Revolutionizing Skin Cancer Detection with AI: A New Tool for Doctors

Recent advancements in artificial intelligence (AI) are transforming the landscape of dermatology, particularly in the early detection and diagnosis of skin cancer. Led by researchers from Monash University, an international team has developed PanDerm, a groundbreaking AI-powered system that analyzes multiple types of skin images simultaneously to assist clinicians in making more accurate diagnoses.

Published in the prestigious journal Nature Medicine, PanDerm is one of the first AI models tailored explicitly for real-world dermatological practice. Unlike previous models limited to analyzing individual image types such as dermoscopic photos, PanDerm integrates diverse data sources—including close-up photographs, dermoscopic images, pathology slides, and full-body images—to provide a comprehensive assessment.

Evaluations of PanDerm demonstrated notable improvements in diagnostic accuracy, enhancing skin cancer detection accuracy by 11% when used by medical professionals and increasing diagnostic performance for various skin conditions by 16.5%, especially when used by non-dermatologists. The system also shows promise in early detection, identifying problematic lesions even before they are visibly detectable by clinicians.

Trained on over 2 million images from 11 institutions across multiple countries, PanDerm harnesses a rich dataset that enables it to understand complex skin conditions better. Associate Professor Zongyuan Ge, an expert in AI and computer vision at Monash University, highlighted that existing AI tools often focus on isolated tasks, limiting their utility. In contrast, PanDerm’s multimodal approach mimics the way dermatologists evaluate skin, leveraging diverse imaging techniques to enhance understanding and accuracy.

Designed as an aid for clinicians, PanDerm supports various clinical tasks including skin cancer screening, assessing recurrence risk, skin typing, mole counting, lesion tracking, and segmentation. The system’s ability to function effectively with minimal labeled data makes it especially valuable in resource-limited settings.

For example, in primary care or busy hospitals where specialist dermatologists may be scarce, PanDerm can assist general practitioners by providing diagnostic probabilities and highlighting subtle changes over time, thus supporting earlier detection and intervention.

Senior co-author Professor Harald Kittler from the Medical University of Vienna emphasized that the collaborative nature of the data used in developing PanDerm ensures its relevance across different healthcare systems worldwide. While still under evaluation, the researchers plan to expand the system's capabilities, including broader condition coverage and ensuring equitable performance across diverse populations.

This innovation represents a significant step towards making dermatological expertise more accessible and consistent globally, ultimately aiming for earlier diagnosis and better outcomes for patients at risk of skin cancer.

Source: MedicalXpress