心房顫動(房颤)是最常見的持續性心率失常，其源於異常不規則的心房電表現，無法正常排空心房內血液，因而可能導致心臟血栓、中風、動脈疾病、血栓栓塞和心衰竭。大多数人在房颤發生前就有房室心肌病，其常見的症狀為左右心房的擴大。若能有效控制房室心肌病，可防止其發展為房顫。
常規的心電圖可以顯示心跳的變化，但缺點是敏感性不佳。透過讓病人運動可以使心臟的工作量急劇增加，使得心電圖中的P-QRS-T波變化比起常規靜態心電圖具有更高的診斷敏感度。其中對於心房性心肌病和陣發性心房顫動的檢測而言，運動過程中的P波變化具有很高的診斷敏感度。
因此，我們提出了一套深度學習系統。將病人的運動心電圖輸入我們的深度學習系統來分類病人是否有心房擴大及心房顫動的病徵。此系統主要分為兩大部分。第一部分利用卷積循環神經網路的深度學習架構，將病人的心電圖輸入以找出P波的位置。第二部分則是利用卷積循環神經網路找到的P波，將其計算完相關參數後輸入平行雙向長短期記憶網路的深度學習架構。此架構可以同步分析不同階段的P波參數，最終分類出病人的病徵。透過我們提出的深度學習系統，可以有效的同時分析病人在不同階段的變化。即使每位病人的資料長度以及各個階段的資料長度皆不同，我們依然可以全面分析病人所有的心電圖資料。這讓我們的架構得到更準確的效果。此外我們的系統不論是常規心電圖或是運動心電圖都可以使用，甚至還可以接受含有常規心電圖及運動心電圖的混合資料。另外我們還證實了含有越多的運動心電圖資料可以讓模型更有效的診斷出患病的病人。

Atrial fibrillation (AF) is the most common type of sustained arrhythmia. AF results from abnormal irregularities in the electrical performance of the atria, which do not empty the atria properly and may lead to heart thrombosis, stroke, arterial disease, thromboembolism, and heart failure. Most people have atrial cardiomyopathy before the onset of atrial fibrillation, and the common symptom is the atrial enlargement of the right and left atria. Atrial cardiomyopathy can be prevented from progressing to atrial fibrillation if it is effectively managed.
A regular electrocardiogram (ECG) can show changes in the heartbeat but has the disadvantage of not being sensitive. By making the patient exercise, a dramatic increase in cardiac workload can be achieved. It makes the P-QRS-T wave change in the exercise ECG more diagnostically sensitive than in a regular ECG. For the detection of atrial cardiomyopathy and paroxysmal atrial fibrillation, P-wave changes during exercise have a high diagnostic sensitivity.
Therefore, we propose a deep learning system. The patient's exercise ECG is entered into our deep learning system to classify whether the patient has features of atrial enlargement and atrial fibrillation. The system is divided into two main parts. The first part uses the deep learning architecture of the Convolutional Recurrent Neural Network (CRNN) to identify the the P-waves location in the patient's ECG. The second part uses the P-wave found by CRNN to calculate the relevant parameters and input to the Parallel Bi-directional Long Short-Term Memory Network (PBLSTM) deep learning architecture. This architecture can analyze the P-wave parameters of different stages simultaneously and finally classify the patient's disease. With our proposed deep learning system, we can effectively analyze the changes of patients in different stages simultaneously. Even though the data length of each patient and the data length of each stage are different, we can still analyze all the data of the patient completely. This allows the proposed system to get more stable and accurate results. In addition, our system can be used for both routine and exercise ECGs as input. It can even accept a mix of regular and exercise ECG data. In addition, we proved that the more exercise ECG data we have, the more effective the model can be in diagnosing patients with the disease.

Verification Letter i
摘要 ii
Abstract iii
List of Figures vi
List of Tables vii
Chapter 1. Introduction 1
Chapter 2. Related works 5
Chapter 3. System Components 8
3.1. Convolutional Neural Network (CNN) 8
3.2. Long Short-Term Memory (LSTM) 9
3.3. Bi-directional Long Short-Term Memory Network (BLSTM) 11
3.4. Convolutional Recurrent Neural Networks (CRNN) 12
3.5. Attention Mechanism (Attention) 13
Chapter 4. Material and Methods 14
4.1. Exercise ECG 15
4.2. Data preprocessing 16
4.2.1. Z-Score Normalization 16
4.2.2. Baseline Adjustment 17
4.3. Find out the P-wave location 18
4.4. Calculating P-wave parameters 21
4.4.1. Peak voltage of P-wave 22
4.4.2. P-wave duration 22
4.4.3. P-front integration 22
4.4.4. P-rear integration 23
4.5. Exercise ECG signal correlation processing 23
4.6. Parallel Bi-directional Long Short-Term Memory Network (PBLSTM) 26
Chapter 5. Experimental Results 29
5.1. Experimental settings and classification distribution 29
5.2. Comparison of the three phases of data input 31
5.3. Comparison of mixing different percentages of regular ECG and exercise ECG data. 33
5.4. Comparison of results 36
Chapter 6. Discussion 40
6.1. Dealing with data imbalance 40
6.2. Comparison of different data forms of exercise ECG and corresponding models 42
6.2.1. Mean value of the parameters 43
6.2.2. All parameters 44
6.2.3. P-waves with imbalance methods 45
6.2.4. All waves with imbalance methods 48
Chapter 7. Concluding Remarks 51
Reference 53

[1] World Health Organization, and Research for International Tobacco Control. WHO report on the global tobacco epidemic, 2008: the MPOWER package. World Health Organization, 2008.
