BiomedCoOp introduces a novel approach that automatically learns optimal prompts for vision-language models in biomedical imaging, reducing reliance on manual prompt engineering.

• Validated across 11 medical datasets spanning 9 modalities and 10 organs
• Outperforms traditional manual prompting methods while requiring minimal human intervention
• Achieves state-of-the-art performance while being more efficient than full model fine-tuning
• Addresses the unique challenges of limited annotations in biomedical imaging

This research significantly improves accessibility of AI tools for medical professionals by eliminating the need for specialized prompt engineering expertise, potentially accelerating adoption of AI in clinical settings while maintaining high accuracy.

BiomedCoOp: Learning to Prompt for Biomedical Vision-Language Models