本研究旨在建立一套基於雙手指穿透式光學感測器之光體積變化描記圖（photoplethysmography, PPG）訊號的機器學習模型，用以評估血液透析通路狹窄程度，提供客觀的儀器輔助結果，提供臨床診斷的第二意見，並可應用於居家監測情境，協助患者及早掌握通路狀況。在資料收集方面，本研究於花蓮慈濟醫學中心心臟血管外科收集來自40位受試者之306筆人工血管（arteriovenous graft, AVG）左前臂的資料PPG訊號資料。考量PPG訊號易受無意識動作影響，導致低頻基線飄移、高頻雜訊與運動偽影，本研究設計了及時訊號品質檢測方法，並且將其整合至收案的APP中，於回診收案時自動排除低品質的訊號，以確保後續分析的可用性與有效性。搭配以都卜勒超音波（Doppler ultrasound）採集血管狹窄程度（degree of stenosis, DOS）作為標註。將雙手PPG訊號以單一心跳週期為單位進行切割，擷取 56 個時域訊號特徵作為模型輸入。在模型建立方面，以五種特徵排序方式搭配四種機器學習分類模型，共計 20 種組合。為提升模型的臨床應用價值，最終採用軟投票方式整合分段預測結果，回推為符合回診需求的單筆訊號預測模式。特徵篩選則採用逐步特徵選擇策略，針對不同模型進行特徵數量的最佳化調整。特徵選擇方面，採用基於逐步特徵選擇（stepwise selection）的策略，針對不同模型動態調整特徵數量。模型訓練與驗證過程中則採用k-group fold交叉驗證（k = 3），以提高對個體間差異的穩定性與泛化能力。最終結果顯示，以極限梯度提升（eXtreme gradient boosting, XGBoost）模型搭配SHAP（SHapley Additive exPlanations）進行特徵排序，並選定其中 18 個關鍵特徵作為模型輸入，於k-group fold達到平均準確率80.10%，優於隨機森林（random forest）的 77.41% 及多層感知神經網絡（multilayer perceptron, MLP）的 78.57%，另外XGBoost於加權平均召回率（weighted average recall）為0.7931，加權平均精確率（weighted average precision）為0.7429同樣優於其他模型的表現。顯示XGBoost在本研究資料中具有最佳分類效能與可解釋性。

This study aims to develop a machine learning model based on photoplethysmography (PPG) signals acquired from bilateral fingertip transmission-type optical sensors, for the purpose of evaluating the degree of vascular access stenosis in hemodialysis patients. The objective is to provide an objective, instrument-assisted result as a secondary reference for clinical diagnosis, as well as to enable home-based monitoring to help patients detect access dysfunction at an early stage.
For data collection, a total of 306 PPG recordings were obtained from the left forearms of 40 patients with arteriovenous grafts (AVG) at the Department of Cardiovascular Surgery, Hualien Tzu Chi Hospital. Considering that PPG signals are easily affected by unconscious movements, resulting in low-frequency baseline drift, high-frequency noise, and motion artifacts, this study designed a real-time signal quality assessment method. The method was integrated into a custom-built mobile application to automatically exclude low-quality signals during clinical data acquisition, ensuring the usability and reliability of subsequent analyses. The degree of stenosis (DOS) measured via Doppler ultrasound was used as the reference label. The bilateral PPG signals were segmented based on individual cardiac cycles, and 56 time-domain features were extracted for model input. In the modeling process, five feature ranking methods were combined with four machine learning classifiers, resulting in a total of 20 model combinations. To enhance the clinical applicability of the model, a soft voting strategy was adopted to aggregate predictions from segmented data into a single prediction per clinical session. Feature selection was performed using a stepwise selection strategy, allowing dynamic adjustment of the number of features based on the characteristics of each model. During model training and validation, k-group fold cross-validation (k = 3) was applied to improve robustness and generalizability across individual differences. The final results showed that the eXtreme Gradient Boosting (XGBoost) model, combined with SHapley Additive exPlanations (SHAP) for feature ranking and using 18 selected key features as input, achieved the highest average accuracy of 80.10% under k-group fold cross-validation. This performance outperformed that of the random forest model 77.41% and the multilayer perceptron (MLP) model 78.57%. Additionally, XGBoost also achieved superior performance in terms of weighted average recall of 0.7931 and weighted average precision of 0.7429, surpassing the other models. These findings indicate that XGBoost demonstrated the best classification performance and interpretability on the dataset used in this study.

