急性低氧性呼吸衰竭（Acute Hypoxemic Respiratory Failure, AHRF）是一種危及生命的臨床狀況，其特徵為氧合功能受損而無顯著高碳酸血症，常見於重症病患中。高流量鼻導管（High-Flow Nasal Cannula, HFNC）因可提供加溫加濕的高流速氧氣，已成為治療AHRF的首選非侵襲性呼吸支持方式。然而，HFNC治療在相當比例的患者中可能失敗，需升級為侵襲性機械通氣。若未能及時識別HFNC治療失敗，可能導致病情惡化、延長加護病房住院時間，甚至提高死亡率。因此，及早且準確地預測HFNC失敗風險，對於優化呼吸照護與改善病患預後具有重要意義。本研究提出一套機器學習預測架構，結合來自胸腔X光影像的放射特徵與完整的生理資料，以預測HFNC治療失敗風險。影像部分首先透過 U-Net 架構對肺部區域進行分割，並以對比受限自適應直方圖均衡化（Contrast Limited Adaptive Histogram Equalization, CLAHE）進行增強。隨後從處理後的影像中提取統計型與基於圖論的紋理特徵。特徵選擇階段採用多重宇宙最佳化演算法（Multi-Verse Optimizer, MVO），並以支援向量機（Support Vector Machine, SVM）作為適應度評估模型。結果顯示，影像與生理特徵的整合能夠提升模型效能。基於SHAP的特徵重要度分析指出，血氧指標、腎功能指標及影像紋理異質性為主要預測因子。此整合式預測方法可望協助臨床醫師及早識別 HFNC 治療高風險病患，促進即時介入，提升臨床處置效率與病人預後。

Acute Hypoxemic Respiratory Failure (AHRF) is a life-threatening condition characterized by impaired oxygenation without significant hypercapnia and is frequently encountered in critically ill patients. High-Flow Nasal Cannula (HFNC) therapy has become a preferred non-invasive respiratory support modality for AHRF due to its ability to deliver heated and humidified oxygen at high flow rates. However, HFNC may fail in a considerable proportion of patients, necessitating escalation to invasive mechanical ventilation. Delayed identification of HFNC failure can lead to worsened clinical outcomes, including increased morbidity, prolonged Intensive Care Unit (ICU) stay, and higher mortality. Therefore, early and accurate prediction of HFNC failure is essential for optimizing respiratory management and improving patient prognosis. This study proposed a machine learning framework that integrates radiomic features extracted from chest X-ray with comprehensive physiological data to predict HFNC failure. Lung regions were segmented using a U-Net architecture, and image contrast was enhanced via Contrast Limited Adaptive Histogram Equalization (CLAHE). From the processed images, both conventional statistical and graph-based topological features were extracted. Feature selection was optimized using the Multi-Verse Optimizer (MVO) algorithm, with a Support Vector Machine (SVM) acting as the fitness evaluator. Our results demonstrate that integrating imaging and physiological features enhances model performance. SHapley Additive exPlanations (SHAP)-based feature importance interpretation revealed key contributors such as oxygenation indices, renal function markers, and image-derived texture heterogeneity. This integrated approach may assist clinicians in timely identification of patients at high risk of HFNC failure, facilitating earlier interventions and improving clinical outcomes.

[1] V. M. Ranieri et al., "Acute respiratory distress syndrome: the Berlin Definition," The Journal of the American Medical Association, vol. 307, no. 23, 2012.
[2] G. Bellani et al., "Epidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries," The Journal of the American Medical Association, vol. 315, no. 8, pp. 788-800, 2016.
[3] L. Papazian et al., "Use of high-flow nasal cannula oxygenation in ICU adults: a narrative review," Intensive Care Medicine, vol. 42, pp. 1336-1349, 2016.
[4] P. Plant, J. Owen, and M. Elliott, "Early use of non-invasive ventilation for acute exacerbations of chronic obstructive pulmonary disease on general respiratory wards: a multicentre randomised controlled trial," The Lancet, vol. 355, no. 9219, pp. 1931-1935, 2000.
