Dermatologists must be ‘actively involved’ in AI development to reduce disparities
Key takeaways:
- Melanoma is often diagnosed later and has worse outcomes in patients with skin of color.
- AI could help alleviate this disparity but must be designed with equitability in mind.
ORLANDO — AI can transform how dermatologists diagnose skin conditions including melanoma; however, without attention to equitability, specifically in skin of color, it could exacerbate disparities in care, according to a speaker here.
“Artificial intelligence is rapidly becoming a potential tool that could provide benefits diagnostically in the clinic,” Adewole S. Adamson, MD, MPP, associate professor of dermatology and internal medicine at Dell Medical School at The University of Texas at Austin, told Healio. “But the problem is, a lot of the algorithms have not been designed on populations that include darker skin types.”
While melanoma is much less prevalent in individuals with skin of color and specifically in darker skin types, it is often diagnosed later and has lower survival rates compared with white patients.
AI algorithms have been developed and are being studied and used to diagnose melanoma; however, studies have shown these algorithms decline in efficacy on darker skin types and even dermatologists have a lower ability to diagnose melanoma in darker skin, Adamson said during his presentation at the Skin of Color Society Scientific Symposium.
DermaSensor, an FDA-approved device for AI melanoma detection, was not studied in skin of color prior to approval, with data leading to its approval including only white patients.
The FDA has issued a requirement for post-market testing in underrepresented populations; however, those studies have yet to be published, Adamson added.
Additionally, Google’s AI health tool, which is in use in Europe, was also not studied in diverse skin types.
“The lack of transparency and potential bias is rife in the design of these AI applications and algorithms,” Adamson said during his presentation. “An algorithm is only as good as the inputs that are used to train them and right now, we have a chance to intervene.”
Dermatologists can work with technology industries to develop equitable AI algorithms, partnering with computer and data science experts to include images of darker skin types and to train AI to recognize not only melanoma lesions, but other skin disorders and inflammatory diseases.
“Because dermatology is a visual field, AI is poised to potentially help us on the diagnostic side,” Adamson told Healio. “But this technology, if we aren’t careful, could increase disparities if it’s not developed with equity in mind. Dermatologists need to be actively involved in the development and implementation of this technology.”