This research introduces V-RAG (Visual Retrieval-Augmented Generation), a framework that reduces hallucinations in medical AI by combining text and visual data from retrieved images.

• Reduces hallucinations in chest X-ray report generation by up to 85%
• Improves medical image captioning with better factual accuracy
• Maintains strong performance while reducing false information
• Demonstrates effective integration of multimodal retrieval techniques

For healthcare applications, this advancement addresses a critical safety concern, as AI-generated medical content must maintain high accuracy standards for clinical use. The approach enables more reliable AI assistance in medical imaging interpretation.

Reducing Hallucinations of Medical Multimodal Large Language Models with Visual Retrieval-Augmented Generation