[5] A. Gray, S. Goodacre, D. E. Newby, M. Masson, F. Sampson, and J. Nicholl, "Noninvasive ventilation in acute cardiogenic pulmonary edema," New England Journal of Medicine, vol. 359, no. 2, pp. 142-151, 2008.
[6] B. L. Ferreyro et al., "Association of noninvasive oxygenation strategies with all-cause mortality in adults with acute hypoxemic respiratory failure: a systematic review and meta-analysis," The Journal of the American Medical Association, vol. 324, no. 1, pp. 57-67, 2020.
[7] A. S. Slutsky and V. M. Ranieri, "Ventilator-induced lung injury," New England Journal of Medicine, vol. 369, no. 22, pp. 2126-2136, 2013.
[8] J. P. Frat et al., "High-flow oxygen through nasal cannula in acute hypoxemic respiratory failure," New England Journal of Medicine, vol. 372, no. 23, pp. 2185-2196, Jun 2015.
[9] 台灣胸腔暨重症加護醫學會, “濕化高流量氧氣重症治療台灣專家共識,” 2021.
[10] J. Bräunlich, M. Köhler, and H. Wirtz, "Nasal highflow improves ventilation in patients with COPD," International Journal of Chronic Obstructive Pulmonary Disease, pp. 1077-1085, 2016.
[11] O. Roca, J. Riera, F. Torres, and J. R. Masclans, "High-flow oxygen therapy in acute respiratory failure," Respiratory Care, vol. 55, no. 4, pp. 408-413, 2010.
[12] R. Williams, N. Rankin, T. Smith, D. Galler, and P. Seakins, "Relationship between the humidity and temperature of inspired gas and the function of the airway mucosa," Critical Care Medicine, vol. 24, no. 11, pp. 1920-1929, 1996.
[13] S. Girod, J. Zahm, C. Plotkowski, G. Beck, and E. Puchelle, "Role of the physiochemical properties of mucus in the protection of the respiratory epithelium," European Respiratory Journal, vol. 5, no. 4, pp. 477-487, 1992.
[14] R. W. Heaton, A. F. Henderson, B. J. Gray, and J. F. Costello, "The bronchial response to cold air challenge: evidence for different mechanisms in normal and asthmatic subjects," Thorax, vol. 38, no. 7, pp. 506-511, 1983.
[15] A. Hasani, T. Chapman, D. McCool, R. Smith, J. Dilworth, and J. Agnew, "Domiciliary humidification improves lung mucociliary clearance in patients with bronchiectasis," Chronic Respiratory Disease, vol. 5, no. 2, pp. 81-86, 2008.
[16] J. Ritchie, A. Williams, C. Gerard, and H. Hockey, "Evaluation of a humidified nasal high-flow oxygen system, using oxygraphy, capnography and measurement of upper airway pressures," Anaesthesia and Intensive Care, vol. 39, no. 6, pp. 1103-1110, 2011.
[17] W. Möller et al., "Nasal high flow clears anatomical dead space in upper airway models," Journal of Applied Physiology, vol. 118, no. 12, pp. 1525-1532, 2015.
[18] R. Parke, S. McGuinness, and M. Eccleston, "Nasal high-flow therapy delivers low level positive airway pressure," British Journal of Anaesthesia, vol. 103, no. 6, pp. 886-890, 2009.
[19] T. Mauri et al., "Physiologic effects of high-flow nasal cannula in acute hypoxemic respiratory failure," American Journal of Respiratory and Critical Care Medicine, vol. 195, no. 9, pp. 1207-1215, 2017.
[20] L. Pisani et al., "Change in pulmonary mechanics and the effect on breathing pattern of high flow oxygen therapy in stable hypercapnic COPD," Thorax, vol. 72, no. 4, pp. 373-375, 2017.
