臺灣博碩士論文加值系統

心房震顫(Artrial fibrillation)是臨床上最常見的一種心律不整。心房震顫發生時，因心房組織快速不正常放電便使心房無法有效的收縮，導致血液無法正常流動，因而容易產生血栓，當血栓隨著血流流出心臟，便容易造成器官的栓塞，如栓塞於腦部導致腦中風。基於心電圖(ECG)分析預測陣發性心房震顫發生的風險會對中風的臨床預防帶來重大的貢獻。過去的研究指出，P波的形態變化有效反映於心房電生理活動的資訊。然而現有的分析方法仰賴於手動標記特徵，然而這是非常耗時且低效率的。並且在標記這些間接特徵的過程中產生的人為誤差難以被量化。此研究的目標是發展一個快速，自動化識別與P波相關的心房震顫發作指標的分析方法(Morphological P-wave Variant Analysis)，並且利用深度學習模型預測陣發性心房震發生的機率。此方法通過開發MVP分析方法進行P波型態變異分析與設計最佳化的深度學習模型(MVPNet)來預測陣發性心房震顫（PAF）的發作。基於將AF發作前5分鐘與至少遠離AF發作 2小時的竇性心律訊號進行分類，從而達到預測心房震顫發作的效果。最新的研究表明，從標準neuECG訊號蒐集方法獲得的皮膚交感神經活動（SKNA）顯示心房震顫與交感神經活性有正相關。我們分析了8例陣發性心房震顫患者，並使用含交感神經訊號的寬頻ECG(neuECG)進行24小時長時間記錄，此SKNA標實驗數據蒐集於美國印第安納大學醫院。我們比較了此方法用於常規ECG和neuECG的分類效果。結果表明，此方法在8位受試者的寬帶ECG實驗中具有89%PAF預測準確率，並且此模型具有出色的靈敏度（92％）和良好的分辨率（85％）。MVP分析和深度學習相結合提供了一種優於傳統風險評估與預測的方法。此方法能夠預先預測PAF發作，提供更多的臨床治療反應時間進而達到預防缺血性中風的效果。

Atrial fibrillation is one of the most frequently asymptomatic arrhythmias but associated with stroke and significant morbidity and mortality. To identify risk stratification individually based on ECG analysis would be a substantial interest in the primary prevention of AF. We aimed to develop a lean and point-of-care identification of individuals at risk for developing AF. We analyzed 8 PAF patients under 24 hours wideband ECG recording with neuECG protocol at Indiana University Hospital. P-wave character in ECG is widely regarded as an important source of information in the diagnosis of atrial conduction pathology. The proposed morphological variant P-wave(MVP) analysis combines deep learning methods to predict the upcoming onset of paroxysmal atrial fibrillation (PAF). We aim to predict the upcoming PAF onset by developing an automatic system to handle these indicators to classify the sinus rhythm of far from and right before PAF onset. We applied MVP analysis with an efficient, lightweight deep learning method to investigate the 24hr wideband ECG recording. A recent study indicates that Skin Sympathetic Nerve Activity (SKNA) obtain form a standard neuECG protocol shows the sympathetic tone link to atrial fibrillation. We spread a comparison between ordinary ECG and neuECG, which improve to an outstanding outcome. The result demonstrated effectiveness in the overall PAF prediction 89% accuracy, with which the excellent sensitivity (92%) and functional specificity (85%) in general through wideband ECG. We conclude that the proposed MVP analysis combined with deep learning offers superior to existing risk assessment methods. The result conceivably presents the clinical scenario of real-time prediction to PAF episode in advance provides more treatment response time, potentially preventing ischemic stroke.

