AI-powered tool boosts skin cancer detection accuracy by 11 per cent, study finds

A new artificial intelligence tool has improved doctors’ ability to detect skin cancer by 11 per cent, according to research from an international team led by Monash University.

The system, named PanDerm, is designed to analyse multiple types of medical images simultaneously to help identify melanoma and other skin diseases more quickly and accurately than current methods.

With skin conditions now affecting 70 per cent of the global population, researchers say the technology could prove especially useful for early detection, which is crucial for better outcomes.

PanDerm was developed using more than two million skin images from 11 institutions across several countries.

It can analyse a wide range of visuals, including close-up photographs, dermoscopic images (taken with a magnifying tool), pathology slides, and total body photographs.

When tested, the tool helped non-dermatologist healthcare professionals improve diagnostic accuracy across a range of skin conditions by 16.5 per cent.

It also flagged suspicious lesions before they were identified by clinicians.

Associate Professor Zongyuan Ge is from Monash University’s Faculty of Information Technology.

The researcher said: “Previous AI tools have struggled to integrate and interpret multiple data types, limiting their usefulness in real-world clinical settings.

PanDerm is designed to work alongside clinicians, helping them interpret complex imaging and make decisions with greater confidence.”

Unlike traditional AI models designed for single tasks, PanDerm was evaluated across several clinical applications—including skin cancer screening, predicting recurrence risk, assessing skin type, mole counting, tracking lesion changes over time, lesion segmentation, and diagnosing a wide range of skin conditions.

In many cases, it delivered best-in-class results with just 5–10 per cent of the labelled data usually required.

Rather than replacing clinicians, PanDerm supports routine diagnostic workflows by processing image types commonly used in dermatology and providing diagnostic probability assessments.

This approach can enhance accuracy, particularly for non-specialists, and help detect subtle lesion changes over time.

First author Siyuan Yan is a PhD student in Monash University’s Faculty of Engineering.

Yan said: “By training PanDerm on diverse imaging modalities, we’ve created a system that mimics how dermatologists synthesise information from different visual sources.

“This allows for more holistic analysis than previous single-modality AI tools.”

Adjunct Professor Victoria Mar, director of the Alfred Health Victorian Melanoma Service, said the system could help identify subtle lesion changes and signs of future spread.

She added: “This kind of support could improve early detection and provide more consistent monitoring for melanoma patients.”