This research introduces a novel framework that enables vision-language models to focus on specific regions in medical images through visual prompts, improving diagnostic accuracy.

• Proposes MedVP, a framework using visual markers (arrows, boxes, dots) to guide AI attention
• Comprehensively evaluates various prompt variations to determine optimal visual guidance methods
• Demonstrates improved performance on medical entity extraction and medical visual question answering
• Shows potential for more precise clinical interpretations with region-specific attention

This innovation matters for healthcare by enabling more accurate AI-assisted diagnostics, allowing clinicians to direct AI attention to regions of interest, and potentially reducing diagnostic errors in medical imaging.

Guiding Medical Vision-Language Models with Explicit Visual Prompts: Framework Design and Comprehensive Exploration of Prompt Variations