来源:最新院系讲座

11月28日 | 许燕勋:A Reinforcement Learning Framework for Learning Optimal Oxygen Treatments in Patients with Acute Respiratory Distress Syndrome

来源:统计学院发布时间:2025-11-24浏览次数:10

时    间:2025年11月28日(周五) 10:00 – 11:00

地    点:普陀校区理科大楼A1714室

报告人:许燕勋    Johns Hopkins University副教授

主持人:周迎春    华东师范大学教授

摘   要:

Acute respiratory failure presents major challenges in selecting the most effective oxygen therapy. Options range from standard oxygen delivery to high-flow nasal cannula (HFNC) and invasive mechanical ventilation (IMV). While IMV can be life-saving, it carries significant risks, and the optimal timing for intubation remains uncertain. HFNC offers a less invasive alternative, but questions remain about which patients benefit most and when escalation to IMV is appropriate. To address these gaps, we developed a reinforcement learning–based approach to identify individualized, dynamic treatment strategies. Using detailed clinical registry data, we modeled hourly treatment decisions based on each patient’s evolving clinical status. To ensure safety and clinical relevance, we introduced the concept of confident states, where treatment follows established guidelines or well-recognized clinician practices. In less certain scenarios, reinforcement learning–derived recommendations guide decision-making. This hybrid framework integrates data-driven insights with clinical expertise, providing a pathway to optimize oxygen therapy decisions and improve outcomes in critical care.

报告人简介:

许燕勋教授现任美国约翰斯·霍普金斯大学应用数学与统计系终身教职教授(博士生导师),并荣膺Joseph & Suzanne Jenniches冠名教授。她长期致力于贝叶斯统计理论与人工智能交叉研究,在强化学习、高维数据分析、非参数统计及不确定性量化等领域取得重要突破。其创新性统计机器学习方法已成功应用于智能医疗健康多个核心领域,包括精准医疗(临床试验设计、癌症基因组学、个性化治疗), 疾病早期诊断(阿尔茨海默病预测模型), 以及电子健康记录智能分析。Xu教授曾获ISBA世界大会Mitchell Prize等重要学术奖项,其指导的多名学生也斩获美国统计学会(ASA)和国际生物统计学会(ENAR)等权威机构的优秀论文奖。目前担任国际贝叶斯分析学会(ISBA)执行委员会财务主管,并任《Bayesian Analysis》主编及多个统计学期刊副主编。作为独立PI,其研究持续获得国立卫生研究院(NIH)及工业界资助,在人工智能驱动的大健康研究领域形成了"基础理论-算法开发-临床转化"的完整创新链,推动了人工智能在医疗健康领域的科学化应用。