- Design
- Retrospective multicentre, multivendor diagnostic accuracy study
- Population
- 787 men with prostate MRI and histological verification, 2014–2025
- Primary outcome
- Sensitivity for Grade Group ≥2 cancer, AI vs PI-RADS
- Effect
- Patient level 97.6% vs 92.6% (non-inferior); lesion level 78.8% vs 88.8% (not non-inferior)
A retrospective multicentre, multivendor diagnostic accuracy study in European Radiology, published on 26 September, compared a commercially available AI risk-classification system with radiologist PI-RADS scoring in 787 men who had prostate MRI and histological verification; 48% had clinically significant cancer (Grade Group 2 or above).
Per patient, AI sensitivity was non-inferior to PI-RADS (97.6% vs 92.6%), and discrimination was slightly higher (AUC 0.80 vs 0.77; difference 0.032, 95% CI 0.001 to 0.064). Per lesion, AI sensitivity was lower (78.8% vs 88.8%) and not non-inferior. Within PI-RADS 3 lesions, the AI upgraded 93% of those that proved significant cancer and labelled 24% of benign ones low risk; a strategy applying it only to PI-RADS 3 avoided about 19% of biopsies while keeping sensitivity at 98.4%.
The practical role is narrow and useful: a second opinion for the equivocal PI-RADS 3 lesion, where biopsy decisions are hardest, while the radiologist still maps every lesion for targeting. The lesion-level shortfall means it cannot replace that mapping.
- Keep radiologist lesion mapping for biopsy targeting; AI missed more individual lesions.
- Consider AI risk output as decision support for PI-RADS 3 lesions, where it best separated significant cancer from benign.
- Report AI and PI-RADS separately so urologists can see where they disagree.
- Audit local performance before relying on any AI threshold; this was a retrospective study.
Why it matters
PI-RADS 3 is where most unnecessary biopsies come from, and this is where the AI added most.
Don't overread it
Retrospective and enriched for cancer (48%); the biopsy-avoidance figure is modelled, not observed.
The statistics, in plain English
Non-inferior means the AI was not meaningfully worse than PI-RADS by a pre-set margin. The AUC difference of 0.032 has a lower bound of 0.001 — statistically significant, clinically small. With half the men having significant cancer, predictive values will be different in a typical referral population.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free