Royal Free London NHS Trust prospective study shows DERM AI matches specialist accuracy in melanoma detection with 95.8% AUROC

Melanoma incidence is rising faster than any other cancer, yet early-stage diagnosis remains a challenge across NHS skin cancer clinics. With a growing and ageing population overwhelming the clinical workforce, there is high demand for more consistent, scalable, and accurate skin lesion assessment tools to reduce variation in diagnosis and improve patient outcomes.

A prospective study led by Royal Free London NHS Trust across seven UK hospitals evaluated Skin Analytics' CE-marked DERM AI platform on over 1,500 skin lesions. Patients with clinically concerning lesions were photographed using two smartphones and a digital camera with lens attachments; DERM analysed the images and produced a classification compared to histopathologically-confirmed diagnoses.

DERM achieved an AUROC of 91.8%–95.8% depending on the camera used. When configured for 100% sensitivity (catching all melanoma cases), it achieved 65% specificity — comparable to clinical specialists who achieved 69.9% specificity and 77.8% AUROC. More than half of melanoma diagnoses identified were stage 0 or less than 1mm deep, highlighting DERM's ability to detect early-stage lesions.

• AI-assisted diagnosis can match specialist-level melanoma detection accuracy in a real-world NHS clinical setting.
• Configuring AI for 100% sensitivity enables a safety-first approach to triage while still reducing unnecessary referrals.
• Early detection at stage 0–I dramatically improves survival rates (>95%), making AI screening tools potentially life-saving at scale.