[1] 台灣腎臟醫學會, “TWRDS 台灣腎病年報.” Accessed: Jun. 23, 2025. [Online]. Available: https://www.tsn.org.tw/twrds_detail.html?year=2021
[2] 耕莘醫院, “2023-03-11 「新國病」贏美國新加坡！台灣9萬洗腎人口居全球第1　得靠「這招」預防-本文另刊載於祝你健康 | 耕莘新聞 | 耕莘訊息 | 天主教耕莘醫療財團法人耕莘醫院.” Accessed: Mar. 12, 2025. [Online]. Available: https://www.cth.org.tw/?aid=110&pid=60&page_name=detail&iid=662
[3] 聯合報, “圖解台灣洗腎人口關鍵數字 能擺脫「洗腎王國」標籤嗎 | 聯合報,” 聯合報數位版. Accessed: Mar. 12, 2025. [Online]. Available: https://vip.udn.com/vip/story/122664/7654988
[4] 台灣腎臟醫學會, “TWRDS 台灣腎病年報.” Accessed: Mar. 12, 2025. [Online]. Available: https://www.tsn.org.tw/twrds_detail.html?year=2023
[5] L. A. Inker, B. C. Astor, C. H. Fox, T. Isakova, J. P. Lash, C. A. Peralta, et al., “KDOQI US Commentary on the 2012 KDIGO Clinical Practice Guideline for the Evaluation and Management of CKD,” Am. J. Kidney Dis., vol. 63, no. 5, pp. 713–735, May 2014.
[6] P. E. Stevens, S. B. Ahmed, J. J. Carrero, B. Foster, A. Francis, R. K. Hall, et al., “KDIGO 2024 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease,” Kidney Int., vol. 105, no. 4, pp. S117–S314, Apr. 2024.
[7] NIDDK, “Peritoneal Dialysis - NIDDK,” National Institute of Diabetes and Digestive and Kidney Diseases. Accessed: Mar. 12, 2025. [Online]. Available: https://www.niddk.nih.gov/health-information/kidney-disease/kidney-failure/peritoneal-dialysis
[8] NIDDK, “Hemodialysis - NIDDK,” National Institute of Diabetes and Digestive and Kidney Diseases. Accessed: Mar. 12, 2025. [Online]. Available: https://www.niddk.nih.gov/health-information/kidney-disease/kidney-failure/hemodialysis
[9] 中國醫藥大學附設醫院, “淺談血液透析血管通路種類 | 衛教單張 - 中國醫藥大學附設醫院.” Accessed: Mar. 12, 2025. [Online]. Available: http://cmuh.cmu.edu.tw/HealthEdus/Detail?no=5267
[10] 馬偕紀念醫院, “馬偕紀念醫院 衛教單張：動靜脈廔管手術.” Accessed: Mar. 12, 2025. [Online]. Available: https://www.mmh.org.tw/know_health_view.php?docid=511
[11] J. H. Lawson, L. E. Niklason, and P. Roy-Chaudhury, “Challenges and novel therapies for vascular access in haemodialysis,” Nat. Rev. Nephrol., vol. 16, no. 10, pp. 586–602, Oct. 2020.
[12] A. A. Alian and K. H. Shelley, “Photoplethysmography,” Hemodynamic Monit. Devices, vol. 28, no. 4, pp. 395–406, Dec. 2014.
[13] J. Park, H. S. Seok, S.-S. Kim, and H. Shin, “Photoplethysmogram Analysis and Applications: An Integrative Review,” Front. Physiol., vol. 12, p. 808451, 2021.
[14] T. Tamura, Y. Maeda, M. Sekine, and M. Yoshida, “Wearable Photoplethysmographic Sensors—Past and Present,” Electronics, vol. 3, no. 2, pp. 282–302, 2014.
