Implementing AI DERM technology into the urgent skin cancer pathway at an NHS district hospital

Skin cancer is the highest referring cancer, with a consistent 10% rise in referrals received by the hospital each year. This is coupled with a rising backlog of routine general dermatology work. There are unfilled consultant dermatology posts and a shortfall of workforce for the population. There is limited dermatology training and an increase use of allied professionals referring into the service. The current system is not sustainable. The aims of this study are to implement artificial intelligence (AI) teledermatology into the urgent skin cancer pathway, to enable fewer face-to-face assessments for benign skin lesions, and to improve access to treatment for patients with true skin cancers. Patients eligible for AI teledermatology are referred via their general practitioner (GP) into directly bookable slots at the hospital for photographs and AI DERM assessment. Patients deemed to have a benign lesion (including benign moles, seborrhoeic keratoses, dermatofibromas and vascular lesions) had a ‘second read’ of the photos from a human dermatologist. If the human agreed, the patient was sent a letter explaining the lesion was benign and discharged back to their GP. If the lesion was deemed as ‘not benign’ (including melanoma, squamous cell carcinoma, basal cell carcinoma, actinic keratosis and dysplastic naevus) the patient was offered a face-to-face appointment in a ‘see and treat’ clinic. After an evaluation period of 18 months, the ‘second read’ was removed and an autonomous AI pathway commenced. As of February 2025, 3830 patients were assessed with AI DERM and 1010 were discharged without the need for a face-to-face appointment (26%). In the move to becoming autonomous the discharge rate increased to 37%. One basal cell carcinoma was missed. AI DERM has helped remove benign lesions from the urgent skin cancer pathway. However, it is expensive and a cost analysis needs to be undertaken to truly evaluate the benefits. This will be discussed in further detail.

Abstract from Crossref