本研究針對台灣慢性腎臟病（CKD）及腎臟病末期（ESRD）患者的高發病率，特別關注伴隨的心血管疾病風險，旨在利用心電圖數據判斷是否為血液透析患者，並探索腎病與心臟健康之間的相關性。研究資料來自高雄醫學大學附設中和紀念醫院（KMUH），陽性資料集包含 5187 筆血液透析患者的心電圖，陰性資料集包含 10733 筆健康檢查者的心電圖。為平衡資料，我們使用降採樣技術將陰性樣本數量降至與陽性樣本相同。本研究開發了一種創新的心電圖影像數位化方法，不僅提高了數據處理的效率和準確性，也使得深度神經網路能夠更精確地識別與腎病相關的心電圖特徵。我們利用深度神經網路技術來學習並分析這些心電圖數據，以期在非侵入性情況下，準確識別腎病末期患者，從而及時提醒患者進行腎臟科回診或開始血液透析，並為處於腎病五期的患者提供早期預警。研究結果顯示，通過深度神經網路模型，我們的系統在診斷腎病末期方面表現優異，準確度達到 97%，且正確陽性率（PPV）和陰性預測值（NPV）也有出色的表現。本研究不僅為心腎相關性研究提供了新的見解，還展現了心電圖作為早期診斷工具的臨床應用潛力。未來，我們計劃進一步優化模型，以實現腎病患者的早期診斷和管理，改善患者的生存質量。

This study focuses on the high prevalence of chronic kidney disease (CKD) and end-stage renal disease (ESRD) in Taiwan, particularly their link to cardiovascular risks. We aimed to use electrocardiogram (ECG) data to identify hemodialysis patients and explore the connection between kidney disease and heart health. Data from Kaohsiung Medical University Hospital (KMUH) included 5,187 ECGs from hemodialysis patients and 10,733 from healthy individuals. We balanced the datasets by down-sampling the negative samples to match the positive ones. A novel ECG digitization method was developed, enhancing data processing efficiency and allowing deep neural networks to better identify kidney disease-related features. Our deep learning model accurately diagnosed ESRD with an accuracy of 97%, showing strong predictive values. This study highlights the potential of ECGs as an early diagnostic tool for kidney disease, and we plan to further optimize the model to improve early diagnosis and management.

目錄：
論文審定書i
誌謝ii
摘要iii
Abstract iv
圖次 vii
表次 viii
第一章 介紹1
1.1 研究背景1
1.1.1 慢性腎臟病 (Chronic Kidney Disease, CKD) 1
1.1.2 心臟與血液透析2
1.2 研究動機與目的3
1.3 論文架構5
第二章 文獻探討6
2.1 心電圖數位化6
2.1.1 心電圖轉換一維訊號6
2.1.2 離散小波轉換7
2.2 深度學習基於 EKG 的疾病預測7
2.3 深度學習神經網路架構8
2.3.1 LSTM (Long-Short Term Memory) 8
2.3.2 殘差網路模型（ResNet）10
2.3.3 壓縮與激勵神經網路（SENet）11
第三章 研究方法12
3.1 資料收集12
3.2 資料預處理13
3.2.1 輸入心電圖類別偵測13
3.2.2 心電圖數位化15
3.2.3 結果比較23
3.3 深度神經網路架構25
3.3.1 基於 LSTM 的一維心電訊號預測模型25
3.3.2 1D-ResSENet (Residual Squeeze-and-Excitation Network) 27
第四章 實驗結果與分析30
4.1 實驗設置與評估指標30
4.2 資料平衡32
4.3 超參數設置33
4.4 模型的表現差異34
4.5 模型在各年齡層的表現38
4.6 模型在不同性別的表現40
4.7 方法比較 42
第五章 結論 44
參考文獻46
圖次
圖1-1 12導程心電圖4
圖2-1舊數位化系統效果與實際心電圖對照7
圖2-2 LSTM單元結構示意圖8
圖3-1本研究實驗流程12
圖3-2不同類別的心電圖示意圖13
圖3-3背景移除後的心電圖14
圖3-4數位化流程示意圖15
圖3-5此二類心電圖導程時長皆為48秒16
圖3-6不同心電圖之電壓比例尺16
圖3-7軌跡追蹤示意圖17
圖3-8 𝐸𝐾𝐺1𝐷生成流程21
圖3-9將𝐸𝐾𝐺1𝐷進行9層小波分解後的心電圖分量22
圖3-10校正基線飄移與濾除噪聲之前後比較範例23
圖3-12數位化結果比較24
圖3-13本論文解決S波失真的具體流程圖24
圖3-14 LSTM模型架構圖25
圖3-15 1D-ResSENet模型架構圖27
圖4-1混淆矩陣30
圖4-2 LSTM模型準確度與損失曲線35
