The ESUR Prostate MRI Working Group has assessed where artificial intelligence stands in MRI-based prostate cancer detection. Its summary is that deep learning tools demonstrate high technical accuracy and that there remains a significant gap between that and actual clinical application.
The group's positions are specific. Deployment should be human-in-the-loop, with AI supporting rather than replacing radiological expertise, explicitly to limit automation bias and legal liability. Prospective validation is needed across multiple vendors, not within the dataset a model was built on. Post-market surveillance is needed to detect algorithmic drift once a tool is running on live scanners. Research priorities named are cost-effectiveness, explainability, and the particular problems of using these tools inside population screening programmes.
The practical translation for a department buying one of these products is that the vendor's accuracy figure is not the question. The questions are whether the tool has been validated on scanners like yours, who is accountable for a missed lesion the tool did not flag, what training the readers get, and how you will know six months in that performance has decayed.
- Ask for prospective validation data on your scanner vendor and field strength, not headline accuracy
- Keep the radiologist as the decision-maker and document it in the reporting workflow
- Plan for post-deployment monitoring before go-live, not after
- Train readers on automation bias explicitly — it is a named risk, not a theoretical one
- Clarify liability for AI-influenced reads with your institution before deployment
Why it matters
It reframes AI procurement from a performance question into a governance one.
Don't overread it
This is a working group special report and consensus position, not a systematic review of tool performance.
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