A study conducted by researchers at the Massachusetts Institute of Technology (MIT) found that doctors and dermatologists accomplish lower diagnostic accuracy rates when examining images of darker skin tones compared to lighter ones. Technologically-assisted diagnosis offered greater improvements when assessing lighter skin.
More than 1,000 practitioners, including dermatologists and general practitioners, categorized an array of 364 images representing various skin diseases across diverse skin shades. Dermatologists accurately characterized about 38% of images, but this lowered to 34% for darker skin images. Meanwhile, accuracy rates dropped by around four percentage points for both specializations when diagnosing darker skin.
The study, a first of its kind to highlight this diagnostic disparity, suggested that the use of lighter skin images in dermatology textbooks and other training resources may contribute to the differential. Moreover, some physicians might lack sufficient familiarity with diagnosing and treating darker skin tones.
With the application of an artificial intelligence (AI) algorithm, the researchers observed that overall diagnostic accuracy improved. The AI mechanism had been trained using around 30,000 images to classify images into nine disease categories. The algorithm demonstrated an overall accuracy of around 47% which, when applied, improved accuracy rates for both dermatologists (up to 60%) and general practitioners (up to 47%).
Interestingly, doctors were more inclined to consider the AI’s suggestions once it had provided a few correct answers, but rarely adopted AI recommendations that were incorrect. This implies that while doctors are proficient at excluding diseases, they tend not to accept AI guidance for a disease they have already ruled out. However, the study found that AI assistance provided a more significant boost to general practitioners’ performance when examining images of lighter skin.
The study raises important considerations for the inclusion of AI assistance in dermatology, an area under ongoing development by multiple companies. The researchers hope their findings will encourage more comprehensive training on darker skin patients in medical schools and textbooks, and guide the implementation of AI assistance programs in the field of dermatology.
