First EU‑Certified AI Triage: A Leap in Breast Cancer Screening

AI DiagnosticsAI TherapyAI & Machine Learning

Imagine a future where a computer can instantly decide which mammograms need a human eye—Vara’s new AI tool just made that a reality in Europe.

TL;DR
• Vara’s AI triage tool receives first EU Class IIb CE certification, enabling autonomous clearance of normal mammograms.
• The system includes ATMON for continuous human oversight, meeting EU AI Act requirements.
• Evidence from the PRAIM study shows improved detection rates and comparable recall rates, supporting rollout.

Summary
Vara, a Berlin‑based AI firm, received EU Class IIb CE certification for an autonomous breast‑cancer triage system that clears normal mammograms without radiologist input and forwards only suspicious cases for expert review. The clearance marks the first EU‑approved autonomous AI tool for breast‑cancer screening and aims to mitigate the growing shortage of radiologists in population‑based programs.
• Deploy ATMON monitoring to track hardware changes, system health, and performance signals, reverting to full radiologist reads when thresholds are breached.
• Leverage seven years of model monitoring data to demonstrate stability and compliance with the EU AI Act’s high‑risk oversight requirements.
• Align rollout with national guidelines, noting that over 50% of Germany’s organized screening programs already incorporate the system.
• Use real‑world evidence from the PRAIM study, which showed a detection rate increase from 5.7 to 6.7 per 1,000 women and a non‑inferior recall rate, to support clinical adoption.
• Establish clear redress pathways for clinicians to report anomalies, ensuring continuous human oversight and transparent performance reporting.

The PRAIM observational study, published in Nature Medicine, provided the evidentiary basis for certification, reporting a breast‑cancer detection rate of 6.7 per 1,000 women in the AI‑supported group versus 5.7 in standard double reading, with recall rates of 37.4 versus 38.3 per 1,000 women and a higher positive predictive value (17.9% vs 14.9%). ATMON continuously monitors site‑specific operating points, detecting deviations in mammography hardware or algorithmic drift and automatically triggering a fallback to full radiologist reads when predefined limits are exceeded.

Implementation will depend on country‑specific clinical readiness and guideline alignment, but Vara notes that more than half of Germany’s organized screening programs have already adopted the AI tool. The company emphasizes that ongoing oversight, transparent performance metrics, and a structured mechanism for addressing performance shifts are essential to maintain patient safety and trust while delivering the efficiency gains promised by autonomous AI.

Will autonomous AI triage become the new standard for breast cancer screening, balancing efficiency with patient safety?