According to a recent study published by Groh et al in Nature Medicine, a novel deep-learning system may help dermatologists diagnose dermatologic diseases more accurately in patients with darker skin tones. Researchers noted that dermatologists often face greater challenges diagnosing the diseases when examining images of darker vs lighter skin tones. In the recent study, 389 dermatologists and 459 primary-care physicians were asked to examine 10 of a total of 364 images of 46 dermatologic diseases such as atopic dermatitis, Lyme disease, secondary syphilis, and cutaneous T-cell lymphoma. Each of them responded with their top three diagnostic predictions and noted whether they would refer the patients for biopsy. The dermatologists achieved overall diagnostic accuracies of 38%, whereas the primary-care physicians achieved overall diagnostic accuracies of 19%. However, when evaluating images of darker skin tones, both groups saw declines in their diagnostic accuracies by four percentage points. Further, the dermatologists and the primary-care physicians were less likely to refer those with darker skin tones for biopsy. Potential factors contributing to these disparities may include a lower exposure to images of darker skin tones in dermatology training materials and less experience diagnosing patients with darker skin tones.
Researchers then introduced a novel deep-learning system capable of analyzing dermatologic images. Having been trained using 30,000 photos, the system had an accuracy rate of 47% in diagnosing dermatologic diseases. The researchers discovered that after utilizing the novel system, the diagnostic accuracies of both dermatologists and primary-care physicians improved by over 33%. Nonetheless, the improvements were greater when diagnosing patients with lighter skin tones than those with darker skin tones. The researchers emphasized that although deep-learning systems may hold the potential to enhance diagnostic capabilities, they may not address disparities across skin tones. They hope their new findings can inspire the wider use of images of dermatologic diseases in individuals with darker skin tones throughout textbooks and other training materials. In a companion press release on the findings from the Massachusetts Institute of Technology, the study authors concluded: “This is one of those situations where [researchers] need empirical evidence to help … figure out how [to] change policies around dermatology education.”