prostate-cancer

AI diagnostic system matches radiologist accuracy for prostate MRI—and could handle half of all scans independently.

This simulation study evaluated whether an AI system could autonomously diagnose a subset of prostate MRI examinations for clinically significant prostate cancer (csPCa), reserving uncertain cases for radiologists. Using data from 500 men across four European centers, AI thresholds were calibrated…

Dan Manning

Dan Manning

— 1 min read
Share
AI diagnostic system matches radiologist accuracy for prostate MRI—and could handle half of all scans independently.

AI diagnostic system matches radiologist accuracy for prostate MRI—and could handle half of all scans independently.

This simulation study evaluated whether an AI system could autonomously diagnose a subset of prostate MRI examinations for clinically significant prostate cancer (csPCa), reserving uncertain cases for radiologists. Using data from 500 men across four European centers, AI thresholds were calibrated and then tested on 400 examinations.

The AI-driven pathway maintained comparable sensitivity (89.0% vs. 89.4% for radiologists) but significantly improved specificity by 11.5% (69.2% vs. 57.7%). The AI autonomously diagnosed 195 of 400 examinations (49%) with 94.7% sensitivity and 94.7% specificity on its autonomous subset.

This triaging model could reduce radiologist workload by nearly half without compromising diagnostic accuracy.

Key Findings

  • AI maintained equivalent sensitivity (89.0%) to radiologists for prostate csPCa
  • AI improved specificity by 11.5% (69.2% vs. 57.7%)
  • AI autonomously handled 49% of examinations with 94.7% sensitivity and specificity
  • 62 radiologists from 4 European centers included in comparison
  • Simulation study (NCT05489341)

Implications

AI-assisted prostate MRI triage could substantially reduce radiologist workload while improving specificity. This has major implications for scaling prostate cancer screening programs globally. Clinical implementation studies are the next critical step.

Caveats

Simulation study with retrospective data; abstract-only. AI performance on autonomous subset may reflect selection bias (cases where AI is most confident). Real-world performance may differ. 4 European centers may not represent all clinical settings.

Source: Radiology. Imaging cancer — 2026-05-01

Read more