Recent studies have shown a disproportionate number of dermatological misdiagnoses in patients with darker skin due to a systemic lack of medical training with diverse skin tones.
Skin conditions can present differently on a variety of skin tones. Rashes, for example, do not always appear red on darker skin. Often, they can appear purple, or indistinguishable from the skin itself. Physicians can also often miss rosacea — a skin condition typically characterized by facial redness, bumps, pimples and skin thickening — in darker skinned patients for similar reasons.
The issue of misdiagnosis may extend to life-threatening health conditions as well, such as melanoma, a type of skin cancer. A study conducted in England found that among 20 pictures of different skin conditions, general practitioners correctly identified the picture of melanoma 87% and 93% of the time in pictures of lighter skinned patients, but only 38% and 69% of the time in darker skinned patients.
Melanoma can also appear in different locations in darker skinned patients compared to lighter skinned patients. Physicians primarily scan their patients’ head, neck, back and legs for signs of melanoma. However, African Americans diagnosed with melanoma have a higher prevalence of developing tumors in areas of the body not commonly checked by physicians.
The knowledge gap on the various ways skin conditions present on different skin tones can be attributed to the lack of diversity in photos in dermatology textbooks.
The issue of disproportionate representation repeatedly appears in studies of dermatology textbooks as well. A study published in the Social Science & Medicine journal in April 2018 analyzed 4,146 images from dermatology textbooks and found that light skin tones were overrepresented and dark skin tones were underrepresented in the four surveyed textbooks.
“I thought back to the dermatology training I had received in medical school. I could recall only three occasions when dark skin was used to demonstrate a skin problem,” said Dr. Neil Singh, a senior teaching fellow at Brighton and Sussex Medical School. “Dark skin had only ever been used to demonstrate these dermatological rarities, and never as part of core teaching on common disorders.”
A 2021 study from the University of Pennsylvania shared similar results. ‘Darker’ skin tones made up only 4-18% of the images depicting skin in the textbooks. Dermatological diseases with a racial predisposition were also not commonly represented.
In addition, researchers compared the results of this study with a 2006 study that analyzed the same textbooks. Although 15 years had passed, the researchers did not find a significant increase in diversity levels — only one textbook had a greater than a 1% increase in representation.
To counteract the human bias institutionalized in medical textbooks, technologies such as artificial intelligence (AI) have presented a promising alternative. A 2020 study from Science Translational Medicine measured the ability of an AI to diagnose dermatology diseases. The algorithm attempted to diagnose 26 of the most common skin conditions in over 16,000 individual cases. It made the correct diagnosis 66% of the time, a similar accuracy rate to that of dermatologists (63%) and a superior rate to that of both primary care physicians (44%) and nurse practitioners (40%).
Still, some professionals believe that engineers need to refine AI technology before integrating it into healthcare practices, as researchers have documented several racial issues when testing AI’s ability to identify humans. For instance, in June 2015, a man reported that his Google Photos automatically sorted photos with his African American friend into a folder labeled “gorillas.”
In another case, a graduate student at the Massachusetts Institute of Technology discovered that the A.I. tools from Microsoft and IBM misidentified sex about one percent of the time when analyzing photos of lighter skinned men. However, the error rate increased when the same A.I. analyzed photos of darker skinned subjects.
In addition, clinical research has historically focused on people with light skin. Engineers would have to use this pre-existing knowledge to program the AI, but the lack of information regarding marginalized communities could significantly hinder its ability to diagnose skin conditions in patients with darker skin when symptoms present differently from the AI’s database.
AI is a promising extension for healthcare in the future, with its ability to become widespread and work quickly and efficiently. However, AI will only work as well as its programmed information, and according to some healthcare professionals, the current database is insufficient and requires significant improvement.
“You could take this type of tech and it could have a big role in helping marginalized communities who can’t get to the dermatologist. Potentially, if you could combine different forms of telemedicine and machine vision, you could access more people and make more educated diagnoses,” said Dr. Carlos Charles, a dermatologist who specializes in treating patients with darker skin tones.