发表机构
IEM Kolkata; School of UEMK Kolkata; Heritage Institute of Technology Kolkata; Indian Institute of Information Technology, Kalyani（印度工程管理学院 Kolkata校区; UEMK Kolkata学院; 加尔各答遗产技术学院; 卡利亚尼印度信息技术学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出CoT-Mediate框架，结合双臂协议与来源控制干预，在VQA-RAD数据集上评估LLaVA-Med和MedGemma，发现上下文位置而非声明来源是医学VLMs使用生成推理的主要决定因素。
AI 中文摘要
医学视觉-语言模型（VLMs）在回答临床问题前会生成思维链（CoT）推理，但该推理是否对预测产生因果影响仍不明确。本文提出CoT-Mediate，这是一种行为框架，可在模型自身生成的推理中扰动单个具有临床意义的属性，并衡量所得预测是否遵循编辑后的推理。该框架结合了双臂协议——比较重新提示的证据与前缀强制延续，以及来源控制干预——仅改变相同推理的归因来源，以分离推理中介与逢迎行为。我们在1000个VQA-RAD样本上分别评估LLaVA-Med和MedGemma。前缀强制延续始终比重新提示产生更高的中介忠实度，而来源分析揭示了不同的模型特定顺从行为。在两种模型中，移除视觉证据会增加对注入推理的依赖，而偏侧性是追踪忠实度最低的临床属性。这些结果表明，注入推理的机制会显著影响测得的忠实度，且上下文位置而非声明的来源是医学VLMs是否使用其生成推理的主要决定因素。
英文摘要
Medical vision-language models (VLMs) generate chain-of-thought (CoT) reasoning before answering clinical questions, but whether this reasoning causally influences predictions remains unclear. We present CoT-Mediate, a behavioral framework that perturbs a single clinically meaningful attribute within a model's own generated reasoning and measures whether the resulting prediction follows the edited reasoning. Our framework combines a dual-arm protocol comparing re-prompted evidence with prefix-forced continuation, together with a provenance-controlled intervention that varies only the attributed source of identical reasoning to disentangle reasoning mediation from sycophancy. We evaluate LLaVA-Med and MedGemma on 1,000 VQA-RAD samples each. Prefix-forced continuation consistently yields higher mediation faithfulness than re-prompting, while the provenance analysis reveals distinct model-specific deference behaviors. Across both models, removing visual evidence increases reliance on injected reasoning, whereas laterality is the least faithfully tracked clinical attribute. These results show that the mechanism used to inject reasoning substantially affects measured faithfulness and that contextual position, rather than stated provenance, is the primary determinant of whether medical VLMs use their generated reasoning.
