A NEW study unveils a groundbreaking AI model that could redefine how skin conditions are diagnosed and managed in clinical settings. Unlike traditional AI tools that focus on narrow tasks, such as identifying skin cancer from a single image type, this advanced model is designed to support the full spectrum of dermatological care by analysing diverse image types and clinical data simultaneously.
Developed using over two million dermatological images collected from 11 institutions across multiple countries, the model integrates four key types of imagery: total-body photographs, standard clinical photos, dermoscopic close-ups, and dermatopathology slides. This makes it uniquely capable of supporting a wide range of tasks, from initial screening and risk assessment to in-depth diagnosis and prognosis across hundreds of skin conditions.
This new system was trained using cutting-edge self-supervised learning techniques, allowing it to learn from vast amounts of unlabelled data, an essential advantage in healthcare, where expert annotations are often scarce. It demonstrated exceptional accuracy across 28 benchmark tests, consistently outperforming existing AI models in both skin cancer assessment and general dermatology.
In clinical trials involving human clinicians, the model helped improve diagnostic accuracy, particularly among general practitioners in primary care, by enhancing their ability to identify and differentiate between complex inflammatory conditions, neoplasms, and pigmentary disorders. In some cases, the model alone even outperformed clinician–AI collaborations, suggesting that its diagnostic reasoning is both robust and reliable.
While the model still requires broader testing across rarer conditions and more demographic groups, this study demonstrates the enormous potential of multimodal, clinically integrated AI systems. By mimicking the holistic, multi-layered approach that dermatologists use in real practice, this new AI model represents a significant step toward smarter, more accessible skin care in a wide range of healthcare settings.
Reference
Yan S et al. A multimodal vision foundation model for clinical dermatology. Nat Med. 2025;DOI:10.1038/s41591-025-03747-y.
