
(Vienna, 06-06-2024) A study led by Monash University in Melbourne, the University of Queensland in Brisbane and the Medical University of Vienna has tested the versatility and reliability of the newly developed AI system “PanDerm” in the diagnosis of skin diseases. The results of the study, which have been published in the leading journal Nature Medicine, show that the open source model delivers very good outcomes for a range of different issues and can be a valuable support tool for medical professionals thanks to its diagnostic accuracy and efficiency.
PanDerm was developed by a research group led by Zongyuan Ge from Monash University in Melbourne, who conducted the current study together with H. Peter Soyer from the University of Queensland in Brisbane and Harald Kittler from MedUni Vienna. PanDerm is considered the first system of its kind to simulate the complex reality of dermatological practice based on over two million different medical image sources, including close-ups, dermatoscopic images, histopathological specimens and full-body images. This multimodal approach enables PanDerm to detect not only skin cancer but also numerous other skin diseases with a high degree of accuracy.
System impressed in a wide range of tasks
The accuracy of the system was evaluated in 28 clinical test scenarios and three studies with medical doctors. PanDerm performed well in a wide range of tasks, such as the differential diagnosis of common and rare skin diseases, the early detection of melanomas, the assessment of skin cancer risk, the evaluation of changes in dermatoscopic images, and prognosis estimates – for example, with regard to the likelihood of metastasis. "The model achieves excellent results, even when fed with only a fraction of the data normally required for new tasks," reports co-study leader Harald Kittler from the Department of Dermatology and the Comprehensive Centre of AI in Medicine at MedUni Vienna. The study showed that medical doctors working with the AI system achieved an 11 percent higher accuracy rate in diagnosing skin cancer; non-specialist doctors achieved a 17 percent higher accuracy rate with the tool. Particularly noteworthy: PanDerm detected early-stage melanomas ten per cent more accurately than specialists and identified suspicious skin changes before they were visible to the human eye.
PanDerm is not a ready-made AI-driven decision-making aid, but a flexible open source model that is freely available to software developers and can be adapted for a wide range of tasks. The relevance of this comprehensive system stems from the frequency and variety of skin diseases: around 70 percent of the world's population is affected by one or more of the over 3,000 skin diseases known in dermatology. "PanDerm can be applied to a wide range of dermatological problems and could be particularly helpful in so-called niche problems," says study co-author Philipp Tschandl, also from the Department of Dermatology and the Comprehensive Centre of AI in Medicine at MedUni Vienna. Further studies are planned before the system is put into clinical use.
Publication: Nature Medicine
A Multimodal Vision Foundation Model for Clinical Dermatology.
Siyuan Yan, Zhen Yu, Clare Primiero, Cristina Vico-Alonso, Zhonghua Wang, Litao Yang, Philipp Tschandl, Ming Hu, Lie Ju, Gin Tan, Vincent Tang, Aik Beng Ng, David Powell, Paul Bonnington, Simon See, Elisabetta Magnaterra, Peter Ferguson, Jennifer Nguyen, Pascale Guitera, Jose Banuls, Monika Janda, Victoria Mar, Harald Kittler, H. Peter Soyer, Zongyuan Ge.
DOI: 10.1038/s41591-025-03747-y
https://www.nature.com/articles/s41591-025-03747-y