[2] Rahman, Faisal, Gene F. Kwan, and Emelia J. Benjamin. "Global epidemiology of atrial fibrillation." Nature Reviews Cardiology 11.11 (2014): 639-654.
[3] Chugh, Sumeet S., et al. "Worldwide epidemiology of atrial fibrillation: a Global Burden of Disease 2010 Study." Circulation 129.8 (2014): 837-847.
[4] Ng, Kuan H., Robert G. Hart, and John W. Eikelboom. "Anticoagulation in patients aged≥ 75 years with atrial fibrillation: role of novel oral anticoagulants." Cardiology and therapy 2.2 (2013): 135-149.
[5] Fragakis, Nikolaos, Gabriele Vicedomini, and Carlo Pappone. "Endurance sport activity and risk of atrial fibrillation–epidemiology, proposed mechanisms and management." Arrhythmia & electrophysiology review 3.1 (2014): 15.
[6] Pelliccia, Antonio, et al. "Prevalence and clinical significance of left atrial remodeling in competitive athletes." Journal of the American College of Cardiology 46.4 (2005): 690-696.
[7] Mohanty, Sanghamitra, et al. "Long‐term outcome of catheter ablation in atrial fibrillation patients with coexistent metabolic syndrome and obstructive sleep apnea: impact of repeat procedures versus lifestyle changes." Journal of cardiovascular electrophysiology 25.9 (2014): 930-938.
[8] Bureekam, Sangrawee, and Thananya Boonyasirinant. "Accuracy of left atrial enlargement diagnosed by electrocardiography as compared to cardiac magnetic resonance in hypertensive patients." J Med Assoc Thai 97.Suppl 3 (2014): S132-S138.
[9] Andlauer, Robin, et al. "Influence of left atrial size on P-wave morphology: differential effects of dilation and hypertrophy." EP Europace 20.suppl_3 (2018): iii36-iii44.
[10] Simonson, Ernst. "Effect of moderate exercise on the electrocardiogram in healthy young and middle-aged men." Journal of applied physiology 5.10 (1953): 584-588.
[11] Irisawa, Hiroshi, and Issei Seyama. "The configuration of the P wave during mild exercise." American heart journal 71.4 (1966): 467-472.
[12] YokOTA, Mitsuhiro, et al. "Analysis of the exercise-induced orthogonal P wave changes in normal subjects and patients with coronary artery disease." Japanese heart journal 27.4 (1986): 443-460.
[13] Andrikopoulos, George K., et al. "Increased variance of P wave duration on the electrocardiogram distinguishes patients with idiopathic paroxysmal atrial fibrillation." Pacing and Clinical Electrophysiology 23.7 (2000): 1127-1132.
[14] Goette, Andreas, et al. "EHRA/HRS/APHRS/SOLAECE expert consensus on atrial cardiomyopathies: definition, characterization, and clinical implication." Ep Europace 18.10 (2016): 1455-1490.
[15] Guichard, Jean-Baptiste, and Stanley Nattel. "Atrial cardiomyopathy: a useful notion in cardiac disease management or a passing fad?." Journal of the American College of Cardiology 70.6 (2017): 756-765.
[16] Sharma, Sanjay, Ahmed Merghani, and Lluis Mont. "Exercise and the heart: the good, the bad, and the ugly." European heart journal 36.23 (2015): 1445-1453.
[17] Attia, Zachi I., et al. "Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram." Nature medicine 25.1 (2019): 70-74.
[18] Ying, Jun, et al. "Classification of exacerbation frequency in the COPDGene cohort using deep learning with deep belief networks." IEEE journal of biomedical and health informatics 24.6 (2016): 1805-1813.
[19] Kido, Shoji, Yasusi Hirano, and Noriaki Hashimoto. "Detection and classification of lung abnormalities by use of convolutional neural network (CNN) and regions with CNN features (R-CNN)." 2018 International workshop on advanced image technology (IWAIT). IEEE, 2018.
