Med-HEAL：通过幻觉感知的上下文学习分析与缓解医学大语言模型中的幻觉
Med-HEAL: Analyzing and Mitigating Hallucinations in Medical LLMs with Hallucination-Aware In-Context Learning
AI总结：
提出Med-HEAL框架，利用临床数据构建幻觉数据集，并通过自我批评和检索增强上下文学习策略缓解医学大语言模型中的幻觉，实验表明自我批评策略可提升多数模型准确率。
摘要：
医学大语言模型中的幻觉对临床决策支持构成严重风险，特别是当模型需要推理复杂的电子健康记录时。然而，现有基准通常缺乏真实的临床背景，并且对如何在实践中缓解幻觉提供的见解有限。我们引入了Med-HEAL，一个利用临床基础数据系统性地识别、分析和缓解医学大语言模型幻觉的框架。基于源自MIMIC-IV出院摘要的EHRNoteQA基准，我们通过评估BioMistral-7B在开放式临床问答任务上的表现构建了一个幻觉数据集。模型输出通过一个结合LLM-as-a-Judge评估（GPT-4o）和医学生评审员人工审核的双重评估流程进行标注，通过自定义的基于网络的评估系统产生正确性判断和推理错误注释。然后，我们利用该数据集研究缓解策略：自我批评流程，其中测试模型审查自己的答案以检测潜在错误，并对标记案例重新生成响应；以及检索增强的上下文学习，该模型暴露于幻觉和纠正示例。在五个开源大语言模型（BioMistral、Llama-3.1、DeepSeek、Qwen2.5和Qwen3）上的实验表明，自我批评策略在无需参数更新的情况下提高了五个模型中三个的准确率（p < 0.05）。Med-HEAL既提供了一个可重用的幻觉数据集，也提供了一个实用的框架，用于研究和缓解医学大语言模型中的幻觉，支持AI系统在临床环境中更安全地部署。我们的代码和数据可在https://github.com/yimingliao-blad/med-heal.git公开获取。
英文摘要：
Hallucinations in medical large language models (LLMs) pose serious risks for clinical decision support, particularly when models must reason over complex electronic health records (EHRs). However, existing benchmarks often lack a realistic clinical context and provide limited insight into how hallucinations can be mitigated in practice. We introduce Med-HEAL, a framework for systematically identifying, analyzing, and mitigating hallucinations in medical LLMs using clinically grounded data. Building on the EHRNoteQA benchmark derived from MIMIC-IV discharge summaries, we construct a hallucination dataset by evaluating BioMistral-7B on open-ended clinical question answering tasks. Model outputs are labeled through a dual evaluation pipeline that combines LLM-as-a-Judge assessment (GPT-4o) with human auditing by medical student reviewers, producing correctness judgments and annotations of reasoning errors via a custom web-based evaluation system. We then leverage this dataset to investigate mitigation strategies: a self-critique pipeline, in which the test model reviews its own answers to detect potential errors and regenerates responses for flagged cases, and retrieval-augmented in-context learning (RA-ICL), which exposes the model to hallucinated and corrected examples. Experiments across five open-source LLMs-BioMistral, Llama-3.1, DeepSeek, Qwen2.5, and Qwen3, show that the self-critique strategy improves accuracy for three of five models (p < 0.05) without requiring parameter updates. Med-HEAL provides both a reusable hallucination dataset and a practical framework for studying and mitigating hallucinations in medical LLMs, supporting safer deployment of AI systems in clinical environments. Our code and data are publicly available at https://github.com/yimingliao-blad/med-heal.git.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