[15] J. M. Szafron, E. E. Heng, J. Boyd, J. D. Humphrey, and A. L. Marsden, “Hemodynamics and Wall Mechanics of Vascular Graft Failure,” Arterioscler. Thromb. Vasc. Biol., vol. 44, no. 5, pp. 1065–1085, May 2024.
[16] P. Roy-Chaudhury, B. S. Kelly, M. A. Miller, A. Reaves, J. Armstrong, N. Nanayakkara, et al., “Venous neointimal hyperplasia in polytetrafluoroethylene dialysis grafts,” Kidney Int., vol. 59, no. 6, pp. 2325–2334, Jun. 2001.
[17] L. Li, C. M. Terry, Y.-T. E. Shiu, and A. K. Cheung, “Neointimal hyperplasia associated with synthetic hemodialysis grafts,” Kidney Int., vol. 74, no. 10, pp. 1247–1261, Nov. 2008.
[18] D. Santoro, F. Benedetto, P. Mondello, N. Pipitò, D. Barillà, F. Spinelli, et al., “Vascular access for hemodialysis: current perspectives,” Int. J. Nephrol. Renov. Dis., vol. 7, no. null, pp. 281–294, Jul. 2014.
[19] 翁宏嘉 and 黎思源, “血液透析血管通路狹窄的機轉,” 腎臟與透析, vol. 31, no. 2, pp. 58–62, 2019.
[20] L. Jia, L. Wang, F. Wei, H. Yu, H. Dong, B. Wang, et al., “Effects of wall shear stress in venous neointimal hyperplasia of arteriovenous fistulae,” Nephrology, vol. 20, no. 5, pp. 335–342, 2015.
[21] D.-F. Yeih, Y.-S. Wang, Y.-C. Huang, M.-F. Chen, and S.-S. Lu, “Physiology-based diagnosis algorithm for arteriovenous fistula stenosis detection,” Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. Annu. Int. Conf., vol. 2014, pp. 4619–4622, 2014.
[22] H.-Y. Wang, C.-H. Wu, C.-Y. Chen, and B.-S. Lin, “Novel Noninvasive Approach for Detecting Arteriovenous Fistula Stenosis | IEEE Journals & Magazine | IEEE Xplore.” Accessed: Mar. 12, 2025. [Online]. Available: https://ieeexplore.ieee.org/document/6754131
[23] S. Julkaew, T. Wongsirichot, K. Damkliang, and P. Sangthawan, “DeepVAQ : an adaptive deep learning for prediction of vascular access quality in hemodialysis patients,” BMC Med. Inform. Decis. Mak., vol. 24, no. 1, p. 45, Feb. 2024.
[24] P.-Y. Chiang, P.-P. Chao, T.-Y. Tu, Y.-H. Kao, C.-Y. Yang, D.-C. Tarng, et al., “Machine Learning Classification for Assessing the Degree of Stenosis and Blood Flow Volume at Arteriovenous Fistulas of Hemodialysis Patients Using a New Photoplethysmography Sensor Device,” Sensors, vol. 19, no. 15, 2019.
[25] Y.-C. Du and A. Stephanus, “Levenberg-Marquardt Neural Network Algorithm for Degree of Arteriovenous Fistula Stenosis Classification Using a Dual Optical Photoplethysmography Sensor,” Sensors, vol. 18, no. 7, 2018.
[26] A. Stephanus and Y. C. Du, “K-NN based algorithm for degree of stenosis classification using dual non-invasive photoplethysmography system,” J. Phys. Conf. Ser., vol. 1450, no. 1, p. 012052, Feb. 2020.
[27] H.-Y. Wang, C.-H. Wu, C.-Y. Chen, and B.-S. Lin, “Novel noninvasive approach for detecting arteriovenous fistula stenosis,” IEEE Trans. Biomed. Eng., vol. 61, no. 6, pp. 1851–1857, Jun. 2014.
[28] Y. Yan, C. C. Poon, and Y. Zhang, “Reduction of motion artifact in pulse oximetry by smoothed pseudo Wigner-Ville distribution,” J. NeuroEngineering Rehabil., vol. 2, no. 1, p. 3, Mar. 2005.
