急性心肌梗塞（Acute myocardial infarction，AMI），是一種嚴重的心臟狀態，梗塞的主要原因是因為部分心肌的血液循環突然中斷，造成心肌無法獲得足夠的氧氣及營養，嚴重時會引起心肌壞死而影響心臟功能，甚至危及患者的生命，心肌梗塞通常是冠狀動脈疾病所引起，造成動脈粥樣硬化斑塊破裂引起血管阻塞，有些檢查有助於心肌梗塞的診斷，像是心電圖（ECG）、驗血、冠狀動脈造影等，其中心電圖記錄心臟跳動時心肌的電位變化，透過觀察電位的異常變化可偵測不同型態的心肌梗塞。
本研究應用人工智慧影像分析技術對心電圖進行分析以偵測心肌梗塞的發生，影像分析技術的核心是基於殘差網路（ResNet）發展，並且針對心電圖的影像特性改良原始架構中的殘差區塊，在改良的方法中通過Xavier初始化來平衡所有卷積濾波器的梯度尺度的大小，權值進行隨機初始化，同時使用正則化（batch normalization）來減小過擬合問題的影響，並且調整一般將正則化放在卷積層後與激活函數前位置做法，改將正則化延遲到激活函數後執行。

本研究使用的是12導程ECG影像，99張正常，99張發生急性心肌梗塞，並且通過前處理(灰階/裁切)及10組交叉驗證後套入模型訓練，最終結果顯示Test準確度為100%，本研究改良後的ResNet在訓練時可以得到更低的error，透過Heatmap證實AMI發生前的訊號偵測更精準，並且通過自訂評估函數挑選訓練較佳的Model，再將挑選後的Model搭配Flask運用在實際平台上。

臨床上，對急性心肌梗塞的診斷需要通過病史、心電圖、驗血才有辦法確診，醫師的經驗對於正確判讀扮演非常關鍵的角色，但是急性心肌梗塞通常都發生的很突然，因此不一定有機會做完整的診斷，本研究通過人工智慧技術分析心電圖的方式，可以及時評估是否發生急性心肌梗塞，幫助醫師更及時的做出臨床診斷給予患者適當的治療。

Acute myocardial infarction (AMI) is anacute and severe state of the heart. The main cause of infarction is the suddeninterruption of blood circulation in the myocardia, disabling them to obtainsufficient oxygen and nutrition.
In severe cases, AMI may affect the heart functiondue to myocardial necrosis, and even endanger the lives of patients.
One of the most common causes of AMI is coronary artery diseases, in which atheroscleroticplaque rupture leads to blockage of some blood vessels. Many examinations help to diagnose AMI, such as electrocardiograms (ECG), blood tests,coronary angiography, etc. Electrocardiograms record the electric potentials of the heart and can identify different types of AMI by observing the abnormal changes.
This study uses artificial intelligence image analysis techniques to analyze the electrocardiograms for the detection of the occurrence of AMI. The core of this image analysis technique is based on the development of the residual networks (ResNet). Also, the residual blocks in the original model architecture are improved based on the image characteristics of the electrocardiograms.
12-lead electrocardiogram images were analyzed in this study, 99 of which were normal and 99 had AMI. After preprocessing and 10-fold cross-validation, model training was performed. The final results showed that the test set accuracy was 100%.
The improved ResNet designed in this study also achieved lower error during training. Clinically, the patient’s medical history, electrocardiogram,and blood test are required to confirm the diagnosis of AMI.
The experience of the physician plays a key role in the correct interpretation of clinical data.However, AMI usually occurs suddenly, causing a lack of time and resource for a complete diagnosis.
By analyzing electrocardiograms with artificial intelligence techniques, this study provides a way to predict the occurrence of AMI efficiently,and help physicians make clinical diagnoses and give patients appropriate treatment promptly.

致謝辭 i
中文摘要 ii
Abstract iv
目錄 vi
圖目錄 vii
表目錄 ix
第一章 緒論 1
1.1 研究動機與目的 1
1.2 論文架構 2
第二章 相關研究與背景 3
2.1卷積神經網路(Convolutional Neural Network) 7
2.2 殘差神經網路(Residual Network) 8
2.2.1關於殘差模塊(residual block) 11
2.2.2 恆等對映(Identity Mapping) 改進 13
2.3二維卷積神經網絡進行心電圖心律不整分類 16
2.3.1 資料預處理 16
2.3.2 圖像類別 17
2.3.3 分類器模型選擇 18
2.4 熱力圖 23
第三章 系統架構 24
3.1 流程架構圖 24
3.1.1影像前處理 26
3.1.2深度學習模型 28
3.1.3交叉驗證處理 32
第四章 結果與討論 34
4.1 實驗結果 34
4.2 結果討論 38
4.3 研究環境 38
第五章 結論與未來展望 39
參考文獻 40

[1] Anderson JL, Adams CD, Antman EM, Bridges CR, Califf RM, Casey DE, “2012 ACCF/AHA focused update incorporated into the ACCF/AHA 2007 guidelines for the management of patients with unstable angina/non-STelevation myocardial infarction: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines.” J Am Coll Cardiol, 2013;
[2] Zhang B, Zhang W, Huang RC,” Gender and age differences associated with prehospital delay in Chinese patients presenting with ST elevation myocardial infarction.” Journal of Cardiovascular Nursing, 2014;
[3] Wu JR, Moser DK, Riegel B, McKinley S, Doering LV. “Impact of prehospital delay in treatment seeking on in-hospital complications after acute myocardial infarction. Journal of Cardiovascular Nursing.” Journal of Cardiovascular Nursing. 2011;
[4] Nawar EW, Niska RW, Xu J. “National Hospital Ambulatory Medical Care Survey: 2005 emergency department summary.” Adv Data, 2007;
[5] Thygesen K, Alpert JS, White HD, Jaffe AS, Apple FS, Galvani M,”Universal definition of myocardial infarction. “ Eur Heart J, 2007;
[6] Roffi M, Patrono C, Collet J-P, Mueller C, Valgimigli M, Andreotti F, “ 2015 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. “Eur Heart J, 2016;
[7] ECG Learning Center. Eccles Health Sciences Library University of Utah.
[8] 陳文彬、潘祥林，診斷學（第8版）120、486、487頁，人民衛生出版社。ISBN 978-7-117-17077-2
[9] See images of ECG electrodes here 網際網路檔案館的存檔，存檔日期2011-04-05.or here lead_dia. Library.med.utah.edu. [2009-08-15].
[10] AHA Diagnostic ECG Electrode Placement 網際網路檔案館的存檔，存檔日期2009-11-22.
[11] Yan LeCun, "Gradient-based learning applied to document recognition." Proceedings of the IEEE, 1988;
[12] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv, 2014;
[13] K. He, X. Zhang, S. Ren, and J. Sun. “Deep residual learning for image recognition,” Proceedings of the IEEE conference on computer vision and pattern recognition, 2016;
[14] K. He, X. Zhang, S. Ren, and J. Sun, “Identity mappings in deep residual networks,” European Conference on Computer Vision, 2016;
[15] T. J. Jun, H. M. Nguyen, D. Kang, D. Kim, D. Kim, Y.-H. Kim, "ECG arrhythmia classification using a 2-D convolutional neural network" arXiv, 2018;
[16] Bolei Zhou, Aditya Khosla, Àgata Lapedriza, Aude Oliva, and Antonio Torralba. “ILearning deep features for discriminative localizationi.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016;