[21] J. F. Fraser, A. J. Spooner, K. R. Dunster, C. M. Anstey, and A. Corley, "Nasal high flow oxygen therapy in patients with COPD reduces respiratory rate and tissue carbon dioxide while increasing tidal and end-expiratory lung volumes: a randomised crossover trial," Thorax, vol. 71, no. 8, pp. 759-761, 2016.
[22] F. T. Brink, T. Duke, and J. Evans, "High-flow nasal prong oxygen therapy or nasopharyngeal continuous positive airway pressure for children with moderate-to-severe respiratory distress?," Pediatric Critical Care Medicine, vol. 14, no. 7, pp. E326-E331, Sep 2013.
[23] S. Inoue, Y. Tamaki, S. Sonobe, J. Egawa, and M. Kawaguchi, "A pediatric case developing critical abdominal distension caused by a combination of humidified high-flow nasal cannula oxygen therapy and nasal airway," JA Clinical Reports, vol. 4, no. 1, p. 4, 2018.
[24] J. D. Ricard et al., "Use of nasal high flow oxygen during acute respiratory failure," Intensive Care Medicine, vol. 46, no. 12, pp. 2238-2247, 2020.
[25] B. J. Kang et al., "Failure of high-flow nasal cannula therapy may delay intubation and increase mortality," Intensive Care Medicine, vol. 41, pp. 623-632, 2015.
[26] M. Nishimura, "High-flow nasal cannula oxygen therapy in adults: physiological benefits, indication, clinical benefits, and adverse effects," Respiratory Care, vol. 61, no. 4, pp. 529-541, Apr 2016.
[27] A. Grünewaldt, M. Gaillard, and G. Rohde, "Predictors of high-flow nasal cannula (HFNC) failure in severe community-acquired pneumonia or COVID-19," Internal and Emergency Medicine, pp. 1-9, 2024.
[28] J. Q. Xu et al., "A novel risk-stratification models of the high-flow nasal cannula therapy in COVID-19 patients with hypoxemic respiratory failure," Frontiers in Medicine, vol. 7, Dec 2020.
[29] O. Roca et al., "An index combining respiratory rate and oxygenation to predict outcome of nasal high-flow therapy," American Journal of Respiratory and Critical Care Medicine, vol. 199, no. 11, pp. 1368-1376, Jun 2019.
[30] C. T. Lun, C. K. Leung, H. P. Shum, and S. O. So, "Predictive factors for high-flow nasal cannula failure in acute hypoxemic respiratory failure in an intensive care unit," Lung India, vol. 39, no. 1, pp. 5-11, 2022.
[31] K. Rotheray et al., "What is the relationship between the Glasgow coma scale and airway protective reflexes in the Chinese population?," Resuscitation, vol. 83, no. 1, pp. 86-89, Jan 2012.
[32] M. Xue et al., "A retrospective study to predict failure of high-flow oxygen therapy for acute hypoxic respiratory failure," International Journal of Emergency Medicine, vol. 18, no. 1, pp. 98, 2025.
[33] W. H. Cho et al., "High-flow nasal cannula therapy for acute hypoxemic respiratory failure in adults: a retrospective analysis," Internal Medicine, vol. 54, no. 18, pp. 2307-2313, 2015.
[34] H. Yu et al., "Machine learning models compared with current clinical indices to predict the outcome of high flow nasal cannula therapy in acute hypoxemic respiratory failure," Critical Care, vol. 29, no. 1, pp. 101, 2025.
[35] G. Pappy, M. Aczon, R. Wetzel, and D. Ledbetter, "Predicting high flow nasal cannula failure in an intensive care unit using a recurrent neural network with transfer learning and input data perseveration: Retrospective analysis," JMIR Medical Informatics, vol. 10, no. 3, pp. e31760, 2022.
[36] L. Yang et al., "Optimal machine learning methods for prediction of high-flow nasal cannula outcomes using image features from electrical impedance tomography," Computer Methods and Programs in Biomedicine, vol. 238, pp. 107613, 2023.
[37] J. N. Hasoon et al., "COVID-19 anomaly detection and classification method based on supervised machine learning of chest X-ray images," Results in Physics, vol. 31, pp. 105045, Dec 2021.
