MEDA: Medical-Oriented Activation Editing for Hallucination Mitigation in Medical Large Vision-Language Model

Despite notable advances in automated medical image interpretation, Medical Large Vision-Language Models (Med-LVLMs) continue to suffer from severe hallucinations, posing critical safety risks in clinical deployment. Editing LVLM activations has shown promise for mitigating hallucination with minimal cost. However, due to the requirements of medical domain expertise, existing methods struggle to capture imaging manifestations and diagnostic principles that are critical for clinical interpretation, thereby limiting their effectiveness. To address these limitations, we propose the first MEDicaloriented Activation Editing (MEDA) method by integrating Query-decisive Manifestation Steering (QMS) and Principle-driven Diagnosis Induction (PDI) to promote Med-LVLM's expertise elicitation. Specifically, QMS retrieves positive query-decisive imaging manifestations as trusted guidance for activation steering, while PDI constructs positive principle-embedded diagnostic prompts to induce expert-like clinical reasoning. Extensive experiments across multiple modalities demonstrate that MEDA efficiently improves the response factuality for both VQA and report generation tasks, achieving up to 10.6% improvement on IU-Xray, while exhibiting generalization and few-shot robustness for practical application.