超越准确性：多模态医学推理中的视觉定位评估

Recent research indicates that text-only reinforcement learning with verifiable rewards (RLVR) can match or outperform image-text RLVR on multimodal medical visual question answering (VQA) benchmarks, suggesting that current evaluation protocols may fail to adequately measure causal visual dependence. A counterfactual evaluation framework is proposed, utilizing real, blank, and shuffled images across four medical VQA benchmarks: PathVQA, PMC-VQA, SLAKE, and VQA-RAD. Beyond accuracy, the study measures Visual Reliance Score (VRS), Image Sensitivity (IS), and introduces Hallucinated Visual Reasoning Rate (HVRR) to identify instances where models generate visual claims despite producing image-invariant answers. Findings reveal that while RLVR enhances accuracy, it degrades visual grounding, with text-only RLVR achieving superior accuracy.