- Design
- Retrospective multinational development and external validation study
- Population
- 1,974 external patients from 20 cities in 12 countries
- Primary outcome
- Agreement with standard-of-care diagnosis of grade group ≥2 cancer
- Effect
- 83.0% to 87.9% agreement; external AUC 0.84 to 0.89
SCARLET-1 trained an AI system on 20,471 prostate MRIs from 14 countries and tested it on 1,974 patients from 20 further cities in 12 countries, in one diagnostic and two organised screening datasets.
Agreement with standard-of-care diagnosis of grade group 2 or higher cancer was 83.0%, 87.9% and 87.7%, meeting the prespecified definition of interchangeability with an average radiologist in each. External AUCs were 0.86, 0.89 and 0.84. Performance was consistent across age, risk and ethnicity, but varied a lot between sites — 28% of sites had an AUC below 0.80 — and fell with poorer image quality.
This is strong retrospective evidence that AI can support prostate MRI reading, including in screening. The site variation is the caution: a department adopting AI needs to check performance on its own scanners and protocols, and image quality remains the radiologist's responsibility.
- Validate any prostate MRI AI tool on local cases before relying on it
- Monitor image quality, as AI performance drops with poorer scans
- Keep a radiologist read in the pathway while local performance is established
- Track agreement between AI and final histology as an ongoing audit
Why it matters
AI could expand prostate MRI capacity for screening, but only where it works on the local equipment.
Don't overread it
Retrospective validation — it does not show patient outcomes improve when AI is used in practice.
The statistics, in plain English
An AUC of 0.86 means the AI ranks a patient with significant cancer above one without about 86% of the time. 'Interchangeable' was defined against an average radiologist's agreement, not an expert's, and site-level AUCs ranged widely.
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