摘要 i
Abstract ii
Acknowledgment iii
Tables vi
Figures vii
Chapter 1 Introduction 1
1.1 Background 1
1.1.1 Atrial fibrillation characteristic 3
1.1.2 Risk factors and predictors 5
1.2 Related work 8
1.3 Research objective 10
Chapter 2 Methods 12
2.1 System overview diagram 12
2.2 Study population 14
2.3 Preprocessing 15
2.4 Morphological Variant P-wave Analysis 17
2.4.1 Morphological P-wave 18
2.4.2 Normalized P-wave Segmentation 21
2.4.3 Continues Wavelet Transform (CWT) 22
2.4.4 Spectral subtraction 25
2.5 Deep Learning Classification 27
2.5.1 Light Weight Convolutional Neural Network 28
2.5.2 Separable Convolution 29
2.5.3 MVPNet 31
Chapter 3 Results 34
3.1 ECG and neuECG Classification Accuracy 34
3.2 Inter-Subject Prediction Accuracy 35
3.3 Model Comparison 37
3.4 Gradient Activation Map 39
Chapter 4 Discussion 41
4.1 Compare with Other Methods 41
4.2 Explainable Artificial Intelligence 43
4.3 SKNA in neuECG 49
Chapter 5 Conclusion 50
Reference 53

[1] S. S.Chugh et al., "Worldwide epidemiology of atrial fibrillation: A global burden of disease 2010 study," Circulation, vol. 129, no. 8, pp. 837–847, 2014.
[2] G. Y. H.Lip et al., "Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: The Euro Heart Survey on atrial fibrillation," Chest, vol. 137, no. 2, pp. 263–272, 2010.
[3] M. H.Kim, S. S.Johnston, B. C.Chu, M. R.Dalal, andK. L.Schulman, "Estimation of total incremental health care costs in patients with atrial fibrillation in the united states," Circ. Cardiovasc. Qual. Outcomes, vol. 4, no. 3, pp. 313–320, 2011.
[4] A.Brandes, M. D.Smit, B. O.Nguyen, M.Rienstra, andI. C.VanGelder, "Risk factor management in atrial fibrillation," Arrhythmia Electrophysiol. Rev., vol. 7, no. 2, pp. 118–127, 2018.
[5] "Atrial Fibrillation | cdc.gov," Centers for Disease Control and Prevention. [Online]. Available: https://www.cdc.gov/heartdisease/atrial_fibrillation.htm. [Accessed: 13-Jul-2020].
[6] A.Narin, Y.Isler, M.Ozer, andM.Perc, "Early prediction of paroxysmal atrial fibrillation based on short-term heart rate variability," Phys. A Stat. Mech. its Appl., vol. 509, pp. 56–65, 2018.
[7] P. E.Dilaveris andJ. E.Gialafos, "Future concepts in P wave morphological analyses," Card. Electrophysiol. Rev., vol. 6, no. 3, pp. 221–224, 2002.
[8] Y. K.İçen, H. E.Sümbül, andM.Koç, "Increased P wave duration and dispersion is associated with catheter-related atrial fibrillation during electrophysiological study," Eur. Res. J., vol. 5, no. 4, pp. 691–696, 2018.
[9] A.Martínez, R.Alcaraz, andJ. J.Rieta, "Morphological variability of the P-wave for premature envision of paroxysmal atrial fibrillation events," Physiol. Meas., vol. 35, no. 1, pp. 1–14, 2014.
[10] S.Pezzuto et al., "Beat-to-beat P-wave morphological variability in patients with paroxysmal atrial fibrillation: An in silico study," Europace, vol. 20, no. November, pp. III26–III35, 2018.
[11] A. P.Persson et al., "Heart rate and premature atrial contractions at 24hECG independently predict atrial fibrillation in a population-based study," heart, pp. 287–291, 2019.
[12] L.Clavier, J. M.Boucher, R.Lepage, J. J.Blanc, andJ. C.Cornily, "Automatic P-wave analysis of patients prone to atrial fibrillation," Med. Biol. Eng. Comput., vol. 40, no. 1, pp. 63–71, 2002.
[13] R.Alcaraz, A.Martínez, andJ. J.Rieta, "Role of the P-wave high frequency energy and duration as noninvasive cardiovascular predictors of paroxysmal atrial fibrillation," Comput. Methods Programs Biomed., vol. 119, no. 2, pp. 110–119, 2015.
[14] P. S.Chen, L. S.Chen, M. C.Fishbein, S. F.Lin, andS.Nattel, "Role of the autonomic nervous system in atrial fibrillation: Pathophysiology and therapy," Circ. Res., vol. 114, no. 9, pp. 1500–1515, 2014.
[15] P. E.Dilaveris andJ. E.Gialafos, "P-wave dispersion: A novel predictor of paroxysmal atrial fibrillation," Annals of Noninvasive Electrocardiology, vol. 6, no. 2. pp. 159–165, 2001.
