Medical Technology · uk
AI-assisted prostate MRI interpretation: NICE draft supports five tools, with two years to gather further evidence
Identifying suspicious lesions is only the start of diagnosis. NICE proposes allowing the NHS to use five AI tools while evidence is collected. The real test is whether they can accurately identify cancers that need treatment and reduce unnecessary biopsies.
When faced with suspicious findings on prostate MRI, doctors must assess not only whether cancer is present, but also whether further investigation with a biopsy is warranted. On October 23, the National Institute for Health and Care Excellence (NICE) published draft guidance supporting five artificial intelligence tools as options for the National Health Service (NHS) to help detect prostate cancer during a two-year evidence collection period. The recommendation opens the door to clinical use while leaving a question for everyday practice: can AI improve patient outcomes?
These tools mark areas where cancer may be present on prostate magnetic resonance images for radiologists to examine further. MRI interpretation helps assess whether a biopsy or other tests are needed, but suspicious imaging findings alone do not constitute a confirmed diagnosis. Both NICE and reporting by *Pulse Today* emphasize that a radiologist will still review every scan, considering the original images alongside the AI results before making care decisions.
The five tools listed in the draft are AI-Rad Companion Prostate MR, DeepHealth Prostate, ProstatID, Prostate Intelligence (Pi), and QP-Prostate. Two other products, hProstate and mdprostate, were not recommended for NHS funding because clinical or cost-effectiveness evidence was unavailable for assessment. The draft remains open for consultation and must undergo further committee consideration before final guidance is published.
Support for conditional use does not mean the five tools have been shown to consistently improve diagnosis. The NICE committee believes they have the potential to improve the accuracy and consistency of MRI interpretation, particularly in services with limited specialist interpretation experience. However, the available research evidence is limited, and findings vary across studies, leaving insufficient evidence to establish their benefits in routine NHS practice.
Prabhakar Rajan, a professor at Barts Cancer Institute, Queen Mary University of London, noted through the UK Science Media Centre that a gap remains between performing well at identifying or ruling out suspicious abnormalities and improving patient outcomes, and that this gap needs to be examined. He said evaluation should focus on whether the tools can reliably identify clinically significant cancers, reduce unnecessary biopsies, and avoid overdiagnosing slow-growing tumors. Whether performance is equitable across different patient populations and NHS sites also needs to be tested.
These questions will determine AI’s practical value in the diagnostic pathway. Rising demand for prostate MRI is putting pressure on imaging services, but potential efficiencies in interpretation must be weighed alongside subsequent tests, patient outcomes, and overall costs. The two-year evidence collection period must establish not only how many lesions the software can mark, but whether those markings help doctors make more appropriate decisions.