A new artificial intelligence (AI) model developed by an international research team is poised to improve the diagnosis of skin conditions, offering support for both dermatologists and non-dermatology professionals. The model, named PanDerm, is designed as a clinical decision support tool to assist with skin cancer detection and the assessment of various skin conditions. Capable of processing multiple skin images at once, it delivers diagnostic probability assessments, which could enhance diagnostic accuracy across a wide range of clinical tasks.
The team behind PanDerm, led by AI and machine learning experts from Monash University, trained the model on over two million skin images sourced from 11 medical institutions worldwide. These images included close-up photos, dermoscopic images, pathology slides, and total body photographs. PanDerm was trained to detect skin cancers, assess risk, track changes in moles, and even identify a variety of other skin conditions. It can also predict the recurrence and spread of skin cancers and evaluate the characteristics of lesions.
The model demonstrated significant potential in clinical performance. In a series of diagnostic tests, PanDerm outperformed clinicians in detecting early-stage melanoma, the most aggressive form of skin cancer, by 10%. Additionally, it helped dermatologists improve their diagnostic accuracy from dermoscopic images by 11%, reaching an 80% accuracy rate. Non-dermatologists, including general practitioners and clinical assistants, also benefited from PanDerm’s assistance, showing a 16.5% improvement in identifying and distinguishing conditions like inflammatory dermatoses and pigmentary disorders.
PanDerm proved to be superior to existing AI models in various tasks, including assessing skin cancer risk, detecting malignancy, and predicting metastasis, even when trained with only a fraction of the labeled data typically required. Researchers believe this is a significant step forward, particularly in primary care settings, where access to dermatologists is often limited. According to the team, PanDerm has the potential to fill gaps in dermatological expertise, supporting clinicians in diagnosing skin conditions more accurately.
The AI model’s ability to process multiple types of skin images and integrate them into a cohesive diagnostic framework sets it apart from previous AI models, which have struggled to support various clinical workflows. Monash University associate professor Zongyuan Ge noted that PanDerm’s integrated approach could be particularly valuable in busy or resource-limited healthcare settings. This integration allows for more seamless support in the diagnosis and management of skin conditions, even when access to specialist care is not readily available.
The research also highlighted PanDerm’s potential to detect melanoma at an earlier stage, which could significantly improve outcomes for patients. Victoria Mar, a lead co-author of the study, emphasized the importance of early detection, especially for aggressive cancers like melanoma, where timely intervention can be life-saving.
The team plans to conduct further clinical evaluations to ensure that PanDerm performs consistently across different patient populations and healthcare environments. As AI continues to transform medical diagnostics, the researchers believe PanDerm could play a key role in improving skin disease care, particularly in settings where dermatologists are scarce.
This development comes amid a broader trend of AI integration in skin cancer detection. In Australia, the world’s first AI-powered pop-up skin care clinic has been established to help detect melanoma early. This initiative, led by the health charity Skin Check Champions, aims to reduce the number of skin cancer diagnoses by half while increasing the number of screenings by 25%. Meanwhile, in South Korea, the first locally developed AI-powered smartphone app for skin cancer diagnosis has been approved, signaling growing interest in AI-driven solutions for skin health worldwide.
Related Topics:
