Healthcare (Commonwealth Union) – The detection of melanoma and various other skin conditions is set to become quicker and more precise thanks to a new artificial intelligence (AI) tool that examines several types of medical images at once. The system was developed by a global team of scientists led by Monash University.
Appearing recently in Nature Medicine, the tool—named PanDerm—is among the first AI models specifically designed to support everyday clinical dermatology by processing a range of image formats, such as close-up photographs, dermoscopic scans, pathology slides, and full-body images.
Extensive testing revealed that PanDerm boosted the accuracy of skin cancer diagnoses by 11 percent when used by doctors. It also helped non-specialist healthcare providers improve their accuracy in identifying other skin diseases by 16.5 percent.
The model showed researchers the ability to spot skin cancers at an early stage, flagging suspicious lesions even before doctors noticed them.
Trained using over two million skin images, PanDerm drew data from 11 institutions across multiple countries, covering four distinct types of medical imaging.
AI and computer vision expert Associate Professor Zongyuan Ge, a lead co-author of the study from Monash University’s Faculty of Information Technology, explained that current AI models in dermatology are often restricted to narrow tasks—like identifying skin cancer using dermoscopic images, which are magnified views captured with a specialized tool.
Associate Professor Ge indicated that most existing models fall short when it comes to processing and integrating multiple types of medical images, which limits their effectiveness in real-world clinical environments.
“PanDerm is a tool designed to work alongside clinicians, helping them interpret complex imaging data and make informed decisions with more confidence.”
In contrast to traditional models that are trained for a single purpose, PanDerm was tested across a broad array of clinical applications. These included tasks like screening for skin cancer, estimating the likelihood of cancer recurrence or spread, evaluating skin types, counting moles, monitoring changes in lesions, diagnosing various skin conditions, and identifying lesion boundaries.
Despite using only 5–10% of the usual labelled data, PanDerm consistently achieved top-tier performance.
Within clinical environments, PanDerm serves as a decision-support tool, analyzing the diverse range of skin images typically used by medical professionals. It evaluates these visuals and generates diagnostic probability scores, enhancing clinicians’ ability to interpret images more accurately.
This capability is especially beneficial for improving diagnostic precision among general practitioners, detecting subtle lesion evolution over time, and evaluating patient risk profiles.
Siyuan Yan, the lead author and a PhD candidate at Monash University’s Faculty of Engineering, highlighted that the system’s success was largely due to its multimodal design.
“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 synthesising information from various visual sources,” explained Mr Yan.
“This allows for more holistic analysis of skin diseases than previous single-modality AI systems.”
Researchers of the study indicated that with skin conditions now affecting 70% of people worldwide, early diagnosis is vital and can significantly improve treatment success.
Professor Victoria Mar, Director of the Victorian Melanoma Service at Alfred Health and lead co-author of the study, said PanDerm has potential in identifying slight changes in skin lesions over time, offering insights into their biological behaviour and possible risk of spreading.
Professor Mar pointed out that this type of tool could aid in earlier detection and provide more reliable ongoing assessment for individuals at risk of melanoma.
Professor H. Peter Soyer, Director of the University of Queensland Dermatology Research Centre and a lead co-author of the study, explained that variations in imaging and diagnostic methods may occur because of the differing resources available in urban, regional, and rural healthcare settings.
He indicated that the key advantage of PanDerm is its capacity to enhance and integrate with current clinical practices.
