A Dermatology AI Trained on 2 Million Skin Images

PanDerm, published in Nature Medicine in 2025, learned from over 2 million real skin images and helped non-specialists read skin photos 16.5% more accurately.

In 2025, Nature Medicine published PanDerm, a dermatology AI built by researchers at Monash University with dermatology centres in Australia, Austria, Italy and Spain. It learned from more than 2 million real-world skin images collected at 11 clinical institutions, across four kinds of skin imaging, including ordinary clinical photographs.

The team tested it on 28 different tasks, from spotting skin cancer to telling apart common and rare skin conditions. It reached state-of-the-art results on every task, often while using only 10% of the labelled examples older models needed.

The most practical finding came from the reader studies. When non-dermatologist healthcare providers used PanDerm on clinical photos covering 128 skin conditions, their diagnoses were 16.5% more accurate. Experienced clinicians also improved their skin cancer accuracy by 11% with its help.

For skin professionals, the message is clear: AI that has seen millions of faces and skin types is now a reliable second pair of eyes. It does not replace the expert in the room, it makes every assessment sharper and more consistent.

Independent research on AI skin analysis in general, not a test of DiemSkin.