[20] Shin, Hoo-Chang, et al. "Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning." IEEE transactions on medical imaging 35.5 (2016): 1285-1298.
[21] Liang, Fei, Cong Shen, and Feng Wu. "An iterative BP-CNN architecture for channel decoding." IEEE Journal of Selected Topics in Signal Processing 12.1 (2018): 144-159.
[22] Cui, Zongyong, et al. "SAR unlabeled target recognition based on updating CNN with assistant decision." IEEE Geoscience and Remote Sensing Letters 15.10 (2018): 1585-1589.
[23] Yu, Yang, et al. "An unsupervised convolutional feature fusion network for deep representation of remote sensing images." IEEE Geoscience and Remote Sensing Letters 15.1 (2017): 23-27.
[24] Scarpa, Giuseppe, Sergio Vitale, and Davide Cozzolino. "Target-adaptive CNN-based pansharpening." IEEE Transactions on Geoscience and Remote Sensing 56.9 (2018): 5443-5457.
[25] Gupta, Harshit, et al. "CNN-based projected gradient descent for consistent CT image reconstruction." IEEE transactions on medical imaging 37.6 (2018): 1440-1453.
[26] Porumb, Mihaela, et al. "A convolutional neural network approach to detect congestive heart failure." Biomedical Signal Processing and Control 55 (2020): 101597.
[27] Kiranyaz, Serkan, et al. "1-d convolutional neural networks for signal processing applications." ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019.
[28] Kiranyaz, Serkan, et al. "1D convolutional neural networks and applications: A survey." Mechanical systems and signal processing 151 (2021): 107398.
[29] Lu, Peng, et al. "ECG Classification Based on Long Short-Term Memory Networks." The International Conference on Healthcare Science and Engineering. Springer, Singapore, 2018.
[30] Zihlmann, Martin, Dmytro Perekrestenko, and Michael Tschannen. "Convolutional recurrent neural networks for electrocardiogram classification." 2017 Computing in Cardiology (CinC). IEEE, 2017.
[31] Gamboa, John Cristian Borges. "Deep learning for time-series analysis." arXiv preprint arXiv:1701.01887 (2017).
[32] Attia, Zachi I., 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." The Lancet 394.10201 (2019): 861-867.
[33] Rajpurkar, Pranav, et al. "Cardiologist-level arrhythmia detection with convolutional neural networks." arXiv preprint arXiv:1707.01836 (2017).
[34] Yildirim, Özal. "A novel wavelet sequence based on deep bidirectional LSTM network model for ECG signal classification." Computers in biology and medicine 96 (2018): 189-202.
[35] Attia, Zachi I., 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." The Lancet 394.10201 (2019): 861-867.
[36] Christopoulos, Georgios, et al. "Artificial Intelligence–Electrocardiography to Predict Incident Atrial Fibrillation: A Population-Based Study." Circulation: Arrhythmia and Electrophysiology 13.12 (2020): e009355.
[37] Abrishami, Hedayat, et al. "P-QRS-T localization in ECG using deep learning." 2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI). IEEE, 2018.
[38] Jo, Yong-Yeon, et al. "Explainable artificial intelligence to detect atrial fibrillation using electrocardiogram." International Journal of Cardiology 328 (2021): 104-110.
[39] Schmidhuber, Jürgen, and Sepp Hochreiter. "Long short-term memory." Neural Comput 9.8 (1997): 1735-1780.
[40] Yang, Zichao, et al. "Hierarchical attention networks for document classification." Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: human language technologies. 2016.
[41] Ahmad, Masood, and C. Gunnar Blomqvist. "P wave changes during exercise in normal subjects and in patients with hemodynamic evidence of left atrial overload." Journal of electrocardiology 11.4 (1978): 361-368.
[42] Keras : https://keras.io
[43] Tensor-Flow : https://www.tensorflow.org
[44] Prechelt, Lutz. "Early stopping-but when?." Neural Networks: Tricks of the trade. Springer, Berlin, Heidelberg, 1998. 55-69.
[45] Mei, Qianwen, and Tangyou Liu. "ECG diagnosis based on one-dimensional convolutional neural network." IOP Conference Series: Materials Science and Engineering. Vol. 768. No. 7. IOP Publishing, 2020.
[46] He, Runnan, et al. "Automatic cardiac arrhythmia classification using combination of deep residual network and bidirectional LSTM." IEEE Access 7 (2019): 102119-102135.
[47] Ribeiro, Antônio H., et al. "Automatic diagnosis of the 12-lead ECG using a deep neural network." Nature communications 11.1 (2020): 1-9.