[38] P. Rajpurkar et al., "Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists," PLoS Medicine, vol. 15, no. 11, pp. e1002686, Nov 2018.
[39] M. D. Li et al., "Automated assessment and tracking of COVID-19 pulmonary disease severity on chest radiographs using convolutional siamese neural networks,", Radiology: Artificial Intelligence, vol. 2, no. 4, pp. e200079, Jul 2020.
[40] O. Ronneberger, P. Fischer, and T. Brox, "U-net: Convolutional networks for biomedical image segmentation," Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, proceedings, part III, vol. 18, pp. 234-241, 2015.
[41] K. J. Zuiderveld, "Contrast limited adaptive histogram equalization," Graphics Gems, vol. 4, no. 1, pp. 474-485, 1994.
[42] M. E. H. Chowdhury et al., "Can AI help in screening viral and COVID-19 pneumonia?," IEEE Access, vol. 8, pp. 132665-132676, 2020.
[43] T. Rahman et al., "Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images," Computers in Biology and Medicine, vol. 132, pp. 104319, 2021.
[44] J. J. Van Griethuysen et al., "Computational radiomics system to decode the radiographic phenotype," Cancer Research, vol. 77, no. 21, pp. e104-e107, 2017.
[45] J. Iacovacci and L. Lacasa, "Visibility graphs for image processing," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 4, pp. 974-987, 2019.
[46] T. T. Wong and P. Y. Yeh, "Reliable accuracy estimates from k-fold cross validation," IEEE Transactions on Knowledge and Data Engineering, vol. 32, no. 8, pp. 1586-1594, Aug. 2020.
[47] Y. Zhang, et al., “Towards augmented kernel extreme learning models for bankruptcy prediction: Algorithmic behavior and comprehensive analysis,” Neurocomputing, vol. 430, pp. 185–212, Mar. 2021.
[48] M. Dash and H. Liu, "Feature selection for classification," Intelligent Data Analysis, vol. 1, no. 1, pp. 131-156, 1997.
[49] S. Mirjalili, S. M. Mirjalili, and A. Hatamlou, “Multi-verse optimizer: A nature-inspired algorithm for global optimization,” Neural Computing and Applications, vol. 27, no. 2, pp. 495–513, 2016.
[50] H. Faris, M. A. Hassonah, A. M. Al-Zoubi, S. Mirjalili, and I. Aljarah, "A multi-verse optimizer approach for feature selection and optimizing SVM parameters based on a robust system architecture," Neural Computing and Applications, vol. 30, pp. 2355-2369, 2018.
[51] S. M. Lundberg and S.-I. Lee, "A unified approach to interpreting model predictions," Advances in Neural Information Processing Systems, vol. 30, 2017.
[52] D. Trachsel, B. W. McCrindle, S. Nakagawa, and D. Bohn, "Oxygenation index predicts outcome in children with acute hypoxemic respiratory failure," American Journal of Respiratory and Critical Care Medicine, vol. 172, no. 2, pp. 206-211, 2005.
[53] A. V. Kozlov and J. Grillari, "Pathogenesis of multiple organ failure: the impact of systemic damage to plasma membranes," Frontiers in Medicine, vol. 9, p. 806462, 2022.
[54] A. Assinger, W. C. Schrottmaier, M. Salzmann, and J. Rayes, "Platelets in sepsis: an update on experimental models and clinical data," Frontiers in Immunology, vol. 10, p. 1687, 2019.
[55] M. Liu et al., "Signalling pathways involved in hypoxia‐induced renal fibrosis," Journal of Cellular and Molecular Medicine, vol. 21, no. 7, pp. 1248-1259, 2017.
[56] M. Malek, J. Hassanshahi, R. Fartootzadeh, F. Azizi, and S. Shahidani, "Nephrogenic acute respiratory distress syndrome: a narrative review on pathophysiology and treatment," Chinese Journal of Traumatology, vol. 21, no. 1, pp. 4-10, 2018.