NIH Reporter #1OT2OD038079-01

A Model Editing Framework for Participatory Multimodal AI in Dermatology

This award funds the development of a participatory AI model editing framework in dermatology to improve diagnostic accuracy and stakeholder involvement, enabling dermatologists to correct AI model outputs using natural language.

Description

The proposed research aims to develop new methods that enable highly-skilled dermatologists to edit multimodal AI models to achieve dermatologist-level performance in diagnosing skin disease. This will help primary care physicians in underserved and rural areas in the U.S. to use multimodal AI models for referral decisions based on multiple imaging modalities such as clinical photography and dermoscopy. The project proposes a participatory framework for model editing that allows dermatologists to fix AI model outputs using natural language feedback.

View original record