DermaSynth addresses the critical shortage of training data for medical vision-language models by generating over 92,000 synthetic image-text pairs for dermatology applications.

• Leverages 45,205 dermatology images (both clinical and dermatoscopic) to create a comprehensive dataset
• Uses state-of-the-art LLMs and self-instruct methods to generate diverse and clinically relevant text descriptions
• Enables fine-tuning of specialized models like DermatoLlama for dermatology tasks
• Demonstrates how synthetic data can overcome privacy and scarcity challenges in medical AI

This research significantly advances the development of AI tools for dermatology diagnostics and education, potentially improving access to specialist-level skin disease assessment in underserved regions.

DermaSynth: Rich Synthetic Image-Text Pairs Using Open Access Dermatology Datasets