MedGym：面向动态医疗治疗强化学习的统一基准
A Unified Benchmark for Dynamic Medical Treatment Reinforcement Learning
研究者提出 MedGym，一个基于连续时间框架的动态治疗推荐基准环境，利用 Physics-Informed Neural Networks 从临床数据构建可配置的医疗 RL 基准。它支持在不规则治疗时机和患者个体动态下直接对比离散时间与连续时间方法，并可从个性化和轨迹级安全等临床视角进行评估，为连续时间动态治疗的医疗 RL 方法提供更贴近现实的标准化评测。
Authors:Yuepeng Wang, Ken Kawano, Yoshihiko Fujisawa, Yongqi Zhou, Akifumi Wachi, Mehrshad Sadria, Lei Zhou, Richard Weiss, Katsuki Fujisawa, Ying Chen, Xin Liu, Kyoung-Sook Kim, Xiao Hu, Sebastien Gros, Xun Shen
Abstract:Medical treatment recommendation poses several challenges to reinforcement learning (RL): patient physiology evolves in continuous time, measurements and interventions are performed at irregular intervals, and treatment effects vary substantially across individuals. Existing RL formulations and simulated environments, however, are based on discrete-time MDPs with fixed decision intervals. Thus, it remains difficult to evaluate whether RL methods can handle time-interval-dependent disease progression, personalized treatment response, and safety between consecutive measurement points. To address this gap, we introduce MedGym, a benchmark environment for dynamic treatment recommendation. MedGym models longitudinal patient evolution in a continuous-time framework and constructs a configurable medical RL benchmark from clinical data by using Physics-Informed Neural Networks. The resulting benchmark enables direct comparison between discrete-time and continuous-time methods under irregular treatment timing and patient-specific dynamics. Furthermore, MedGym supports evaluation from clinically important perspectives, such as personalization and trajectory-level safety. By providing a standardized and configurable benchmark for continuous-time dynamic treatment, MedGym enables more realistic and informative evaluation of medical RL methods.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.01028 [cs.LG] |
| (or arXiv:2606.01028v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.01028 arXiv-issued DOI via DataCite |
Submission history
From: Yuepeng Wang [view email]
[v1]
Sun, 31 May 2026 05:36:03 UTC (7,372 KB)
[v2]
Mon, 21 Sep 2026 09:19:53 UTC (5,636 KB)
[v3]
Thu, 24 Sep 2026 07:32:59 UTC (5,636 KB)
[v4]
Fri, 9 Oct 2026 01:04:18 UTC (5,636 KB)
来源：arXiv：cs.LG · arxiv.org