監測連續動脈壓，是手術中實時監控血壓的黃金準則，以往需採用侵入式的方式，利用插入血管內的壓力感測器來取得，但臨床上會有諸多限制，需有經驗之醫護人員施打、耗材昂貴、有血液感染的風險、對血管造成傷害、血管剝離…等，且臨床上連續動脈壓監測的置入困難程度會受到成人或孩童而有所差異，年齡以及疾病對於血管硬化的影響也會影響置入困難程度。
受到上述的限制，本研究利用既有的心電圖波形、血氧波形資料及外部資料集，發展創新之深度學習模型後產生動脈連續波形、血壓的預測，既有研究主要著重在即時波形的轉化，例如將血氧波形圖、心電圖轉化為實時的動脈血壓連續波形；抑或是針對未來高低血壓進行分類預測，而在實務上，資料前處理至模型運算完畢產生結果仍有時間差，因此本研究將利用醫療場域既有的生理資訊，無需添加額外設備或耗材來進行量測，並針對過往僅預測收縮壓(SBP)以及舒張壓(DBP)的不足，進行未來一秒之連續動脈壓預測。
本研究將針對波形、SBP以及DBP就多項指標進行評估，並且在不同的資料集上進行實驗，以驗證本研究所提出之模型架構泛化性。並根據既有研究所提出之模型架構進行比較，以及單一、雙重訊號使用後結果的差別，也針對訊號輸入源之不同秒數進行遮罩，判斷在輸入之訊號窗口內，哪一秒資訊遺失對模型影響最大，而從最後的實驗結果可以得知，就波形、SBP及DBP的RMSE、相關係數上以及AAMI、BHS標準上，本研究所提出之模型誤差皆優於既有研究。

Monitoring continuous arterial pressure is the gold standard for real-time monitoring of blood pressure during surgery. In the past, invasive methods were used to obtain pressure sensors inserted into blood vessels, but there are many clinical limitations, such as the experienced medical personnel, expensive supplies, risk of bloodstream infection, injury to blood vessels, vessel dissection, etc. The degree of difficulty in placing continuous arterial pressure monitoring sensor in clinic varies depending on whether the patient is an adult or a child. Age and the effect of disease on vascular sclerosis may also affect the degree of insertion difficulty.
Due to the above limitations, this study uses existing Electrocardiography (ECG) waveforms, Photoplethysmography(PPG) waveforms and external data sets to develop an innovative deep learning model to generate continuous arterial waveforms and blood pressure predictions. Therefore, this study uses the existing physiological information in the medical field to make measurements without adding additional equipment or consumables, and to predict the future continuous arterial pressure in one second to address the shortcomings of predicting only systolic blood pressure (SBP) and diastolic blood pressure (DBP) in the past.
This study evaluates the waveform, SBP, and DBP on various indicators and conduct experiments on different data sets to verify the generalizability of the proposed model structure. The model structure is compared with the existing model structure, and the difference between the results of single and dual signal use. The error of the proposed model is better than that of the existing studies in terms of waveform, the root-mean-square error of SBP and DBP, correlation coefficient, the Association for the Advancement of Medical Instrumentation(AAMI) standards , and the British Hypertension Society(BHS) standards.

