Vara, a Berlin-based AI company focused on population-based early detection of cancer, has been granted CE certification under the EU Medical Device Regulation (Class IIb) for an autonomous AI tool for breast cancer screening. The certification marks the first such clearance for autonomous AI in breast cancer screening.  

The tool triages mammograms without radiologist involvement, classifying normal scans and routing abnormal scans to a radiologist for review. The company noted that the tool is intended to help address a shrinking radiology workforce in screening programs where mammograms are typically read twice.  

Vara’s tool includes a safety system called ATMON (Autonomous-Triage Monitoring), which supervises the autonomous AI models. With ATMON, each site’s operating points are set and monitored, with the system able to track changes in mammography hardware, system health, and daily performance signals. If needed, the system can revert any site to full radiologist reads whenever a signal moves outside defined limits. This system provides the continuous human oversight required by the EU AI Act for high-risk AI systems. The company noted that it has monitored its AI models for 7 years.  

"Studies have shown that AI can improve cancer detection. But even the best AI model's performance can shift once it operates outside the controlled conditions of a study," said Jonas Muff, CEO and co-founder of Vara. "For AI to deliver its positive impact on patient outcomes, we need to notice when this happens. We believe our ATMON system can be a template for deploying AI safely across healthcare." 

Supporting Evidence 

A large, observational, real-world study evaluating AI-supported double reading in population-based mammography screening in Germany provided supporting evidence for the certification. The PRAIM study, published in Nature Medicine, reported a breast cancer detection rate of 6.7 per 1,000 women in the AI-supported screening group vs 5.7 per 1,000 women in the control group, which underwent standard double reading.  

The recall rate in the AI-supported group was noninferior to that in the control group, at 37.4 vs 38.3 per 1,000 women, respectively. The positive predictive value of recall was 17.9% in the AI-supported group vs 14.9% in the control group. 

The rollout of Vara AI will depend on national guidelines and country-specific clinical readiness. The company noted that more than 50% of Germany’s organized breast cancer screening programs already use Vara AI. 

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.