[29] Xu Lisheng, Wang Kuanquan, D. Zhang, and Shi Cheng, “Adaptive baseline wander removal in the pulse waveform,” in Proceedings of 15th IEEE Symposium on Computer-Based Medical Systems (CBMS 2002), Jun. 2002, pp. 143–148.
[30] Y. Maeda, M. Sekine, and T. Tamura, “Relationship Between Measurement Site and Motion Artifacts in Wearable Reflected Photoplethysmography,” J. Med. Syst., vol. 35, no. 5, pp. 969–976, Oct. 2011.
[31] F. Peng, Z. Zhang, X. Gou, H. Liu, and W. Wang, “Motion artifact removal from photoplethysmographic signals by combining temporally constrained independent component analysis and adaptive filter,” Biomed. Eng. OnLine, vol. 13, no. 1, p. 50, Apr. 2014.
[32] B. S. Kim and S. K. Yoo, “Motion artifact reduction in photoplethysmography using independent component analysis,” IEEE Trans. Biomed. Eng., vol. 53, no. 3, pp. 566–568, Mar. 2006.
[33] A. Biswas, M. Singha Roy, and R. Gupta, “Motion Artifact Reduction from Finger Photoplethysmogram Using Discrete Wavelet Transform,” in Recent Trends in Signal and Image Processing, S. Bhattacharyya, A. Mukherjee, H. Bhaumik, S. Das, and K. Yoshida, Eds., Singapore: Springer Singapore, 2019, pp. 89–98.
[34] B. Lee, J. Han, H. J. Baek, J. H. Shin, K. S. Park, and W. J. Yi, “Improved elimination of motion artifacts from a photoplethysmographic signal using a Kalman smoother with simultaneous accelerometry,” Physiol. Meas., vol. 31, no. 12, p. 1585, Oct. 2010.
[35] T. Desquins, F. Bousefsaf, A. Pruski, and C. Maaoui, “A Survey of Photoplethysmography and Imaging Photoplethysmography Quality Assessment Methods,” Appl. Sci., vol. 12, no. 19, 2022.
[36] C. Fischer, B. Dömer, T. Wibmer, and T. Penzel, “An Algorithm for Real-Time Pulse Waveform Segmentation and Artifact Detection in Photoplethysmograms,” IEEE J. Biomed. Health Inform., vol. 21, no. 2, pp. 372–381, Mar. 2017.
[37] C. Fischer, M. Glos, T. Penzel, and I. Fietze, “Extended algorithm for real-time pulse waveform segmentation and artifact detection in photoplethysmograms,” Somnologie, vol. 21, no. 2, pp. 110–120, Jun. 2017.
[38] C. Orphanidou, “Quality Assessment for the Photoplethysmogram (PPG),” in Signal Quality Assessment in Physiological Monitoring: State of the Art and Practical Considerations, C. Orphanidou, Ed., Cham: Springer International Publishing, 2018, pp. 41–63.
[39] Q. Li and G. D. Clifford, “Signal quality and data fusion for false alarm reduction in the intensive care unit,” J. Electrocardiol., vol. 45, no. 6, pp. 596–603, Nov. 2012.
[40] E. Sabeti, N. Reamaroon, M. Mathis, J. Gryak, M. Sjoding, and K. Najarian, “Signal quality measure for pulsatile physiological signals using morphological features: Applications in reliability measure for pulse oximetry,” Inform. Med. Unlocked, vol. 16, p. 100222, Jan. 2019.
[41] T. Pereira, K. Gadhoumi, M. Ma, R. Colorado, K. J. Keenan, K. Meisel, et al., “Robust Assessment of Photoplethysmogram Signal Quality in the Presence of Atrial Fibrillation | IEEE Conference Publication | IEEE Xplore.” Accessed: Mar. 13, 2025. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/8743969
[42] Y. Zhang and J. Pan, “Assessment of photoplethysmogram signal quality based on frequency domain and time series parameters | IEEE Conference Publication | IEEE Xplore.” Accessed: Mar. 13, 2025. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/8302279
[43] D. Roh and H. Shin, “Recurrence Plot and Machine Learning for Signal Quality Assessment of Photoplethysmogram in Mobile Environment,” Sensors, vol. 21, no. 6, 2021.