論文審定書 i
論文公開授權書 ii
摘要 iii
Abstract iv
目錄 v
圖次 vii
表次 viii
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機 1
1.3 研究目的 2
第二章 文獻探討 3
2.1 循環神經網路 (Recurrent Neural Network, RNN) 3
2.1.1 長短期記憶 (Long Short-Term Memory, LSTM) 3
2.1.2 GRU (Gate Recurrent Unit) 4
2.1.3 雙向循環神經網路(Bidirectional RNN, BRNN) 5
2.2 卷積神經網路 (Convolutional Neural Network, CNN) 6
2.2.1 一維卷積神經網路 (1D Convolutional Neural Networks, 1DCNN) 7
2.2.2 時間卷積網路 (Temporal Convolutional Networks, TCN ) 8
2.3 Sequence to Sequence (Seq2seq) 8
2.3.1 注意力機制(Attention Mechanism) 9
2.3.2 Bahdanau Attention 9
2.3.3 Luong Attention 10
2.3.4 自注意力機制(Self-Attention) 11
2.4 濾波器 (Filter) 12
2.4.1 帶通濾波器(Band-pass filter, BPF) 12
2.4.2 Savitzky-Golay濾波器(Savitzky-Golay filter) 12
2.5 血壓預測(Arterial Blood Pressure Predict) 13
2.5.1 舒張壓/收縮壓測量(DBP/SBP) 13
2.5.2 Real-Time Continuous Arterial Blood Pressure 14
第三章 研究方法與步驟 15
3.1 資料前處理(K Preprocessing) 16
3.1.1 取出重疊時段 16
3.1.2 修正取樣 16
3.1.3 取出連續片段 16
3.1.4 進行濾波 16
3.2 資料處理(Processing) 21
3.3 非平衡資料調整 22
3.4 損失函數 23
3.5 超參數優化 23
3.6 模型架構 23
3.6.1 Bi-GRU to GRU + Luong Attention 23
3.6.2 Transformer based Model 25
3.6.3 TCN-GRU 25
3.7 預測區間調整 26
第四章 實驗 28
4.1 資料集介紹 28
4.1.1 MIMIC-II Waveform 資料集 28
4.1.2 K ICU Waveform資料集 28
4.2 評估方式 29
4.3 結果分析 30
第五章 實驗結果 31
5.1 實驗環境與設置 31
5.2 BHS與AAMI標準 31
5.3 對比既有文獻 32
5.4 對比不同模型結果 33
5.5 使用不同通道結果 34
5.6 最佳化超參數探討 35
5.7 遮罩不同秒數結果 37
5.8 討論 40
第六章 結論與建議 42
6.1 結論 42
6.2 未來展望 42
參考文獻 44

[1] 109年國人死因統計結果 ：https://www.mohw.gov.tw/cp-5017-61533-1.html
[2] Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A. A., Ciompi, F., Ghafoorian, M., ... & Sánchez, C. I. (2017). A survey on deep learning in medical image analysis. Medical image analysis, 42, 60-88.
[3] Hemdan, E. E. D., Shouman, M. A., & Karar, M. E. (2020). Covidx-net: A framework of deep learning classifiers to diagnose covid-19 in x-ray images. arXiv preprint arXiv:2003.11055.
[4] Zeroual, A., Harrou, F., Dairi, A., & Sun, Y. (2020). Deep learning methods for forecasting COVID-19 time-Series data: A Comparative study. Chaos, Solitons & Fractals, 140, 110121.
[5] Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. In Advances in neural information processing systems (pp. 3104-3112).
[6] https://colah.github.io/posts/2015-08-Understanding-LSTMs/
[7] Hochreiter, S. (1991). Untersuchungen zu dynamischen neuronalen Netzen. Diploma, Technische Universität München, 91(1).
[8] Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE transactions on neural networks, 5(2), 157-166.
[9] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[10] Greff, K., Srivastava, R. K., Koutník, J., Steunebrink, B. R., & Schmidhuber, J. (2016). LSTM: A search space odyssey. IEEE transactions on neural networks and learning systems, 28(10), 2222-2232.
[11] Jozefowicz, R., Zaremba, W., & Sutskever, I. (2015, June). An empirical exploration of recurrent network architectures. In International conference on machine learning (pp. 2342-2350). PMLR.
[12] Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014). Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555.
[13] Li, Z., & Yu, Y. (2016). Protein secondary structure prediction using cascaded convolutional and recurrent neural networks. arXiv preprint arXiv:1604.07176.
[14] Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE transactions on Signal Processing, 45(11), 2673-2681.
[15] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[16] Graves, A., Jaitly, N., & Mohamed, A. R. (2013, December). Hybrid speech recognition with deep bidirectional LSTM. In 2013 IEEE workshop on automatic speech recognition and understanding (pp. 273-278). IEEE.
[17] https://blog.goodaudience.com/introduction-to-1d-convolutional-neural-networks-in-keras-for-time-sequences-3a7ff801a2cf?fbclid=IwAR0B11L3DXkTHqheCSYpG1SetABlWcREFsnLQf5tSdVbFtcVkkjr9zCdngw
[18] https://pytorch.org/tutorials/intermediate/seq2seq_translation_tutorial.html
[19] Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.
[20] Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473.
[21] Luong, M. T., Pham, H., & Manning, C. D. (2015). Effective approaches to attention-based neural machine translation. arXiv preprint arXiv:1508.04025.
[22] https://blog.floydhub.com/attention-mechanism/
[23] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.
[24] http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2019/Lecture/Transformer%20(v5).pdf
[25] El-Hajj, C., & Kyriacou, P. A. (2021). Deep learning models for cuffless blood pressure monitoring from PPG signals using attention mechanism. Biomedical Signal Processing and Control, 65, 102301.