圖4-3 LSTM模型測試混淆矩陣35
圖4-4 LSTM模型的接收者操作特徵曲線(ROCcurve)36
圖4-5 1D-ResSENet模型準確度與損失曲線37
圖4-6 1D-ResSENet模型測試混淆矩陣37
圖4-7 1D-ResSENet模型的接收者操作特徵曲線(ROCcurve)38
圖4-8年齡分群ROC曲線38
圖4-9年齡分群ROC曲線39
圖4-10性別分群ROC曲線41
圖4-11性別分群ROC曲線41
圖4-12比較相關模型的混淆矩陣（4生理參數輸入）43
表次
表3-1 LSTM模型的詳細架構26
表3-2 1D-ResSENet模型詳細架構29
表4-1訓練使用的超參數33
表4-2 LSTM Model Performance on Different Patient Info(%)34
表4-3 1D-ResSENet Model Performance on Different Patient Info(%)36
表4-4 2Inputs Model Age Distribution(20span)39
表4-5 4Inputs Model Age Distribution(20span)40
表4-6 2Input Model Gender Distribution41
表4-7 4Input Model Gender Distribution42
表4-8各模型方法比較43

[1] G. Eknoyan, N. Lameire, K. Eckardt, B. Kasiske, D. Wheeler, A. Levin, P. Stevens,
R. Bilous, E. Lamb, J. Coresh, et al., “KDIGO 2012 clinical practice guideline for
the evaluation and management of chronic kidney disease,” Kidney Int, vol. 3, no. 1,
pp. 5–14, 2013.
[2] J. Massagué, “Receptors for the TGF-𝛽family,” Cell, vol. 69, no. 7, pp. 1067–1070,
1992.
[3] G. C. Blobe, W. P. Schiemann, and H. F. Lodish, “Role of transforming growth factor
𝛽in human disease,” New England Journal of Medicine, vol. 342, no. 18, pp. 1350–
1358, 2000.
[4] W. G. Couser, G. Remuzzi, S. Mendis, and M. Tonelli, “The contribution of chronic
kidney disease to the global burden of major noncommunicable diseases,” Kidney
International, vol. 80, no. 12, pp. 1258–1270, 2011.
[5] P. W. Eggers, “Has the incidence of end-stage renal disease in the USA and other
countries stabilized?,” Current Opinion in Nephrology and Hypertension, vol. 20,
no. 3, pp. 241–245, 2011.
[6] M. R. Weir and V. J. Dzau, “The renin-angiotensin-aldosterone system: a specific
target for hypertension management,” American Journal of Hypertension, vol. 12,
no. S9, pp. 205S–213S, 1999.
[7] A. K. Bidani and K. A. Griffin, “Pathophysiology of hypertensive renal damage:
implications for therapy,” Hypertension, vol. 44, no. 5, pp. 595–601, 2004.
[8] C. Mogensen, “Microalbuminuria predicts clinical proteinuria and early mortality
in maturity-onset diabetes,” New England Journal of Medicine, vol. 310, no. 6,
pp. 356–360, 1984.
[9] J. M. Forbes and M. E. Cooper, “Mechanisms of diabetic complications,” Physio-
logical Reviews, vol. 93, no. 1, pp. 137–188, 2013.
[10] C. K. D. P. Consortium et al., “Association of estimated glomerular filtration rate
and albuminuria with all-cause and cardiovascular mortality in general population
cohorts: a collaborative meta-analysis,” The Lancet, vol. 375, no. 9731, pp. 2073–
2081, 2010.
