CHICAGO — Artificial intelligence (AI) has shown its promise in skin tone analysis for aesthetic outcomes and laser safety, as well as in dermatology resident training programs and other uses, but biased databases and subjective outcomes are slowing advances, speakers said at the American Society for Dermatologic Surgery (ASDS) 2025 Annual Meeting.

Cheryl Burgess, MD, medical director and president of the Center for Dermatology and Dermatologic Surgery in Washington, DC, said internet searches for “beautiful women” show the cultural biases that have tainted databases. Although these searches have produced slightly more inclusive images over the past 10 years, she said, they still heavily feature White women.

Burgess highlighted the cultural biases that lead to AI models being trained with too few images of skin of color (SOC) and noted that more analysis should be done on how the learning models are being developed and trained.

She highlighted a 10-year literature review published last year in International Journal of Dermatology that addressed the problems of datasets that underrepresent SOC in AI programs.

“Current AI programs inevitably do worse at identifying lesions in SOC,” the review authors wrote. “Additionally, only 30% of the programs identified in this review had data reported on their use in dermatology, specifically in SOC. Significant development of these applications is required for the accurate depiction of darker skin tone images in datasets.”

Outcomes Subjective

At the meeting, Vivian Bucay, MD, who practices at Bucay Center for Dermatology and Aesthetics in San Antonio, said that she has found AI helpful for sorting through “most-cited papers” or “high-impact journals.”

But she pointed out that one of the ways aesthetic dermatology is different from disease-based dermatology is the subjectivity of outcomes. Beauty is in the eye of the beholder, which could be the physician, but ultimately, is the patient.

In addition to the lack of data in AI models with SOC, bias creeps in other ways. Bucay noted that pictures can be enhanced with angles and lighting. “We need really well-annotated image datasets. They’re not there. Those are things we need to work on to reduce variability,” she said.

“There really is no universal outcome scale or labeling criteria that would define the ideal aesthetic outcome,” she added. “There’s a lot to do with our judgment and patient satisfaction.”

Real-world validation is another challenge for AI models, Bucay said. “They are trained in the research setting but not in the practical setting. I think we’ve all seen this in our practices: promising tools that didn’t quite pan out to do what they were supposed to do.”

Imperfect Translation in Charts

Jane Yoo, MD, dermatologist and Mohs surgeon who practices in New York City, has had experience testing such tools. She told Medscape Medical News that she experimented for months with an AI scribe program and found mistakes in translations in patient visit summaries. She gave an example of a mistake in translation that could have consequences.

“Anything in your chart can be a legal issue. If it says a patient is ‘depressed’ instead of ‘stressed,’ it’s a huge red flag, and that can affect insurance. The ramifications are huge for the dermatologist because it’s on them. It’s not on the AI company. And they want you to pay $200 a month for this, when I have to make all the corrections?”

Yoo encouraged dermatologists to try AI products and see what works. “You have to figure out where the utility is for you. For a person who is in training, [an AI scribe] could be incredibly helpful because they are still gaining the tools and the vocabulary.” However, she said, seasoned dermatologists “need to have the administrative tasks done for us in a very particular manner. We need an end product that is really finished, where this is still rough around the edges.”

AI is already broadening the use of skin analysis, Yoo said in a presentation at the meeting. AI tools in dermatologist offices can analyze the skin and evaluate its texture and pigmentation. “But now patients are using phone apps to analyze their own skin. “They are looking to their iPhones to see what type of cosmetic treatments they could have and what type of skincare procedures and products they can use,” she added.

“There are many prospects for AI in aesthetic medicine. I think that patients will be able to be more engaged with treatment and services,” Yoo said.

AI may have a role in patient follow-up after Mohs surgery, for instance, to ask how a patient is doing, instead of a medical assistant asking the question, she added. “That’s going to be coming down the pipeline.”

“If you haven’t tried an AI application yet,” Yoo said, “I would encourage you to do so.”

Burgess, Bucay, and Yoo reported having no relevant financial disclosures.

Marcia Frellick is a Chicago-based healthcare journalist and a regular contributor to Medscape Medical News.