臺灣博碩士論文加值系統

左心室收縮功能障礙(Left ventricular systolic dysfunction, LVSD)是一種以左心室收 縮和泵血功能受損為特徵的疾病，導致左心室射血分數(Left ventricular ejection fraction, LVEF)降低和泵血效能減少。左心室收縮功能障礙的早期檢測具有挑戰性，因為早期 患者通常沒有症狀，但它對於老年人群體尤其存在重大的死亡風險。本研究旨在探索 一種非侵入性、經濟有效且長期監測的方法來檢測左心室收縮功能障礙。
本研究探討使用對比學習(Contrastive learning)從大型無標註心電圖(Electrocar- diogram, ECG)資料庫中提取有意義的表徵。通過利用資料本身的內在結構，對比學 習使模型能夠捕捉到潛在的模式和語義訊息。隨後，使用包含左心室射血分數資訊的 少量有標註的資料對模型進行微調，以便準確預測 LVSD。為了提供更全面的心臟功 能評估，本研究還包括了心震圖(Seismocardiogram, SCG)和心旋圖(Gyrocardiogram, GCG)等除了心電圖以外的額外特徵。
該研究利用台北榮民總醫院的心電圖歷史數據庫進行無監督預訓練，並使用對比學 習從中提取有價值的特徵。在台北新光吳火獅紀念醫院心臟超音波室和台北榮民總醫 院心導管室使用可穿戴嵌入式設備收集包括心電圖、心震圖和心旋圖在內的臨床數據， 以微調模型並預測左心室收縮功能障礙。結果在臨床應用中具有顯著的臨床相關性， 具有 90.0%的召回率和 89.2%的特異度，為醫療專業人員提供有價值的幫助。通過提 高左心室收縮功能障礙的召回率和特異度，這項研究有助於早期檢測、確診和監測左 心室收縮功能障礙，從而改善患者的預後並減少死亡率。此外，對比學習的應用還為 未來研究提供了新的方向，例如在其他心臟疾病的診斷和監測中，利用對比學習從無 標註資料中提取有意義的特徵。

Left ventricular systolic dysfunction (LVSD) is a condition characterized by impaired con- traction and pumping function of the left ventricle, leading to reduced left ventricular ejection fraction (LVEF) and decreased pumping efficiency. Detecting LVSD early is tough as initial- stage patients often show no symptoms, but it carries a high mortality risk, especially in the elderly. This study aims to investigate a noninvasive, cost-effective, and long-term monitoring method for detecting LVSD.
The study explores the use of contrastive learning to extract meaningful representations from a large unlabeled Electrocardiogram (ECG) database. By leveraging the inherent structure of the data, contrastive learning enables the model to capture the underlying patterns and semantic information. Subsequently, a small amount of labeled data containing LVEF information is used to fine-tune the model for accurate prediction of LVSD. To provide a more comprehensive assessment of heart function, the study incorporates additional features beyond ECG, including seismocardiogram (SCG) and gyrocardiogram (GCG).
The research utilizes data from the ECG historical database of Taipei Veterans General Hos- pital (VGHTPE) for unsupervised pre-training using contrastive learning. Clinical data, in- cluding ECG, SCG, and GCG, are collected using a wearable embedded device in the Cardiac Ultrasound Room at Taipei Shin Kong Wu Ho-Su Memorial Hospital (SKH) and the Cardiac Catheterization Room at VGHTPE, for the purposes of fine-tuning the model and predicting LVSD. The results hold significant clinical relevance, providing valuable assistance to health- care professionals in clinical applications, with a recall of 90.0% and a specificity of 89.2%.

