New framework combats medical MLLM hallucinations with evidence injection

Researchers have developed a novel, training-free framework to enhance the trustworthiness of medical multimodal large language models (MLLMs). This system, called Synergistic Perception-Reasoning Governance, addresses hallucinations by injecting verifiable anatomical evidence during inference. It recalibrates visual perception through ROI-guided activation modulation and anchors textual reasoning by mapping anatomical coordinates to semantic tokens. Evaluations on multiple datasets using various MLLMs demonstrated significant improvements in accuracy and a substantial reduction in hallucinations. AI

IMPACT This framework could improve the reliability of medical AI tools by reducing hallucinations in diagnostic and reporting tasks.

RANK_REASON Research paper detailing a new framework for medical MLLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• InternVL-3.5-8B/38B
• LLaVA-1.5-7B
• LLaVA-Med-1.5-7B
• MedSAM
• Qwen3-VL-8B/32B
• Synergistic Perception-Reasoning Governance

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →