A study from Massachusetts Institute of Technology (MIT) involving over 1,000 doctors, including dermatologists and general practitioners, revealed that accuracy of diagnosing skin diseases based on images is lower when the patient has darker skin. Dermatologists correctly identified 38% of the images they viewed, but this rate dropped to 34% when the images displayed darker skin. The study found some improvement when doctors received help from an artificial intelligence (AI) algorithm, though the enhancement was more notable with lighter-skinned patients.
The researchers suggest that possible reasons for this discrepancy could be that dermatology textbooks and training materials predominantly feature lighter skin tones, and some doctors may have less experience in treating patients with darker skin. This is also the first study to highlight such discrepancies in diagnosis based on skin tone.
To assess diagnostic accuracy, the research team compiled 364 images depicting 46 different skin diseases across a range of skin shades. With the majority of these images featuring one of eight inflammatory skin diseases, each of the study participants was required to make top three predictions about the disease represented in the images. The diagnostic abilities of both dermatologists and general practitioners dropped by about four percentage points when they assessed images of darker skin.
In the latter part of the study, the same participants were provided additional images to analyze with the assistance of an AI algorithm developed by the researchers. The AI tool, trained on about 30,000 images, enhanced the diagnostic accuracy for dermatologists to up to 60% and for general practitioners to up to 47%. However, the overall improvement was less notable with images of darker skin.
These findings highlight a need for more inclusive training materials that better represent patients with darker skin hues. They could also contribute to the development and deployment of AI assistance programs in dermatology, which many companies are currently exploring. The researchers hope to see their discoveries stimulate change within medical schools and textbooks to incorporate more comprehensive training on patients with darker skin. The research was funded by the MIT Media Lab Consortium and the Harold Horowitz Student Research Fund.