[16] F.Censi et al., "P-wave Variability and Atrial Fibrillation," Sci. Rep., vol. 6, pp. 1–7, 2016.
[17] A.Doytchinova et al., "Simultaneous noninvasive recording of skin sympathetic nerve activity and electrocardiogram," Hear. Rhythm, vol. 14, no. 1, pp. 25–33, 2017.
[18] T.Kusayama et al., "Simultaneous noninvasive recording of electrocardiogram and skin sympathetic nerve activity (neuECG)," Nat. Protoc., vol. 15, no. 5, pp. 1853–1877, 2020.
[19] A.Uradu et al., "Skin sympathetic nerve activity precedes the onset and termination of paroxysmal atrial tachycardia and fibrillation," Hear. Rhythm, vol. 14, no. 7, pp. 964–971, Jul.2017.
[20] D. M.German, M. M.Kabir, T. A.Dewland, C. A.Henrikson, andL. G.Tereshchenko, "Atrial Fibrillation Predictors: Importance of the Electrocardiogram," Ann. Noninvasive Electrocardiol., vol. 21, no. 1, pp. 20–29, 2016.
[21] D.Filos, I.Chouvarda, D.Tachmatzidis, V.Vassilikos, andN.Maglaveras, "Beat-to-beat P-wave morphology as a predictor of paroxysmal atrial fibrillation," Comput. Methods Programs Biomed., vol. 151, pp. 111–121, 2017.
[22] A. E.Buxton andM. E.Josephson, "The role of P wave duration as a predictor of postoperative atrial arrhythmias," Chest, vol. 80, no. 1, pp. 68–73, 1981.
[23] O. A.Centurión et al., "P-Wave Morphology, Amplitude, Duration and Dispersion in Atrial Arrhythmias," J. Cardiol. Ther., vol. 6, pp. 1–6, 2018.
[24] Q.Li et al., "Ventricular ectopic beat detection using a wavelet transform and a convolutional neural network," Physiol. Meas., vol. 40, no. 5, 2019.
[25] C.Hao, S.Wibowo, M.Majmudar, andK. S.Rajput, "Spectro-Temporal Feature Based Multi-Channel Convolutional Neural Network for ECG Beat Classification," pp. 5642–5645, 2019.
[26] S.Mousavi, F.Afghah, A.Razi, andU. R.Acharya, "ECGNET: Learning where to attend for detection of atrial fibrillation with deep visual attention," 2018.
[27] M.Limam andF.Precioso, "Atrial fibrillation detection and ECG classification based on convolutional recurrent neural network," Comput. Cardiol. (2010)., vol. 44, pp. 1–4, 2017.
[28] S.Ross-Howe andH. R.Tizhoosh, "Atrial Fibrillation Detection Using Deep Features and Convolutional Networks," no. Bhi, pp. 1–4, 2019.
[29] A. M. L. S.M. Christopher, "Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network," Nat. Med., vol. 25, no. 1, pp. 65–69, Jan.2019.
[30] Y.Xia, N.Wulan, K.Wang, andH.Zhang, "Detecting atrial fibrillation by deep convolutional neural networks," Comput. Biol. Med., vol. 93, no. December 2017, pp. 84–92, 2018.
[31] A.Narin, Y.Isler, M.Ozer, andM.Perc, "Early prediction of paroxysmal atrial fibrillation based on short-term heart rate variability," Phys. A Stat. Mech. its Appl., vol. 509, pp. 56–65, 2018.
[32] Z.Li, H.Derksen, J.Gryak, H.Ghanbari, P.Gunaratne, andK.Najarian, "A Novel Atrial Fibrillation Prediction Algorithm Applicable to Recordings from Portable Devices," Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. EMBS, vol. 2018-July, pp. 4034–4037, 2018.
[33] S.Mousavi, F.Afghah, A.Razi, andU. R.Acharya, "Ecgnet: Learning where to attend for detection of atrial fibrillation with deep visual attention," 2019 IEEE EMBS Int. Conf. Biomed. Heal. Informatics, BHI 2019 - Proc., 2019.
[34] F.González, R.Alcaraz, andJ. J.Rieta, "The physionet QT database: Study on the reliability of P-wave manual annotations under noisy recordings," Comput. Cardiol. (2010)., vol. 44, no. May 2020, pp. 1–4, 2017.
