从案例到智能系统：人工智能驱动的单肺通气肺段级精准供氧闭环控制框架——一项概念验证研究

From Case to Intelligent System: An AI-Driven Closed-Loop Framework for Segmental Oxygen Delivery during One-Lung Ventilation—A Proof-of-Concept Study

摘要: 背景：单肺通气(one-lung ventilation, OLV)是胸外科手术中常用的通气方式，但术中低氧血症仍是其最常见且最具挑战性的并发症之一。传统处理方法如提高吸氧浓度、调整呼吸机参数或短暂双肺通气等往往具有滞后性，难以实现低氧的早期预警与精准干预。为此，本文提出一种人工智能驱动的肺段级精准供氧闭环控制系统，用于单肺通气期间低氧血症的智能化管理。方法：本研究首先基于一例单肺通气期间发生严重低氧血症的病例，在支气管镜引导下实施肺段级精准供氧并成功改善患者氧合状态。在此基础上，构建人工智能驱动的肺段级精准供氧系统架构，包括监测层、决策层、控制层及支气管镜执行层，形成闭环控制系统。同时，建立低氧风险预测模型、氧气流量控制模型及供氧肺段选择模型，构成“预测–决策–控制”的智能供氧模型体系。最后，通过概念验证研究与模拟实验，对人工智能供氧策略与传统供氧策略在低氧预警时机、氧合稳定性及干预次数等方面进行比较。结果：在病例中，当单肺通气期间患者血氧饱和度降至约73%且传统干预无效时，通过支气管镜引导向远离手术区域的肺段持续供氧后，患者血氧饱和度迅速恢复并维持在95%以上，且未对手术视野造成明显影响。模拟实验结果表明，与传统供氧策略相比，人工智能供氧策略能够更早进行低氧预警，减少低氧持续时间，提高最低血氧水平，并减少供氧干预次数，显示出更好的氧合稳定性。结论：本研究提出的人工智能驱动肺段级精准供氧系统可实现单肺通气期间低氧血症的早期预测与精准干预，构建了围术期氧合管理的闭环控制框架。该研究从病例出发，通过系统设计与概念验证研究，证明了人工智能在单肺通气精准供氧中的潜在应用价值，为未来智能麻醉与智能呼吸管理系统的研究提供了新的技术路径。

Abstract: Background: One-lung ventilation (OLV) is a commonly used ventilation method in thoracic surgery, but intraoperative hypoxemia remains one of the most common and challenging complication. Conventional management strategies, such as increasing the fraction of inspired oxygen, adjusting ventilator parameters, or applying intermittent two-lung ventilation, are often reactive and may not provide early prediction or precise intervention. This study proposes an artificial intelligence (AI)-driven segmental oxygen delivery system with a closed-loop control framework for intelligent hypoxemia management during OLV. Methods: This study was initiated from a clinical case in which severe hypoxemia occurred during OLV and was successfully corrected by bronchoscopic-guided segmental oxygen delivery. Based on this case, an AI-driven segmental oxygen delivery system was developed, consisting of a Monitoring Layer, Decision Layer, Control Layer, and Bronchoscope Execution Layer, forming a closed-loop control system. Three AI models were designed, including a hypoxemia prediction model, an oxygen flow control model, and a segment selection model, corresponding to prediction, decision-making, and control processes. A proof-of-concept study and simulation experiments were conducted to compare the AI-based oxygen delivery strategy with conventional oxygenation strategies in terms of hypoxemia prediction timing, oxygenation stability, and intervention frequency. Results: In the clinical case, when SpO₂ dropped to approximately 73% during OLV and conventional interventions were ineffective, continuous oxygen delivery to a non-operative lung segment under bronchoscopic guidance rapidly restored SpO₂ to above 95% without significantly interfering with the surgical field. Simulation results suggested that the AI-based strategy could provide earlier hypoxemia warning, reduce the duration of hypoxemia, improve minimum SpO₂ levels, and reduce the number of oxygenation interventions compared with conventional strategies, indicating improved oxygenation stability. Conclusion: The proposed AI-driven segmental oxygen delivery system enables early prediction and precise intervention for hypoxemia during OLV and establishes a closed-loop framework for perioperative oxygenation management. This study translates a case-based clinical technique into an intelligent system framework and demonstrates its feasibility through a proof-of-concept study, providing a potential pathway toward intelligent anesthesia and precision ventilation.