A new study from MIT researchers reveals that doctors are less accurate when diagnosing skin diseases in patients with darker skin, based solely on images. The study surveyed more than 1,000 dermatologists and general practitioners, finding that while dermatologists accurately characterized about 38% of the images, they only accurately diagnosed 34% of those featuring darker skin.
It was also found that an artificial intelligence algorithm could improve doctors’ accuracy, however, the improvements were more prominent in diagnosing patients with lighter skin. This may be due to the fact that dermatology textbooks and training materials predominantly show lighter skin tones, and some doctors may have less experience with patients with darker skin.
Matt Groh PhD, the lead author of the study, commented that while doctors likely don’t intend to perform less effectively on certain groups of people, the lack of knowledge and experience could result in decreased accuracy. He expressed the need for empirical evidence to help develop policies around dermatology education.
The study participants were shown images representing 46 different skin diseases across various skin shades. Dermatology specialists were more accurate at classifying the images, identifying 38% correctly compared to general practitioners who identified 19% correctly. However, both groups lost about four percentage points in accuracy when diagnosing conditions based on images of darker skin.
The research team also developed an AI algorithm that had an accuracy rate of about 47%. When used by the doctors, it improved accuracy rates for both dermatologists (up to 60%) and general practitioners (up to 47%). The study found that doctors were more likely to accept suggestions from the AI algorithm when it proved to be correct, but would rarely incorporate AI suggestions they deemed to be incorrect. General practitioners showed a greater improvement on images of lighter skin than darker skin when using the AI assistance.
The researchers hope the findings will encourage medical schools to incorporate more training on patients with darker skin, and that it can guide the development and deployment of AI assistance programs in dermatology. The research was funded by the MIT Media Lab Consortium and the Harold Horowitz Student Research Fund.