[26] Zadi, A. S., Alex, R., Zhang, R., Watenpaugh, D. E., & Behbehani, K. (2018). Arterial blood pressure feature estimation using photoplethysmography. Computers in biology and medicine, 102, 104-111.
[27] Slapničar, G., Mlakar, N., & Luštrek, M. (2019). Blood pressure estimation from photoplethysmogram using a spectro-temporal deep neural network. Sensors, 19(15), 3420.
[28] Chowdhury, M. H., Shuzan, M. N. I., Chowdhury, M. E., Mahbub, Z. B., Uddin, M. M., Khandakar, A., & Reaz, M. B. I. (2020). Estimating blood pressure from the photoplethysmogram signal and demographic features using machine learning techniques. Sensors, 20(11), 3127.
[29] Fan, X., Wang, H., Zhao, Y., Li, Y., & Tsui, K. L. (2021). An Adaptive Weight Learning-Based Multitask Deep Network for Continuous Blood Pressure Estimation Using Electrocardiogram Signals. Sensors, 21(5), 1595.
[30] Sideris, C., Kalantarian, H., Nemati, E., & Sarrafzadeh, M. (2016, May). Building continuous arterial blood pressure prediction models using recurrent networks. In 2016 IEEE International Conference on Smart Computing (SMARTCOMP) (pp. 1-5). IEEE.
[31] Sadrawi, M., Lin, Y. T., Lin, C. H., Mathunjwa, B., Fan, S. Z., Abbod, M. F., & Shieh, J. S. (2020). Genetic deep convolutional autoencoder applied for generative continuous arterial blood pressure via photoplethysmography. Sensors, 20(14), 3829.
[32] Sadrawi, M., Lin, Y. T., Lin, C. H., Mathunjwa, B., Hsin, H. T., Fan, S. Z., ... & Shieh, J. S. (2021). Non-invasive hemodynamics monitoring system based on electrocardiography via deep convolutional autoencoder. Sensors, 21(18), 6264.
[33] Aguirre, N., Grall-Maës, E., Cymberknop, L. J., & Armentano, R. L. (2021). Blood pressure morphology assessment from photoplethysmogram and demographic information using deep learning with attention mechanism. Sensors, 21(6), 2167.
[34] Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng CK, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23):e215-e220 [Circulation Electronic Pages; http://circ.ahajournals.org/content/101/23/e215.full]; 2000 (June 13). PMID: 10851218; doi: 10.1161/01.CIR.101.23.e215
[35] Kachuee, M., Kiani, M. M., Mohammadzade, H., & Shabany, M. (2015, May). Cuff-less high-accuracy calibration-free blood pressure estimation using pulse transit time. In 2015 IEEE international symposium on circuits and systems (ISCAS) (pp. 1006-1009). IEEE.
[36] Harfiya, L. N., Chang, C. C., & Li, Y. H. (2021). Continuous blood pressure estimation using exclusively photopletysmography by LSTM-based signal-to-signal translation. Sensors, 21(9), 2952.
[37] Qin, K., Huang, W., & Zhang, T. (2021). Deep generative model with domain adversarial training for predicting arterial blood pressure waveform from photoplethysmogram signal. Biomedical Signal Processing and Control, 70, 102972.
[38] Sideris, C., Kalantarian, H., Nemati, E., & Sarrafzadeh, M. (2016, May). Building continuous arterial blood pressure prediction models using recurrent networks. In 2016 IEEE International Conference on Smart Computing (SMARTCOMP) (pp. 1-5). IEEE.
[39] O'Brien, E., Petrie, J., Littler, W., de Swiet, M., Padfield, P. L., O'Malley, K., ... & Atkins, N. (1990). The British Hypertension Society protocol for the evaluation of automated and semi-automated blood pressure measuring devices with special reference to ambulatory systems. Journal of hypertension, 8(7), 607-619.
[40] O'brien, E., Waeber, B., Parati, G., Staessen, J., & Myers, M. G. (2001). Blood pressure measuring devices: recommendations of the European Society of Hypertension. Bmj, 322(7285), 531-536
[41] Zerveas, G., Jayaraman, S., Patel, D., Bhamidipaty, A., & Eickhoff, C. (2021, August). A transformer-based framework for multivariate time series representation learning. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (pp. 2114-2124).