Artificial intelligence in dermatology: current applications and future horizons

Tobias Sangers (Leiden University Medical Center, the Netherlands) is a PhD candidate investigating artificial intelligence (AI) applications in dermatology. He shared some of the latest evidence on AI in dermatology and explained the possible applications for dermatologists, now and in the future.

“We see that generative AI applications, like ChatGPT, are booming, with over 100 million users within 2 months after its launch,” said Mr Sangers [1]. Many developments in image- generating AI are emerging. With the application SORA, AI can produce life-like videos based on a short line of text. “And we are only at the start of the AI revolution, with many developments to come.”

“Within the field of medicine, more and more applications of AI are emerging every day,” continued Mr Sangers. He showed a study in which ChatGPT was compared with internal medicine specialists and residents for solving clinical cases [2]. AI was significantly better at solving clinical cases than the specialists and residents. “This is an example illustrating that AI can be an asset in the world of medicine,” according to Sangers.

From diagnosis to patient care: the expanding role of AI in dermatology

In dermatology, AI can help us with diagnostics, prediction, and patient interactions. “The advantage of AI with regard to diagnostics in dermatology can be found in image- processing algorithms or deep neural networks,” expressed Mr Sangers. For example, such an algorithm could receive a melanoma-like image as input and decide whether it is, in fact, a melanoma or not. “These deep neural networks are black boxes because we are unsure which steps are taken before the output is generated,” added Mr Sangers.

One study assessed the performance of diagnostic AI in dermatology by comparing its ability to diagnose suspicious melanoma-like skin abnormalities with that of dermatologists. It also examined the combined accuracy of both AI and dermatologists. The results showed that the combined diagnostic accuracy of the dermatologist and AI was significantly higher than that of the dermatologist alone (AUC 0.97 vs 0.90; P=0.005; see Figure) [3]. “This study provides evidence for the use of AI to improve diagnostics in clinical practice,” concluded Sangers. However, there are differing viewpoints and findings. Mr Sangers also shared a small study that showed that the accuracy of ChatGPT Vision in diagnosing melanoma is not yet ready for clinical practice [4]. “I would recommend against using ChatGPT Vision for diagnostic purposes yet, but to stick to specialised tools.”

Figure: Diagnostic accuracy of AI versus dermatologists [3]



AUC, area under the curve.

Next, it was explained that the traditional approach for predicting whether a person may get a skin disease is based on risk factors such as age, skin type, BMI, genetic factors, and smoking. A recent study demonstrated that AI could extract risk factors from participants’ facial images. Moreover, estimating the risk of skin cancer based on AI-processed facial pictures gave a better predictive value than estimating this risk based on 18 known risk factors [5]. “This was a proof-of-concept study and not yet ready for implementation in clinical practice,” commented Mr Sangers. “Nonetheless, it indicates that AI can help us to select the patients we should follow more carefully or whom we should invite for screening programmes.”

Finally, Mr Sangers discussed the possibility of building a chatbot in OpenAI (the developer of ChatGPT) to improve patient interactions. “In the OpenAI environment, one can make a chatbot that, for example, helps patients with constitutional eczema to answer questions about their condition and helps them to use their home therapy correctly,” outlined Mr Sangers. “In my opinion, AI-produced chatbots can help us inform patients more thoroughly and contribute to achieving better treatment outcomes.” The chatbot can be designed to use specific data, like guidelines or other reliable, high-quality scientific evidence. “This chatbot can be built in about 1 hour and provide personalised advice, taking into account a person’s age and intelligence.”

Concluding his presentation, Mr Sangers emphasised that he had only scratched the surface of the countless AI applications within the field of dermatology. “I hope this helps gain more understanding of where we are in 2024 in the context of AI in dermatology.”

1. Sangers T. AI in de dermatologie. Waar staan we anno 2024? Dermatologendagen 2024, 11–12 April, Amsterdam, the Netherlands.
2. Cabral S, et al.  JAMA Internal Medicine. 2024; April 1. DOI: 10.1001/ jamainternmed.2024.0295.
3. Winkler JK, et al. JAMA Dermatology. 2023;159:621-627.
4. Shifai N, et al. J Am Acad Dermatol. 2024;90:1057-1059.
5. Liu X, et al. EClinicalMedicine. 2024;71:102550.

Copyright ©2024 Medicom Medical Publishers



Posted on 31 May 20244 June 2024