摘要..................... i
Abstract..................... ii
Table of Contents..................... iii
List of Figures..................... vi
List of Tables..................... viii
1 Introduction..................... 1
1.1 Motivation and Background..................... 1
1.1.1 Left Ventricular Systolic Dysfunction..................... 1
1.1.2 Cardiogram..................... 2
1.1.3 Contrastive Learning..................... 3
1.1.4 Transfer Learning..................... 4
1.2 Related Work..................... 5
1.2.1 Screening LVSD using an AI-enabled ECG..................... 5
1.2.2 Contrasive Learning............................ 6
1.3 Contribution.................................... 7
2 Real-Time Cardiovascular Care Systems....................... 8
2.1 Wearable Embedded Device ........................... 8
2.1.1 ECG Sensor................................ 8
2.1.2 Motion Sensor .............................. 10
2.2 IoT Platform.................................... 10
2.2.1 IoT Server ................................ 10
2.2.2 IoT Gateway ............................... 11
2.3 High Performance Computing Server ...................... 11
2.4 Mobile App.................................... 12
3 Left Ventricular Systolic Dysfunction Detection Algorithm..................... 14
3.1 Problem Statement ................................ 14
3.2 Contrastive Learning Framework......................... 15
3.2.1 Scenario 1................................. 15
3.2.2 Scenario 2................................. 16
3.2.3 Scenario 3................................. 18
3.2.4 Data Augmentation............................ 18
3.2.5 Encoder Architecture........................... 19
3.2.6 Projection Head.............................. 21
3.2.7 Contrastive Loss ............................. 22
3.2.8 Cross-Entropy with Softmax Loss .................... 23
3.3 Supervised Learning ............................... 24
4 Experimental Results.................................. 27
4.1 Dataset ...................................... 27
4.2 Data Preprocessing ................................ 31
4.3 Experimental Results of Scenario 1........................ 32
4.3.1 Results of the Pre-training Stage ..................... 33
4.3.2 Results of the Fine-tuning Stage 1..................... 33
4.4 Experimental Results of Scenario 2........................ 35
4.4.1 Results of the Pre-training Stage ..................... 35
4.4.2 Results of the Fine-tuning Stage 1..................... 35
4.4.3 Results of the Fine-tuning Stage 2..................... 37
4.5 Experimental Results of Scenario 3........................ 39
4.5.1 Results of the Fine-tuning Stage 2..................... 39
5 Discussion........................................ 41
5.1 Comparison with Other Methods ......................... 41
5.2 Input Incorporation of Additional Sex and Age Features 42
5.3 Fine-tuning with Additional Linear Layers..................... 44
5.4 The Impact of Training and Testing Data Distribution..................... 45
6 Conclusion ....................................... 47
References ......................................... 49

[1] J. F. Pombo, B. L. Troy, and R. O. RUSSELL JR, “Left ventricular volumes and ejection
fraction by echocardiography,” Circulation, vol. 43, no. 4, pp. 480–490, 1971.
[2] M. M. Redfield, S. J. Jacobsen, J. C. Burnett, D. W. Mahoney, K. R. Bailey, and R. J.
Rodeheffer, “Burden of systolic and diastolic ventricular dysfunction in the community:
appreciating the scope of the heart failure epidemic,” Jama, vol. 289, no. 2, pp. 194–202,
2003.
[3] H. Mehta, A. Armstrong, K. Swett, S. J. Shah, M. A. Allison, B. Hurwitz, S. Bangdiwala,
R. Dadhania, D. W. Kitzman, W. Arguelles et al., “Burden of systolic and diastolic left
ventricular dysfunction among hispanics in the united states: insights from the echocardiographic study of latinos,” Circulation: Heart Failure, vol. 9, no. 4, p. e002733, 2016.
[4] P. Jong, S. Yusuf, M. F. Rousseau, S. A. Ahn, and S. I. Bangdiwala, “Effect of enalapril on
12-year survival and life expectancy in patients with left ventricular systolic dysfunction:
a follow-up study,” The Lancet, vol. 361, no. 9372, pp. 1843–1848, 2003.
[5] T. A. McDonagh, K. McDonald, and A. S. Maisel, “Screening for asymptomatic left ventricular dysfunction using b-type natriuretic peptide,” Congestive Heart Failure, vol. 14,
pp. 5–8, 2008.
[6] A. Taebi, B. E. Solar, A. J. Bomar, R. H. Sandler, and H. A. Mansy, “Recent advances in
seismocardiography,” Vibration, vol. 2, no. 1, pp. 64–86, 2019.
[7] P. Castiglioni, A. Faini, G. Parati, and M. Di Rienzo, “Wearable seismocardiography,”
in 2007 29th annual international conference of the IEEE engineering in medicine and
biology society. IEEE, 2007, pp. 3954–3957.
[8] M. Jafari Tadi, E. Lehtonen, A. Saraste, J. Tuominen, J. Koskinen, M. Teräs, J. Airaksinen, M. Pänkäälä, and T. Koivisto, “Gyrocardiography: A new non-invasive monitoring
method for the assessment of cardiac mechanics and the estimation of hemodynamic variables,” Scientific reports, vol. 7, no. 1, pp. 1–11, 2017.
[9] S. Sieciński, P. S. Kostka, and E. J. Tkacz, “Gyrocardiography: A review of the definition,
history, waveform description, and applications,” Sensors, vol. 20, no. 22, p. 6675, 2020.
[10] L. Ericsson, H. Gouk, C. C. Loy, and T. M. Hospedales, “Self-supervised representation learning: Introduction, advances, and challenges,” IEEE Signal Processing Magazine,
vol. 39, no. 3, pp. 42–62, 2022.
[11] A. v. d. Oord, Y. Li, and O. Vinyals, “Representation learning with contrastive predictive
coding,” arXiv preprint arXiv:1807.03748, 2018.
[12] T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive
learning of visual representations,” in International conference on machine learning.
PMLR, 2020, pp. 1597–1607.
[13] K. He, H. Fan, Y. Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised
visual representation learning,” in Proceedings of the IEEE/CVF conference on computer
vision and pattern recognition, 2020, pp. 9729–9738.
[14] J.-B. Grill, F. Strub, F. Altché, C. Tallec, P. Richemond, E. Buchatskaya, C. Doersch,
B. Avila Pires, Z. Guo, M. Gheshlaghi Azar et al., “Bootstrap your own latent-a new approach to self-supervised learning,” Advances in neural information processing systems,
vol. 33, pp. 21 271–21 284, 2020.
