. 1685, 2020.
[39] G. Van Steenkiste, G. van Loon, and G. Crevecoeur, “Transfer learning in ECG classifi-cation from human to horse using a novel parallel neural network architecture,”ScientificReports, vol. 10, no. 1, pp. 1–12, 2020.
[40] H. Tung, C. Zheng, X. Mao, and D. Qian, “Multi-lead ECG classificationvia an information-based attention convolutional neural network,”arXiv preprintarXiv:2003.12009, 2020.
84BIBLIOGRAPHY
[41] Y. Liu, K. Wang, Q. Li, R. He, Y. Yuan, and H. Zhang, “Weakly supervisedarrhythmia detection based on deep convolutional neural network,”arXiv preprintarXiv:2012.05641, 2020.
[42] D. L. Donoho and J. M. Johnstone, “Ideal spatial adaptation by wavelet shrinkage,”Biometrika, vol. 81, no. 3, pp. 425–455, 1994.
[43] V. Nair and G. E. Hinton, “Rectified linear units improve restricted boltzmann ma-chines,” inICML, 2010.
[44] P. Ramachandran, B. Zoph, and Q. V. Le, “Searching for activation functions,”arXivpreprint arXiv:1710.05941, 2017.
[45] S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training byreducing internal covariate shift,”arXiv preprint arXiv:1502.03167, 2015.
[46] M. Lin, Q. Chen, and S. Yan, “Network in network,”arXiv preprint arXiv:1312.4400,2013.
[47] S.-H. Huang, “A congestive heart failure detection system via multi-input deep learningnetwork,” Master’s thesis, National Tsing Hua University, 2018.
[48] H.-S. Shiue, “Risk assessment in patients suffering from congestive heart failure vialong-term and short-term heart rate variability,” Master’s thesis, National Tsing HuaUniversity, 2017.
[49] U. R. Acharya, H. Fujita, S. L. Oh, Y. Hagiwara, J. H. Tan, M. Adam, and R. San Tan,“Deep convolutional neural network for the automated diagnosis of congestive heartfailure using ECG signals,”Applied Intelligence, vol. 49, no. 1, pp. 16–27, 2019.