臺灣博碩士論文加值系統

心電圖是一種量測心臟電流傳導的訊號，其量測方式快速、容易取得、非侵入式且低成本。隨著人工智能的進步，許多研究應用深度學習模型於心電圖建立人工智能心臟疾病快速篩檢工具。然而，大量且具準確標記的資料是深度學習模型能夠達到高準確度的基本關鍵之一。近幾年的發展中，由於電子醫療記錄的普及，無標記的醫學資料越來越普遍；因此，相較於之前著重在監督式學習的研究，本研究專注在無監督式特徵學習去得到通用的心電圖特徵。本研究的⽅法包含兩個階段，分別為無監督式訓練階段以及監督式訓練階段，並透過這兩個階段可以讓模型不需要標註的資訊就可以得到通用的心電圖特徵，並且訓練簡單的線性與非線性的分類器就能夠擁有預測的能力。本研究提出心跳抽樣的方法去使用無監督式度量學習在高維度空間中有鑑別度的特徵。藉由實驗結果驗証，本研究所提出的方法可提升無監督式學習的能力，並且縮小無監督式學習與監督式學習的差距。相較於其他方法，本研究提取的特徵在不同心臟疾病當中擁有更好的轉移學習的能力。尤其當有標註的資料變少時，一般監督式學習方法的表現力會隨之下降，但本研究的方法仍能維持在比監督式學習更好的表現力。因此，本研究提出之方法能夠同時在真實且大量資料集中擁有優異的表現。

Electrocardiogram is a powerful signal measuring for the electrical conduction of the heart, which is a quick, ubiquitous, non-invasive, and low-cost test to diagnose various heart diseases. With the merging success of artificial intelligence (AI), lots of studies apply deep neural networks targeting AI-enabled heart disease screening tool based on electrocardiogram. Nevertheless, clear labels are an essential prerequisite for supervised deep learning models to reach high prediction performance. Recently, unlabeled data becomes more accessible due to the increasing prevalence of electronic medical records. In contrast to supervised learning studies, this work focuses on unsupervised representation learning to attain general electrocardiogram representation. The proposed framework consists of two phases, the unsupervised training phase and the supervised training phase, which learn the general features without the knowledge of labels and train simple linear, non-linear classifiers for prediction tasks, respectively. We propose a heartbeat sampling method to adopt unsupervised deep metric learning techniques for discriminative representation in the high dimensional feature space. The experimental results indicate that our method outperforms other alternatives significantly and reduces the performance gap between supervised learning and unsupervised learning. Comparing to other methodologies, our learning method can achieve higher performance when transferring representation between different heart disease prediction tasks. While supervised learning incurs performance degradation when the number of input data decreases, our proposed models can retain high performance to surpass supervised learning. The proposed approach offers to be robust for large-scale real-world datasets to achieve consistent performance improvement.

1 Introduction 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research Aims and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Related Works 7
2.1 Unsupervised Representation Learning . . . . . . . . . . . . . . . . . . . . 7
2.2 Metric Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Electrocardiogram Classification . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Unsupervised Representation Learning for Biomedical Signals . . . . . . . 11
3 Proposed Method 13
3.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Proposed Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2.1 Heartbeat Sampling Method . . . . . . . . . . . . . . . . . . . . . . 15
3.2.2 Model Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2.3 Metric Learning Loss . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.4 Heartbeats Ensemble Method . . . . . . . . . . . . . . . . . . . . . 20
3.3 Proposed Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4 Experimental Evaluation 24
4.1 Experimental Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.2 Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.3 Hyperparameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.4 Experimental Setups . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1.5 Compared Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1.6 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2.1 External Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2.2 Internal Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Summary of the Experimental Evaluation . . . . . . . . . . . . . . . . . . 44
5 Conclusion and Future Works 45
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
v

[1] Zachi I Attia, Suraj Kapa, Francisco Lopez-Jimenez, Paul M McKie, Dorothy J
Ladewig, Gaurav Satam, Patricia A Pellikka, Maurice Enriquez-Sarano, Peter A
Noseworthy, Thomas M Munger, et al. “Screening for cardiac contractile dysfunction
using an artificial intelligence–enabled electrocardiogram”. In: Nature medicine 25.1
(2019), pp. 70–74.
[2] Zachi I Attia, Peter A Noseworthy, Francisco Lopez-Jimenez, Samuel J Asirvatham,
Abhishek J Deshmukh, Bernard J Gersh, Rickey E Carter, Xiaoxi Yao, Alejandro A
Rabinstein, Brad J Erickson, 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”. In: The Lancet 394.10201 (2019),
pp. 861–867.
[3] Philip Bachman, R Devon Hjelm, and William Buchwalter. “Learning representations
by maximizing mutual information across views”. In: Advances in Neural
Information Processing Systems. 2019, pp. 15509–15519.
[4] Shaojie Bai, J Zico Kolter, and Vladlen Koltun. “An empirical evaluation of generic
convolutional and recurrent networks for sequence modeling”. In: arXiv preprint
arXiv:1803.01271 (2018).
[5] Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. “Deep clustering
for unsupervised learning of visual features”. In: Proceedings of the European
Conference on Computer Vision (ECCV). 2018, pp. 132–149.
[6] Carlos Carreiras, Ana Priscila Alves, André Lourenço, Filipe Canento, Hugo Silva,
Ana Fred, et al. “BioSPPy: Biosignal Processing in Python”. In: Accessed on 3.28
(2015), p. 2018.
[7] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. “A simple
framework for contrastive learning of visual representations”. In: arXiv preprint
arXiv:2002.05709 (2020).
[8] Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter
Abbeel. “Infogan: Interpretable representation learning by information maximizing
generative adversarial nets”. In: Advances in neural information processing systems.
2016, pp. 2172–2180.
[9] Susan Cheng, Michelle J Keyes, Martin G Larson, Elizabeth L McCabe, Christopher
Newton-Cheh, Daniel Levy, Emelia J Benjamin, Ramachandran S Vasan, and
Thomas J Wang. “Long-term outcomes in individuals with prolonged PR interval
or first-degree atrioventricular block”. In: Jama 301.24 (2009), pp. 2571–2577.
[10] Gari D Clifford, Ikaro Silva, Benjamin Moody, Qiao Li, Danesh Kella, Abdullah
Shahin, Tristan Kooistra, Diane Perry, and Roger G Mark. “The PhysioNet/computing
in cardiology challenge 2015: reducing false arrhythmia alarms in the ICU”.
In: 2015 Computing in Cardiology Conference (CinC). IEEE. 2015, pp. 273–276.
[11] Carl Doersch, Abhinav Gupta, and Alexei A Efros. “Unsupervised visual representation
learning by context prediction”. In: Proceedings of the IEEE International
Conference on Computer Vision. 2015, pp. 1422–1430.
[12] Abdur R Feyjie, Reza Azad, Marco Pedersoli, Claude Kauffman, Ismail Ben Ayed,
and Jose Dolz. “Semi-supervised few-shot learning for medical image segmentation”.
In: arXiv preprint arXiv:2003.08462 (2020).
[13] Jean-Yves Franceschi, Aymeric Dieuleveut, and Martin Jaggi. “Unsupervised scalable
representation learning for multivariate time series”. In: Advances in Neural
Information Processing Systems. 2019, pp. 4652–4663.
[14] Antonio Fratini, Mario Sansone, Paolo Bifulco, and Mario Cesarelli. “Individual
identification via electrocardiogram analysis”. In: Biomedical engineering online 14.1
(2015), p. 78.
[15] Gabriel Garcia, Gladston Moreira, David Menotti, and Eduardo Luz. “Inter-patient
ECG heartbeat classification with temporal VCG optimized by PSO”. In: Scientific
reports 7.1 (2017), pp. 1–11.
[16] Behzad Ghazanfari, Fatemeh Afghah, Kayvan Najarian, Sajad Mousavi, Jonathan
Gryak, and James Todd. “An Unsupervised Feature Learning Approach to Reduce
False Alarm Rate in ICUs”. In: 2019 41st Annual International Conference of the
IEEE Engineering in Medicine and Biology Society (EMBC). IEEE. 2019, pp. 349–
353.
[17] Ary L Goldberger, Luis AN Amaral, Leon Glass, Jeffrey M Hausdorff, Plamen Ch
Ivanov, Roger G Mark, Joseph E Mietus, George B Moody, Chung-Kang Peng,
and H Eugene Stanley. “PhysioBank, PhysioToolkit, and PhysioNet: components
of a new research resource for complex physiologic signals”. In: circulation 101.23
(2000), e215–e220.
[18] Raia Hadsell, Sumit Chopra, and Yann LeCun. “Dimensionality reduction by learning
an invariant mapping”. In: 2006 IEEE Computer Society Conference on Computer
Vision and Pattern Recognition (CVPR’06). Vol. 2. IEEE. 2006, pp. 1735–
1742.
[19] Pat Hamilton. “Open source ECG analysis”. In: Computers in cardiology. IEEE.
2002, pp. 101–104.
[20] Giuseppe Jurman, Samantha Riccadonna, and Cesare Furlanello. “A comparison of
MCC and CEN error measures in multi-class prediction”. In: PloS one 7.8 (2012).
[21] Mohammad Kachuee, Shayan Fazeli, and Majid Sarrafzadeh. “Ecg heartbeat classification:
A deep transferable representation”. In: 2018 IEEE International Conference
on Healthcare Informatics (ICHI). IEEE. 2018, pp. 443–444.
[22] Richard E Katholi and Daniel M Couri. “Left ventricular hypertrophy: major risk
factor in patients with hypertension: update and practical clinical applications”. In:
International journal of hypertension 2011 (2011).
[23] Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. “Revisiting self-supervised
visual representation learning”. In: Proceedings of the IEEE conference on Computer
Vision and Pattern Recognition. 2019, pp. 1920–1929.
[24] Nikos Komodakis and Spyros Gidaris. “Unsupervised representation learning by
predicting image rotations”. In: 2018.
[25] Ziwei Liu, Zhongqi Miao, Xiaohang Zhan, Jiayun Wang, Boqing Gong, and Stella X
Yu. “Large-scale long-tailed recognition in an open world”. In: Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition. 2019, pp. 2537–
2546.
[26] Dwarikanath Mahapatra. “Semi-supervised learning and graph cuts for consensus
based medical image segmentation”. In: Pattern recognition 63 (2017), pp. 700–709.
[27] Ishan Misra and Laurens van der Maaten. “Self-supervised learning of pretextinvariant
representations”. In: arXiv preprint arXiv:1912.01991 (2019).
[28] George B Moody and Roger G Mark. “The impact of the MIT-BIH arrhythmia
database”. In: IEEE Engineering in Medicine and Biology Magazine 20.3 (2001),
pp. 45–50.
[29] Sajad Mousavi and Fatemeh Afghah. “Inter-and intra-patient ecg heartbeat classification
for arrhythmia detection: a sequence to sequence deep learning approach”.
In: ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and
Signal Processing (ICASSP). IEEE. 2019, pp. 1308–1312.
[30] Siti Nurmaini, Radiyati Umi Partan, Wahyu Caesarendra, Tresna Dewi, Muhammad
Naufal Rahmatullah, Annisa Darmawahyuni, Vicko Bhayyu, and Firdaus Firdaus.
“An Automated ECG Beat Classification System Using Deep Neural Networks
with an Unsupervised Feature Extraction Technique”. In: Applied Sciences
9.14 (2019), p. 2921.
[31] Keiichi Ochiai, Shu Takahashi, and Yusuke Fukazawa. “Arrhythmia Detection from
2-lead ECG using Convolutional Denoising Autoencoders”. In: Proceedings of the
KDD. Vol. 18. 2018.
