AIIMS, Google develop AI-powered app to expand skincare, Pg12

Key Highlights:

• AIIMS and Google are jointly developing an AI-powered dermatology app, named Indus Derma, to improve access to skincare.
• The app will analyze uploaded images of skin conditions using computer vision and large vision-language models to suggest a likely diagnosis.
• Indus Derma aims to assist both patients and non-specialist healthcare providers in identifying and managing common skin conditions.
• The app will provide suggestions on next steps, including referral to a dermatologist, local management, or further investigations.
• The target accuracy rate for the app is 80-90%, significantly higher than the 40% accuracy of non-specialists.

Detailed Insights:

• The app will use AI-powered image analysis to compare uploaded images with extensive skin condition databases.
• Dermatology is well-suited for AI applications due to its visual, pattern recognition nature.
• A key challenge is the data imbalance in existing AI dermatology models, which are predominantly trained on images of white skin, affecting diagnostic accuracy for Indian patients.
• AIIMS and Google are developing Indian foundation models to retrain global AI systems using Indian skin tones and dermatological conditions.
• The app is intended as a decision support tool and not a replacement for doctors; medication suggestions require verification by a qualified medical professional.
• The app is expected to be particularly useful for general physicians, healthcare workers, and patients in areas with limited access to specialists.
• Regulatory approvals, accuracy benchmarks, and extensive testing are required before the app's public release next year.

Scientific/Technical Concepts Involved:

• Computer Vision: A field of AI that enables computers to "see" and interpret images.
• Large Vision-Language Models: AI models that process both visual and textual information to understand and generate content.
• AI Foundation Models: AI models trained on a broad range of data that can be adapted for various specific tasks.