[11] J. A. Kraut and I. Kurtz, “Metabolic acidosis of CKD: diagnosis, clinical character-
istics, and treatment,” American Journal of Kidney Diseases, vol. 45, no. 6, pp. 978–
993, 2005.
[12] V. Jha, G. Garcia-Garcia, K. Iseki, Z. Li, S. Naicker, B. Plattner, R. Saran, A. Y.-M.
Wang, and C.-W. Yang, “Chronic kidney disease: global dimension and perspec-
tives,” The Lancet, vol. 382, no. 9888, pp. 260–272, 2013.
[13] H. Schiffl, S. M. Lang, and R. Fischer, “Daily hemodialysis and the outcome of acute
renal failure,” New England Journal of Medicine, vol. 346, no. 5, pp. 305–310, 2002.
[14] F. Locatelli, B. Canaud, K.-U. Eckardt, P. Stenvinkel, C. Wanner, and C. Zoccali,
“Oxidative stress in end-stage renal disease: an emerging threat to patient outcome,”
Nephrol. Dial. Transplant, vol. 18, pp. 1272–1280, July 2003.
[15] R. W. Schrier, Diseases of the Kidney and Urinary Tract (8th ed.). Williams Wilkins, 2006.
Lippincott
[16] M. J. Shensa et al., “The discrete wavelet transform: wedding the a trous and Mallat
algorithms,” IEEE Transactions on Signal Processing, vol. 40, no. 10, pp. 2464–
2482, 1992.
[17] J. Cheng, Q. Zou, and Y. Zhao, “ECG signal classification based on deep CNN and
BiLSTM,” BMC Medical Informatics and Decision Making, vol. 21, pp. 1–12, 2021.
[18] T. M. Ingolfsson, X. Wang, M. Hersche, A. Burrello, L. Cavigelli, and L. Benini,
“ECG-TCN: Wearable cardiac arrhythmia detection with a temporal convolutional
network,” in 2021 IEEE 3rd International Conference on Artificial Intelligence Cir-
cuits and Systems (AICAS), pp. 1–4, IEEE, 2021.
[19] H. Ma, C. Chen, Q. Zhu, H. Yuan, L. Chen, and M. Shu, “An ECG signal classifica-
tion method based on dilated causal convolution,” Computational and Mathematical
Methods in Medicine, vol. 2021, no. 1, p. 6627939, 2021.
[20] L. Medsker and L. C. Jain, Recurrent neural networks: design and applications.
CRC press, 1999.
[21] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of
the IEEE Conference on Computer Vision and Pattern Recognition, pp. 7132–7141,
2018.
[22] C.-J. Chung, “基於深度神經網路之心臟驟停風險預測系統,” 國立中山大學碩
士論文, 2023.
[23] U. Grenander, Probability and Statistics: The Harald Cramér Volume. Almqvist
Wiksell, 1959.
[24] C. J. De Luca, L. D. Gilmore, M. Kuznetsov, and S. H. Roy, “Filtering the sur-
face EMG signal: Movement artifact and baseline noise contamination,” Journal of
Biomechanics, vol. 43, no. 8, pp. 1573–1579, 2010.
[25] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation,
vol. 9, no. 8, pp. 1735–1780, 1997.
[26] J. L. Ba, J. R. Kiros, and G. E. Hinton, “Layer normalization,” arXiv Preprint
arXiv:1607.06450, 2016.
[27] D. Hendrycks and K. Gimpel, “Gaussian error linear units (gelus),” arXiv Preprint
arXiv:1606.08415, 2016.
[28] S. Santurkar, D. Tsipras, A. Ilyas, and A. Madry, “How does batch normalization
help optimization?,” Advances in Neural Information Processing Systems, vol. 31,
2018.
[29] J. A. Swets, “The Relative Operating Characteristic in Psychology: A technique
for isolating effects of response bias finds wide use in the study of perception and
cognition.,” Science, vol. 182, no. 4116, pp. 990–1000, 1973.
[30] F. Sebastiani, “Machine learning in automated text categorization,” ACM Computing
Surveys (CSUR), vol. 34, no. 1, pp. 1–47, 2002.
[31] C. D. Manning, P. Raghavan, and H. Schütze, Introduction to information retrieval.
Cambridge university press, 2008.