跳到正文
arXiv cs.AI· Hyungyung Lee, Hangyul Yoon, Edward Choi·· 2 小时前AI 评分34

CXReasonAgent:面向胸片 X 光的证据驱动诊断推理 Agent

CXReasonAgent: Evidence-Grounded Diagnostic Reasoning Agent for Chest X-rays

AI 导读

研究者提出 CXReasonAgent,将 LLM 与临床诊断工具结合,基于影像衍生的诊断和视觉证据对胸部 X 光进行证据驱动的诊断推理,回应 LVLM 回答缺乏证据支撑、适应新任务需重新训练的问题。评估采用自建多轮对话基准 CXReasonDail,含 12 项诊断任务的 1,946 段对话,结果显示其回答比 LVLM 更忠实于证据。该工作已被 ML4H 2026 接收,并提供在线演示。

正文 · 原文

View PDF HTML (experimental)

Abstract:Chest X-ray plays a central role in thoracic diagnosis, and its interpretation inherently requires multi-step, evidence-grounded reasoning. However, large vision-language models (LVLMs) often generate plausible responses that are not faithfully grounded in diagnostic evidence and provide limited visual evidence for verification, while also requiring costly retraining to support new diagnostic tasks, limiting their reliability and adaptability in clinical settings. To address these limitations, we present CXReasonAgent, a diagnostic agent that integrates a large language model (LLM) with clinically grounded diagnostic tools to perform evidence-grounded diagnostic reasoning using image-derived diagnostic and visual evidence. To evaluate these capabilities, we introduce CXReasonDail, a multi-turn dialogue benchmark with 1,946 dialogues across 12 diagnostic tasks, and show that CXReasonAgent produces more faithfully grounded responses than LVLMs. These findings highlight the importance of integrating clinically grounded diagnostic tools to enable reliable and verifiable diagnostic reasoning. The demo is available \href{this https URL}{here}.
Comments: Accepted at ML4H 2026
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2602.23276 [cs.AI]
  (or arXiv:2602.23276v3 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2602.23276

arXiv-issued DOI via DataCite

Submission history

From: Hyungyung Lee [view email]
[v1] Thu, 26 Feb 2026 17:51:21 UTC (18,344 KB)
[v2] Tue, 24 Mar 2026 09:05:23 UTC (18,665 KB)
[v3] Wed, 18 Nov 2026 05:55:38 UTC (20,090 KB)

来源:arXiv cs.AI · arxiv.org