Giving doctors an AI-powered head start on skin cancer

Detection of melanoma and a range of other skin diseases will be faster and more accurate with a new AI-powered tool that analyzes multiple imaging types simultaneously, developed by an international team of researchers led by Monash University.

Featured in an article published in Nature Medicine, PanDerm is one of the first AI models built specifically to assist with real-world dermatological medical practice by analyzing multiple types of images, including close-up photos, dermoscopic images, pathology slides, and total body photographs.

A series of evaluations showed PanDerm improved skin cancer diagnosis accuracy by 11% when used by doctors. The model helped non-dermatologist health care professionals improve diagnostic accuracy on various other skin conditions by 16.5%.

It also showed the ability to detect skin cancer early, identifying concerning lesions before clinician detection.

Trained on more than 2 million skin images, data for the model was sourced from 11 institutions in multiple countries, across four types of medical images.

AI and computer vision expert and one of the lead co-authors of the research, Associate Professor Zongyuan Ge from Monash University’s Faculty of Information Technology, said existing AI models for dermatology remain limited to isolated tasks, such as diagnosing skin cancer from dermoscopic images; magnified images of skin captured using a dermoscope tool.

“Previous AI models have struggled to integrate and process various data types and imaging methods, reducing their usefulness to doctors in different real-world settings,” Associate Professor Ge said.

“PanDerm is a tool designed to work alongside clinicians, helping them interpret complex imaging data and make informed decisions with more confidence.”

Unlike existing models, which are trained to perform a single task, PanDerm was evaluated on a wide range of clinical tasks, such as skin cancer screening, predicting the chance of cancer returning or spreading, skin type assessment, mole counting, tracking lesion changes, diagnosing a wide range of skin conditions, and segmenting lesions.

It consistently delivered best-in-class results, often with just 5%–10% of the labeled data normally required.

In clinical settings, PanDerm functions as a support tool that analyzes the spectrum of skin images that doctors routinely use. The system processes these images and provides diagnostic probability assessments, helping clinicians interpret visual data with greater confidence.

This integration is particularly valuable for improving diagnostic accuracy among non-specialists, detecting subtle lesion changes over time, and assessing patient risk levels.

First author and Ph.D. student Siyuan Yan from Monash University Faculty of Engineering said the multimodal approach was key to the system’s success.

“By training PanDerm on diverse data from different imaging techniques, we’ve created a system that can understand skin conditions the way dermatologists do; by synthesizing information from various visual sources,” Mr. Yan said.

“This allows for more holistic analysis of skin diseases than previous single-modality AI systems.”

With skin conditions now impacting 70% of the global population, early detection is crucial and can lead to better treatment outcomes.

Lead co-author of the paper, Alfred Health Victorian Melanoma Service Director, Professor Victoria Mar, said PanDerm shows promise in helping detect subtle changes in lesions over time and providing clues to lesion biology and future risk of spread.

“This kind of assistance could support earlier diagnosis and more consistent monitoring for patients at risk of melanoma,” Professor Mar said.

“In hospitals or clinic settings, doctors use diverse ways and different types of images to diagnose skin cancer or other skin conditions.”

University of Queensland Dermatology Research Center Director and one of the lead co-authors of the research, Professor H. Peter Soyer, said differences in imaging and diagnosis techniques could also arise due to different levels of resources available in urban, regional and rural health care spaces.

“The strength of PanDerm lies in its ability to support existing clinical workflows,” Professor Soyer said.

“It could be particularly valuable in busy or resource-limited settings, or in primary care where access to dermatologists may be limited.

“We have seen that the tool was also able to perform strongly even when trained on only a small amount of labeled data, a key advantage in diverse medical settings where standard annotated data is often limited.”

Senior co-author Professor Harald Kittler from Medical University of Vienna Department of Dermatology said PanDerm demonstrated how global collaboration and diverse clinical data can be used to build AI tools that are not only technically strong but also clinically relevant across different health care systems.

“Its ability to support diagnosis in varied real-world settings, including in Europe, is a step forward in making dermatological expertise more accessible and consistent worldwide,” Professor Kittler said.

Though showing promising research results, PanDerm is currently in the evaluation phase before broader health care implementation.

Looking to the future, the researchers aim to develop more comprehensive evaluation frameworks that address a wider range of dermatological conditions and clinical variants.

The team plans to establish standardized protocols for cross-demographic assessments and further investigate the model’s performance in varied real-world clinical settings, with a particular focus on ensuring equitable performance across different patient populations and health care environments.