NICE has backed five AI tools that help radiologists read prostate MRI scans, but the "recommendation" in this week's headlines is draft guidance still in consultation, and the consultation closes on 12 November 2026. The tools may be used with NHS funding for a two-year evidence-generation period. For diagnostic AI companies, the evidence generation plan matters more than the headline.

What NICE proposed

The draft covers AI-Rad Companion Prostate MR, DeepHealth Prostate, ProstatID, Prostate Intelligence and QP-Prostate. Each tool marks areas of an MRI where cancer may be present, and assigns a risk classification. Every scan would still be read by a radiologist.

NICE did not back hProstate or mdProstate. Coverage differs on why: one report says they need more research, another that there was no clinical or cost-effectiveness data for either.

NICE's topic page describes the problem the tools address. It notes that reading and reporting scans is time-consuming and accuracy can vary with radiologist expertise, at a time of rising demand and workforce shortages.

Conditional use, not routine adoption

The committee judged the current evidence "limited and inconsistent". That is why use is tied to a two-year period in which data are collected. The draft went through two rounds: the first draft guidance consultation ran from 11 August to 1 September 2026, and the second from 23 October to 12 November. A final publication date has not been set.

The commentary shows where the evidence is thin. Prof Prabhakar Rajan of the Barts Cancer Institute said it is important to distinguish strong performance in flagging or ruling out suspicious areas from proof of better patient outcomes. The gaps listed include:

  • Clinical value. Whether the tools reliably detect clinically significant cancers, and reduce unnecessary biopsies and overdiagnosis of indolent tumours.
  • Pathway effects. Impact on biopsy numbers, time waiting for biopsy and diagnosis, and radiologist capacity and workflow.
  • Equity. Whether performance holds across patient groups and NHS settings.

The template: the fracture-detection plan

The closest precedent is NICE's evidence generation plan for AI fracture detection on X-rays. It recommended four tools for use in the NHS while more evidence is generated, and it sets the terms a prostate plan is likely to follow:

  • The companies carry the burden. They are responsible for ensuring data collection and analysis takes place.
  • Study design is stated. Well-conducted randomised controlled trials are the preferred source for comparative effects, if they can address the gaps.
  • Withdrawal is possible. NICE will withdraw the guidance if the companies do not meet the monitoring conditions.
  • A second review follows. After the period, companies submit evidence in a form that can be used for decision making, and NICE assesses whether the tools can be routinely adopted.

The plan also says commissioning and procurement guidance will come from NHS England. NICE's recommendation therefore leaves open how trusts will buy the tools.

What it means for market access teams

The coverage of the draft gives no detection rates or costs, so the public record does not yet show the economic case for any of the five tools. Teams should work from what the decision shows.

  • Plan the evidence before the funding starts. Biopsy rates, time to diagnosis, patient outcomes and subgroup performance are all on NICE's list. Data capture has to be built into trust deployments from the first scan, not added in year two.
  • Treat the trial question early. If randomised evidence is the stated preference, decide now whether a pragmatic trial, a staged rollout or an in-service evaluation can answer it.
  • Price for the review. Two years of NHS funding is a bridge to a second NICE assessment, not a market. The pricing model should hold up when the committee sees real-world resource use.
  • Watch the commissioning route. In the fracture precedent, procurement guidance sits with NHS England, so a NICE recommendation alone may not open trust budgets.

Companies NICE did not back face a different task: producing the evidence that would support a future review.

To discuss how an evidence generation plan fits your diagnostic or AI product, get in touch.