[15] F. Zhuang, Z. Qi, K. Duan, D. Xi, Y. Zhu, H. Zhu, H. Xiong, and Q. He, “A comprehensive
survey on transfer learning,” Proceedings of the IEEE, vol. 109, no. 1, pp. 43–76, 2020.
[16] Z. I. Attia, S. Kapa, F. Lopez-Jimenez, P. M. McKie, D. J. Ladewig, G. Satam, P. A. Pellikka, M. Enriquez-Sarano, P. A. Noseworthy, T. M. Munger et al., “Screening for cardiac
contractile dysfunction using an artificial intelligence–enabled electrocardiogram,” Nature
medicine, vol. 25, no. 1, pp. 70–74, 2019.
[17] D. Adedinsewo, R. E. Carter, Z. Attia, P. Johnson, A. H. Kashou, J. L. Dugan, M. Albus, J. M. Sheele, F. Bellolio, P. A. Friedman et al., “Artificial intelligence-enabled ecg
algorithm to identify patients with left ventricular systolic dysfunction presenting to the
emergency department with dyspnea,” Circulation: Arrhythmia and Electrophysiology,
vol. 13, no. 8, p. e008437, 2020.
[18] J. C. Jentzer, A. H. Kashou, Z. I. Attia, F. Lopez-Jimenez, S. Kapa, P. A. Friedman,
and P. A. Noseworthy, “Left ventricular systolic dysfunction identification using artificial
intelligence-augmented electrocardiogram in cardiac intensive care unit patients,” International journal of cardiology, vol. 326, pp. 114–123, 2021.
[19] A. H. Kashou, J. R. Medina-Inojosa, P. A. Noseworthy, R. J. Rodeheffer, F. LopezJimenez, I. Z. Attia, S. Kapa, C. G. Scott, A. T. Lee, P. A. Friedman et al., “Artificial
intelligence–augmented electrocardiogram detection of left ventricular systolic dysfunction in the general population,” in Mayo Clinic proceedings, vol. 96, no. 10. Elsevier,
2021, pp. 2576–2586.
[20] B. O. d. F. Brito, Z. I. Attia, L. N. A. Martins, P. Perel, M. C. P. Nunes, E. C. Sabino, C. S.
Cardoso, A. M. Ferreira, P. R. Gomes, A. Luiz Pinho Ribeiro et al., “Left ventricular systolic dysfunction predicted by artificial intelligence using the electrocardiogram in chagas
disease patients–the sami-trop cohort,” PLoS neglected tropical diseases, vol. 15, no. 12,
p. e0009974, 2021.
[21] S. Katsushika, S. Kodera, M. Nakamoto, K. Ninomiya, S. Inoue, S. Sawano, N. Kakuda,
H. Takiguchi, H. Shinohara, R. Matsuoka et al., “The effectiveness of a deep learning
model to detect left ventricular systolic dysfunction from electrocardiograms,” International Heart Journal, vol. 62, no. 6, pp. 1332–1341, 2021.
[22] Y.-C. Huang, Y.-C. Hsu, Z.-Y. Liu, C.-H. Lin, R. Tsai, J.-S. Chen, P.-C. Chang, H.-T. Liu,
W.-C. Lee, H.-T. Wo et al., “Artificial intelligence-enabled electrocardiographic screening
for left ventricular systolic dysfunction and mortality risk prediction,” Frontiers in Cardiovascular Medicine, vol. 10, p. 1070641, 2023.
[23] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in
Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp.
770–778.
[24] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale
hierarchical image database,” in 2009 IEEE conference on computer vision and pattern
recognition. Ieee, 2009, pp. 248–255.
[25] D. Kiyasseh, T. Zhu, and D. A. Clifton, “CLOCS: Contrastive learning of cardiac signals across space, time, and patients,” in International Conference on Machine Learning.
PMLR, 2021, pp. 5606–5615.
[26] H. Chen, G. Wang, G. Zhang, P. Zhang, and H. Yang, “CLECG: A novel contrastive learning framework for electrocardiogram arrhythmia classification,” IEEE Signal Processing
Letters, vol. 28, pp. 1993–1997, 2021.
[27] B. Burns, B. Grant, D. Oppenheimer, E. Brewer, and J. Wilkes, “Borg, omega, and kubernetes,” Communications of the ACM, vol. 59, no. 5, pp. 50–57, 2016.
[28] J. Lewis and M. Fowler, “Microservices,” Mar 2014. [Online]. Available: https:
//martinfowler.com/articles/microservices.html
[29] U. Hunkeler, H. L. Truong, and A. Stanford-Clark, “Mqtt-s—a publish/subscribe protocol
for wireless sensor networks,” in 2008 3rd International Conference on Communication
Systems Software and Middleware and Workshops (COMSWARE’08). IEEE, 2008, pp.
791–798.