Deep learning-aided decision support for diagnosis of skin disease across skin tones

This study investigates the accuracy of dermatologists and primary care physicians in diagnosing skin diseases across different skin tones, with and without the aid of a deep learning system (DLS). A large-scale digital experiment involving 389 dermatologists and 459 primary care physicians from 39 countries showed that specialists achieved 38% diagnostic accuracy and generalists 19%, with both groups showing lower accuracy for darker skin tones. While the DLS improved overall accuracy by over 33%, it exacerbated the accuracy gap for generalists across skin tones. The findings highlight the potential benefits of physician-machine partnerships but emphasize that improved overall accuracy doesn't necessarily address bias.

Matthew Groh, Omar Badri, Roxana Daneshjou, Arash Koochek, Caleb Harris, Luis R. Soenksen, P. Murali Doraiswamy, Rosalind Picard