AI in Prostate Cancer: From MRI Reading to Biopsy Slides and Treatment Decisions

AI is entering every step of the prostate cancer pathway. Each step has a different level of proof, and none of it changes the basic screening debate.

The prostate cancer pathway, step by step

Prostate cancer is usually found through a sequence: a PSA blood test, an MRI scan, a biopsy if the MRI is suspicious, and then a decision about whether to treat or monitor. Each step has well-known weaknesses. PSA is unreliable on its own: the UK National Screening Committee notes that about 8 in 10 men with a raised PSA do not have prostate cancer. MRI reading depends on the radiologist's experience. Biopsy grading varies between pathologists. And many diagnosed cancers grow so slowly that treatment, with its risk of incontinence and erectile problems, does more harm than good.

AI tools now target three of these steps. The table shows where each stands.

Step 1: AI reading prostate MRI

The most important evidence comes from PI-CAI, an international study led by Radboud University Medical Center in the Netherlands and published in Lancet Oncology in June 2024. An AI system was developed on 9,207 MRI exams and tested on 1,000 exams from four centres. In a reader study of 400 of those exams, 62 radiologists from 45 centres in 20 countries (median seven years of prostate MRI experience) read the same scans.

• The AI's AUROC, a measure of how well it separates clinically significant cancer from no or slow-growing cancer, was 0.91, against 0.86 for the radiologists pooled.
• Compared with the original clinical reports, the AI produced about 50% fewer false positives at the same specificity, and detected about 20% fewer slow-growing (indolent) cancers at the same sensitivity. Detecting fewer indolent cancers is a benefit here, because those are the ones most likely to lead to overtreatment.

Several companies sell prostate MRI AI tools with CE marking or FDA clearance. Approval status differs by product, version and country, so ask your hospital which tool it uses and how a radiologist checks the result.

Step 2: AI checking biopsy slides

When a biopsy is taken, a pathologist examines thin tissue slices under the microscope, or increasingly on digital scans, and assigns a Gleason grade. AI can act as a second pair of eyes.

On 21 September 2021 the FDA authorised Paige Prostate through the De Novo route, the first AI software of its kind in pathology. In the study the FDA reviewed, 16 pathologists read 527 digital slide images (171 with cancer, 356 benign), once without and once with the software. With the software, cancer detection on individual slides improved by 7.3% on average, without affecting how benign slides were read. The FDA was clear about the limits: the study did not measure the effect on a patient's final diagnosis, which usually rests on several biopsy cores, and a pathologist must make the final call.

In England, NICE published a medtech innovation briefing on Paige Prostate in 2021. It is a briefing, not a recommendation. NICE noted that there were no published studies of prospective use showing improved patient outcomes, and that pricing was subscription-based, starting at about £1 per slide, with integration costs on top. Hospitals also need digital pathology scanners before they can use such tools at all.

Step 3: AI estimating how aggressive a cancer is

After diagnosis, the hardest question is often whether to treat now, and how intensively. ArteraAI Prostate combines digitised biopsy images with clinical information such as PSA, grade and stage to estimate long-term outcomes. Artera announced FDA De Novo marketing authorisation on 13 August 2025 for patients with non-metastatic prostate cancer, describing it as the first AI software authorised to prognosticate long-term outcomes. The company says its approach was validated using data from several phase 3 randomised trials, and that the test was already available in the US as a laboratory-developed test before the authorisation.

A prognostic score is information, not a treatment instruction. It may help you and your team weigh active surveillance against surgery or radiotherapy, or decide whether hormone therapy adds enough benefit, but it should be discussed alongside your other results. Coverage and price vary; we found no official price that applies across countries (no official data).

Screening: where policy stands

AI does not settle whether healthy men should be screened. Policy remains cautious because of overdiagnosis:

• UK: on 28 May 2026 the UK National Screening Committee recommended screening only for men aged 45 to 61 with a confirmed BRCA2 gene change and a family history of breast, ovarian, pancreatic or prostate cancer, with a PSA test every two years. It did not recommend screening all men, Black men, or men with a family history but no BRCA2 change. The government accepted the recommendation for England, with rollout expected in 2027.
• UK research: the TRANSFORM trial began inviting men on 21 November 2025. Funded with £42 million by the NIHR and Prostate Cancer UK, it is testing combinations of PSA, a saliva genetic test and fast MRI in 16,000 men in its first stage, with at least one in ten invited men being Black. Its results may change future policy.
• EU: the Council Recommendation on cancer screening adopted on 9 December 2022 invites member states to evaluate the feasibility and effectiveness of organised prostate screening based on PSA testing followed by MRI, rather than rolling it out straight away. Approaches differ between countries.
• US: PSA screening is a shared decision between you and your doctor; ask how guidance applies to your age and risk.

Risks and realistic expectations

• Overdiagnosis does not disappear: better MRI reading can reduce detection of slow-growing cancers, but any testing programme still finds some cancers that would never have caused harm.
• Generalisability: AI trained on one scanner type, one population or one lab's slides may perform differently elsewhere. Ask whether the tool was validated in a setting like yours.
• Equity: Black men have a higher risk of prostate cancer. Check whether a tool was tested in populations that include men like you.
• Over-reliance: regulators authorised these tools as aids for specialists, not as replacements.

Questions to ask your doctor

• Was AI used to read my MRI or biopsy, and did a specialist review its findings?
• Is the tool approved for this use in this country?
• Would a prognostic test change the treatment options you would offer me?
• Is active surveillance an option for my cancer, and what would monitoring involve?
• Given my family history and genetics, does any screening programme apply to me?

Questions patients ask