Doctors struggle to accurately diagnose skin diseases in patients with darker skin, an MIT study has found. The study examined the diagnostic success rates of more than 1,000 dermatologists and general practitioners, revealing that dermatologists successfully diagnosed approximately 38% of conditions from images, but only 34% of those presenting darker skin. General practitioners showed similar declines in accuracy with darker skin.
Interestingly, the research revealed that artificial intelligence (AI) could aid physicians in improving accuracy, although gains were greater in patients with lighter skin. The disparity could be due to dermatological textbooks and training resources predominantly featuring lighter skin tones and the lack of experience doctors have with darker skin.
The study, which was published in Nature Medicine, constitutes the first known research work demonstrating physician diagnostic disparities across differing skin tones. Through the study, the research team compiled 364 images from sources including textbooks representing 46 skin diseases across various skin shades. The study included 389 board-certified dermatologists, 116 dermatology residents, 459 general practitioners, and 154 other types of doctors who were presented 10 images from this collection and required to provide their top three disease predictions for each image.
The researchers found that dermatology specialists demonstrated a higher accuracy rate, correctly classifying 38% of images, versus 19% accuracy from general practitioners. Both groups’ diagnostic accuracy fell by 4% when attempting to diagnose skin conditions from images of darker skin.
Following this, the researchers implemented an AI algorithm which showed an accuracy rate of about 47%. This algorithm predicted conditions from a new set of images and offered improved accuracy for both professions, with dermatologists increasing to a 60% accuracy rate, and general practitioners to 47%. Despite these improvements, general practitioners demonstrated greater gains in diagnosing lighter skin than darker skin when provided with AI assistance.
The researchers suggested that findings should motivate medical schools and textbooks to include more training focused on patients with darker skin. The insights could also inform the development and use of AI in dermatology, which is a priority for many companies in the sector. The MIT study was funded by the MIT Media Lab Consortium and the Harold Horowitz Student Research Fund.