[32] Hyun Oh Song, Yu Xiang, Stefanie Jegelka, and Silvio Savarese. “Deep metric learning
via lifted structured feature embedding”. In: Proceedings of the IEEE conference
on computer vision and pattern recognition. 2016, pp. 4004–4012.
[33] Aaron van den Oord, Yazhe Li, and Oriol Vinyals. “Representation learning with
contrastive predictive coding”. In: arXiv preprint arXiv:1807.03748 (2018).
[34] Deepak Pathak, Philipp Krahenbuhl, Jeff Donahue, Trevor Darrell, and Alexei A
Efros. “Context encoders: Feature learning by inpainting”. In: Proceedings of the
IEEE conference on computer vision and pattern recognition. 2016, pp. 2536–2544.
[35] Alec Radford, Luke Metz, and Soumith Chintala. “Unsupervised representation
learning with deep convolutional generative adversarial networks”. In: arXiv preprint
arXiv:1511.06434 (2015).
[36] Pritam Sarkar and Ali Etemad. “Self-supervised learning for ecg-based emotion
recognition”. In: ICASSP 2020-2020 IEEE International Conference on Acoustics,
Speech and Signal Processing (ICASSP). IEEE. 2020, pp. 3217–3221.
[37] Florian Schroff, Dmitry Kalenichenko, and James Philbin. “Facenet: A unified embedding
for face recognition and clustering”. In: Proceedings of the IEEE conference
on computer vision and pattern recognition. 2015, pp. 815–823.
[38] Shawn Tan, Guillaume Androz, Ahmad Chamseddine, Pierre Fecteau, Aaron Courville,
Yoshua Bengio, and Joseph Paul Cohen. “Icentia11K: An Unsupervised Representation
Learning Dataset for Arrhythmia Subtype Discovery”. In: arXiv preprint
arXiv:1910.09570 (2019).
[39] Soo-Kng Teo, Jian Cheng Wong, Bo Yang, Feng Yang, Ling Feng, Toon Wei Lim,
and Yi Su. “Reducing false arrhythmia alarms in the ICU”. In: 2015 Computing in
Cardiology Conference (CinC). IEEE. 2015, pp. 1177–1180.
[40] Xun Wang, Xintong Han, Weilin Huang, Dengke Dong, and Matthew R Scott.
“Multi-similarity loss with general pair weighting for deep metric learning”. In:
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
2019, pp. 5022–5030.
[41] Martin Wattenberg, Fernanda Viégas, and Ian Johnson. “How to Use t-SNE Effectively”.
In: Distill (2016). doi: 10.23915/distill.00002. url: http://distill.
pub/2016/misread-tsne.
[42] Xueting Yan, Ishan Misra, Abhinav Gupta, Deepti Ghadiyaram, and Dhruv Mahajan.
“ClusterFit: Improving Generalization of Visual Representations”. In: Proceedings
of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.
2020, pp. 6509–6518.
[43] Dong Yi, Zhen Lei, Shengcai Liao, and Stan Z Li. “Deep metric learning for person
re-identification”. In: 2014 22nd International Conference on Pattern Recognition.
IEEE. 2014, pp. 34–39.
[44] Ye Yuan, Guangxu Xun, Qiuling Suo, Kebin Jia, and Aidong Zhang. “Wave2vec:
Deep representation learning for clinical temporal data”. In: Neurocomputing 324
(2019), pp. 31–42.
[45] Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, and Lucas Beyer. “S4l: Selfsupervised
semi-supervised learning”. In: Proceedings of the IEEE international
conference on computer vision. 2019, pp. 1476–1485.
[46] Richard Zhang, Phillip Isola, and Alexei A Efros. “Colorful image colorization”. In:
European conference on computer vision. Springer. 2016, pp. 649–666.
[47] Muhammad Zubair, Jinsul Kim, and Changwoo Yoon. “An automated ECG beat
classification system using convolutional neural networks”. In: 2016 6th international
conference on IT convergence and security (ICITCS). IEEE. 2016, pp. 1–
5.