X-Ragent: tool-augmented RL for chest X-ray reasoning
A new arXiv paper introduces X-Ragent, a post-training framework for tool-augmented reinforcement learning aimed at evidence-grounded chest X-ray (CXR) reasoning. The method decouples learned auxiliary-evidence seeking from validity-gated quantitative measurement acquisition, initializes Qwen3-VL-8B via agentic supervised fine-tuning on Rad-CoT (a 7.7K-sample corpus of causally ordered tool-augmented trajectories plus report-style diagnostic reasoning), then optimizes with GRPO using an accuracy-gated reward that applies measurement-consistency and tool-cost terms only to correct trajectories. The authors report the strongest results across three CXR benchmarks, including an 8.6-point gain on the fully held-out ChestAgentBench.
It illustrates a way to train medical imaging agents to seek evidence selectively and reject invalid measurements, which could make diagnostic reasoning more faithful.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2610.16811 (cs)
[Submitted on 12 Oct 2026]
Title:X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray Reasoning
Authors:Yi Wang, Zhenlong Yuan, Yuchan Guo, Maolei Huang, Zijian Song, Dapeng Zhang, Rongbin Tan, Jiyuan Wang, Zhiyuan Wang, Fangfang Lin, Shuhan Song, Huawei Cao
View a PDF of the paper titled X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray Reasoning, by Yi Wang and 11 other authors