[35] S. P.Shashikumar, G. D.Clifford, A. J.Shah, andS.Nemati, "Detection of paroxysmal atrial fibrillation using attention-based bidirectional recurrent neural networks," Proc. ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., pp. 715–723, 2018.
[36] C.Liu, J.Wong, A.Doytchinova, P. S.Chen, andS. F.Lin, "Method for Detection and Quantification of Noninvasive Skin Sympathetic Nerve Activity," 2018 Int. Conf. Syst. Sci. Eng. ICSSE 2018, no. 1, pp. 1–4, 2018.
[37] M. J.Burke andM.Nasor, "Wavelet based analysis and characterization of the ECG signal," J. Med. Eng. Technol., vol. 28, no. 2, pp. 47–55, 2004.
[38] J.Qin andP.Sun, "Applications and Comparison of Continuous Wavelet Transforms on Analysis of A-wave Impulse Noise," Arch. Acoust., vol. 40, no. 4, pp. 503–512, 2015.
[39] M. J.Burke andM.Nasor, "ECG analysis using the Mexican-hat wavelet," no. 1, pp. 1–6, 2001.
[40] R. M.Rao andJ. O.Chapa, "Algorithms for designing wavelets to match a specified signal," IEEE Trans. Signal Process., vol. 48, no. 12, pp. 3395–3406, 2000.
[41] R.Martin, "Spectral subtraction based on minimum statistics," Power, vol. 6, no. 1, p. 8, 1994.
[42] Z. I.Attia et al., "An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction," Lancet, vol. 394, no. 10201, pp. 861–867, 2019.
[43] O.Faust, Y.Hagiwara, T. J.Hong, O. S.Lih, andU. R.Acharya, "Deep learning for healthcare applications based on physiological signals: A review," Comput. Methods Programs Biomed., vol. 161, no. April, pp. 1–13, 2018.
[44] A. G.Howard et al., "MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications," 2017.
[45] Z.Wang, J.Zhu, T.Yan, andL.Yang, "A new modified wavelet-based ECG denoising," Comput. Assist. Surg., vol. 24, no. sup1, pp. 174–183, 2019.
[46] B.Pourbabaee, M. J.Roshtkhari, andK.Khorasani, "Deep Convolutional Neural Networks and Learning ECG Features for Screening Paroxysmal Atrial Fibrillation Patients," IEEE Trans. Syst. Man, Cybern. Syst., vol. 48, no. 12, pp. 2095–2104, 2018.
[47] K.Simonyan andA.Zisserman, "Very deep convolutional networks for large-scale image recognition," 3rd Int. Conf. Learn. Represent. ICLR 2015 - Conf. Track Proc., pp. 1–14, 2015.
[48] K.He, X.Zhang, S.Ren, andJ.Sun, "Deep residual learning for image recognition," Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 2016-Decem, pp. 770–778, 2016.
[49] S.Ioffe andC.Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," 32nd Int. Conf. Mach. Learn. ICML 2015, vol. 1, pp. 448–456, 2015.
[50] R. R.Selvaraju, M.Cogswell, A.Das, R.Vedantam, D.Parikh, andD.Batra, "Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization," Int. J. Comput. Vis., vol. 128, no. 2, pp. 336–359, 2020.
[51] K. H.Boon, M.Khalil-Hani, M. B.Malarvili, andC. W.Sia, "Paroxysmal atrial fibrillation prediction method with shorter HRV sequences," Comput. Methods Programs Biomed., vol. 134, pp. 187–196, 2016.
[52] J.Cho, Y.Kim, andM.Lee, "Prediction to Atrial Fibrillation Using Deep Convolutional Neural Networks," 2018, pp. 164–171.
[53] A. C. C.Yang andH. W.Yin, "Prediction of paroxysmal atrial fibrillation by footprint analysis," Comput. Cardiol., pp. 401–404, 2001.
[54] B.Hickey andC.Heneghan, "Screening for paroxysmal atrial fibrillation using atrial premature contractions and spectral measures," Comput. Cardiol., vol. 29, no. 1, pp. 217–220, 2002.
[55] A.Barredo Arrieta et al., "Explainable Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI," Inf. Fusion, vol. 58, pp. 82–115, 2020.