[44] J. Azar, A. Makhoul, R. Couturier, and J. Demerjian, “Deep recurrent neural network-based autoencoder for photoplethysmogram artifacts filtering,” Comput. Electr. Eng., vol. 92, p. 107065, Jun. 2021.
[45] 蔡延隆, “使用雙手指脈搏血氧儀進行血液透析人工血管狹窄之預測__臺灣博碩士論文知識加值系統.” Accessed: Mar. 21, 2025. [Online]. Available: https://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi/login?o=dnclcdr&s=id=%22112NCKU5442015%22.&searchmode=basic
[46] Nordic, “nRF52832 - Nordic Semiconductor.” Accessed: Mar. 24, 2025. [Online]. Available: https://www.nordicsemi.com/Products/nRF52832
[47] TI.com, “AFE4404 產品規格表、產品資訊與支援 | TI.com.” Accessed: Mar. 24, 2025. [Online]. Available: https://www.ti.com/product/zh-tw/AFE4404#product-details
[48] All States M.E.D, “GE Logiq e Portable Ultrasound Machine,” All States M.E.D. Accessed: Apr. 09, 2025. [Online]. Available: https://www.allstatesmed.com/products/ge-logiq-e-portable-ultrasound-machine
[49] Y.-C. Du and A. Stephanus, “A Novel Classification Technique of Arteriovenous Fistula Stenosis Evaluation Using Bilateral PPG Analysis,” Micromachines, vol. 7, no. 9, p. 147, Aug. 2016.
[50] Y. Liang, M. Elgendi, Z. Chen, and R. Ward, “An optimal filter for short photoplethysmogram signals,” Sci. Data, vol. 5, no. 1, p. 180076, May 2018.
[51] D. G. Lapitan, D. A. Rogatkin, E. A. Molchanova, and A. P. Tarasov, “Estimation of phase distortions of the photoplethysmographic signal in digital IIR filtering,” Sci. Rep., vol. 14, no. 1, p. 6546, Mar. 2024.
[52] M.-T. Wu, I.-F. Liu, Y.-H. Tzeng, and L. Wang, “Modified photoplethysmography signal processing and analysis procedure for obtaining reliable stiffness index reflecting arteriosclerosis severity,” Physiol. Meas., vol. 43, no. 8, Aug. 2022.
[53] heartpy: Heart Rate Analysis Toolkit. Python. Accessed: Jun. 18, 2025. [OS Independent]. Available: https://github.com/paulvangentcom/heartrate_analysis_python
[54] M. Elgendi, “On the Analysis of Fingertip Photoplethysmogram Signals,” Curr. Cardiol. Rev., vol. 8, no. 1, pp. 14–25, Feb. 2012.
[55] S. R. Alty, N. Angarita-Jaimes, S. C. Millasseau, and P. J. Chowienczyk, “Predicting Arterial Stiffness From the Digital Volume Pulse Waveform,” IEEE Trans. Biomed. Eng., vol. 54, no. 12, pp. 2268–2275, Feb. 2007.
[56] Y. Kurylyak, F. Lamonaca, and D. Grimaldi, “A Neural Network-based method for continuous blood pressure estimation from a PPG signal,” in 2013 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), May 2013, pp. 280–283.
[57] M. B. Kursa and W. R. Rudnicki, “Feature Selection with the Boruta Package,” J. Stat. Softw., vol. 36, pp. 1–13, Sep. 2010.
[58] S. Lundberg and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” Nov. 25, 2017, arXiv: arXiv:1705.07874. doi: 10.48550/arXiv.1705.07874.
[59] L. Breiman, “Random Forests,” Mach. Learn., vol. 45, no. 1, pp. 5–32, Oct. 2001.
[60] J. H. Friedman, “Greedy function approximation: A gradient boosting machine.,” Ann. Stat., vol. 29, no. 5, pp. 1189–1232, Oct. 2001.
[61] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, in KDD ’16. New York, NY, USA: Association for Computing Machinery, 13 2016, pp. 785–794.