Bradford NHS clinic says AI skin-lesion triage lets it see 32 patients a session instead of 24

A dermatology clinic at St Luke's Hospital in Bradford can now see 32 patients in each session, compared with 24 before it introduced an AI skin-lesion tool, the BBC reported on 5 October. The figures come from Bradford Teaching Hospitals NHS Foundation Trust, which adopted the system in April.

What the trust reports

The tool is DERM (Deep Ensemble for Recognition of Malignancy), made by the London-based company Skin Analytics. According to the BBC's report:

• A healthcare assistant takes three photographs of a suspicious mole or lesion and uploads them; the system analyses them in minutes.
• Patients with benign cases are offered advice and discharged; those identified as suspicious are referred to a specialist in a one-stop clinic.
• "In Bradford, every image is double-checked by a clinician to make sure the diagnosis is correct," the BBC reports.
• The number of unnecessary biopsies has dropped by about 10% since the AI was introduced, "according to the trust".
• The trust gets around 5,000 referrals a year for suspected skin cancer, of which about 8%, or 400 patients, are found to have malignant cancer, again according to the trust.

Consultant plastic surgeon Zakir Shariff told the BBC the technology had had a "big impact" on waiting times and the number of unnecessary procedures. The BBC reports that Bradford was the first trust in Yorkshire to adopt it.

The BBC also describes a case that shows where the human check sits. Laurence Patten, 73, was referred by his GP with a suspicious mole. The AI flagged the images as suspicious; a specialist's examination then suggested the mole was likely benign, and it will be removed and tested to confirm. Dr Nader Ghaderi, the associate specialist who examined him, said the AI was right to flag the case.

The accuracy figure is the vendor's

The BBC reports that Skin Analytics "claims the AI is 99.9% accurate at ruling out melanoma cases". On its own website the company states this more precisely as "99.9 % NPV for melanoma" — a negative predictive value, meaning the share of lesions the system calls not-melanoma that turn out not to be melanoma. The company's performance page puts it in a table of six rows (one NPV row and five sensitivity rows, four of them against a company-set target), drawn from its own post-market surveillance between December 2023 and November 2025. The melanoma NPV row reads 99.9% against a 99% target, on 89,790 lesions. These are the company's figures, not an independent audit.

The count of sites also varies by source. The BBC says DERM "is now deployed in 40 NHS hospital trusts". Skin Analytics' own homepage, read today, says "Deployed in more than 25 NHS trusts". We could not confirm the 40 figure from a primary source.

What NICE has and has not said

The BBC refers to the "National Institute for Clinical Excellence"; the body's current name is the National Institute for Health and Care Excellence (NICE). Its early value assessment, first published on 1 May 2025 and now filed as HealthTech guidance 746 (last updated 17 March 2026), says DERM "can be used within teledermatology services in the NHS during the evidence generation period as an option" for adults referred on the urgent suspected skin cancer pathway. That period is three years, after which NICE will assess whether it can be routinely adopted.

The recommendation comes with conditions and stated gaps:

• It applies only if the evidence in NICE's evidence generation plan is being generated, and once the tool has appropriate regulatory approval, including NHS England's DTAC approval.
• NICE asks for "a healthcare professional review for people with black or brown skin" and regular monitoring of DERM's performance. Its guidance says the evidence "is mostly for skin lesions in people with white skin, but a small amount of data suggests that automated DERM is also diagnostically accurate in people with black or brown skin", and that more is needed "to be certain that automated DERM does not incorrectly detect or miss skin cancer in people with black or brown skin".
• NICE says "it is unclear whether DERM could free up capacity" for non-urgent skin conditions that need face-to-face assessment.
• Its committee noted clinical experts' concern that DERM classifies non-cancer lesions into a limited set of types, and that people with other conditions may be moved to a non-urgent pathway rather than diagnosed at their first appointment.

On the upside, in its May 2025 announcement NICE said early evidence suggests automated DERM "could approximately halve the number of referrals to dermatologists within the urgent skin cancer pathway compared to using teledermatology alone". Its guidance says that estimate applies to eligible lesions.

The UK medicines regulator has looked at this class of tool too, though not this product: the skin-cancer triage tool in Phase 2 of the MHRA's AI Airlock sandbox was a different company's. We covered that report in July: the MHRA sandbox report warned that as a tool becomes more trusted, human review "may become less rigorous".

Why it matters

On our reading, Bradford's numbers are a local, trust-reported snapshot rather than a trial: 32 versus 24 patients a session and about 10% fewer unnecessary biopsies, with a clinician still checking every image. NICE's own guidance says the national evidence on capacity and on accuracy in darker skin is still being gathered, and its recommendation lasts only as long as that evidence is being generated.