“Is AI Coming for My Job?”: Familiarizing Clinicians with AI Tools and Their Role in Dermatology

At the 5th Annual Elevate-Derm West Conference this week in Scottsdale, Arizona, Eileen Cheever, MPAS, PA-C, presented a session titled, “AI in Dermatology.” The lecture focused on how artificial intelligence (AI) can be utilized within dermatological practice, while also recognizing the limitations of current technology.

TRANSCRIPT

Cheever: My name is Eileen Cheever. I'm a dermatology PA out of Central Massachusetts. I practice at ClearView dermatology and I've been a dermatology PA since 2008. I see patients in a family dermatology practice settings - so all ages, all stages!

Dermatology Times: What topics did you cover during your session titled, "AI in Dermatology"?

Cheever: I was so thrilled to be able to present a lecture on AI in dermatology at Elevate in Scottsdale. It's such an interesting topic that so many of my colleagues have questions about. “What is it?” Is the number one question that I get a lot. “What really is AI, and how can I use it? Is AI coming for my job?” I always tell my colleagues: No, AI probably isn't coming for your job, but somebody who's using AI probably will, so it's probably a good idea to get familiar with what it is and how it's currently being used in dermatology.

My lecture really focused on being able to become more knowledgeable about what's out there and how AI is being used, but also to leverage that information in understanding that there's definitely limitations. There can be pitfalls and drawbacks of using AI and really making sure that the audience was aware of that. One of the most interesting topics that we covered during the course of my lecture was a publication that was in the Journal of the American Academy of Dermatology, in the editorial section actually submitted by a group out of UMass and I am a Massachusetts based practitioner. This study was published in February of 2024, and it assessed the ability of an AI chatbot to translate a dermatopathology report into patient friendly language. They took those biopsy reports that we see every day in the clinic, fed that report into an AI chatbot, and asked it to translate it into language that patients could understand. They took a dozen of those reports, and they involved some really common neoplastic or inflammatory skin disorders, fed that into ChatGPT with a specific request to simplify the report into patient friendly language. Then they had 30 physicians rate the translation that ChatGPT spit out and how accurate it was. What they found across all the diagnoses, whether it be neoplastic or inflammatory in nature, most of the physicians on the review panel agreed or strongly agreed, that the ChatGPT translations were complete. They were accurate, they were understandable, and, most importantly, unlikely to cause harm. A very interesting tool to possibly think about using if you're having those conversations with your patients. How do I explain this complex biopsy report in a way that the patient can understand? Perhaps using AI to help you do that could make the communication process a little bit more simple, very, very interesting. One of the topics that got some of the most robust feedback after the lecture.

What I think was also interesting that we covered during the course of the lecture was artificial intelligence in dermatology, challenges, and advancements in skin of color. This was a literature review from April 2024 and it seeked to identify and address gaps in AI usage in dermatology, but specifically regarding skin of color. One of the biggest challenges that we face across the board right now is so many of the diagnostics that are out there are based in a somewhat archaic Fitzpatrick skin type system, and we know the images that are out there right now for AI skin of color is largely underrepresented when these AI chatbots or these training data sets are being done. These programs, these AI chatbots, are actually being trained on lighter skin tones. Image quality, accuracy, lack of standardization in terms of data sets and the photos and the data sets are really a huge issue when it comes to AI accuracy with skin of color. Really the paper focuses on solutions to this and really identifying that there is a need for a customized approach for skin of color when we're using AI. We want to avoid data input biases, especially for these AI chatbots when they're undergoing that training. Really diversifying the photos that they're trained on, and, of course, train our clinicians who are taking these photographs to properly photograph skin of color, provide more diverse data sets, and having those AI software systems disclose if their software was trained on a diverse amount of skin tones with the images that chatbots are using. Very, very interesting area of research as well.

[This transcript has been edited for clarity.]