Vara’s Breakthrough CE Certification and the Next Chapter for Cancer Screening in Europe
For the first time, an AI system has been certified to make independent decisions in breast cancer screening.
This a guest post by Jonas Muff, founder and CEO of our portfolio company Vara.
Vara, a Berlin-based health tech company specializing in breast imaging, has received the only CE certificate on the market for an AI that can serve as an independent second reader in mammography, both for screening and diagnostic use.
Across Europe, national screening programs face a capacity crisis: more women are eligible for breast cancer screening than ever before, while the number of radiologists declines.
The “double-reading” standard, where two radiologists independently interpreting each mammogram, remains best practice. But as workloads rise by 30–40%, this gold standard is at risk of becoming unsustainable.
AI as an independent second reader has been expected for years. Until now, no system had met the regulatory, ethical, and evidence bar to make it a reality.
How Vara got there step by step
Independent AI in healthcare isn’t mainly about writing code. It’s about earning trust.
Vara’s path reflects a decade of deliberate progress through the four milestones that regulators and clinicians have long demanded:
a) Data foundation: Vara partnered with screening centers across Germany, Europe and emerging markets to build one of the world’s largest curated breast-imaging datasets with far more than 15 million biopsy- and 2 year follow-up proven mammograms.
b) Retrospective validation: Early models showed strong performance but, like most AI in medicine, still lived in the “retrospective” world – safe in research, hard to translate into practice.
c) Clinical integration: Vara then embedded its AI as a support system (“concurrent read”) across the German National Screening Program, now covering ~40 % of all centers. This workflow allowed radiologists to use AI safely while building real-world evidence across a whole nation.
d) Real-world prospective evidence: In 2025, Vara initiated and co-authored the PRAIM study, published in Nature Medicine – the largest real-world, prospective AI study in healthcare. Across 460,000 women screened in 12 centers, PRAIM demonstrated statistically significant improvement in cancer-detection rate (+17.6% in cancer detection rate), reduction in false positives (-20%), and meaningful workload reduction for radiologists (-56.7%).
e) Building real-time monitoring: If an AI system takes independent decisions, it is crucial (also as part of the new EU AI Act) to ensure human oversight and understand how real-world adaptations impact the models’ performance. Vara is the only company on the market with a real-time monitoring system built into its software.
Only after that sequence – data, validation, integration, real-world evidence, and accountability – did Vara apply for regulatory recognition of AI as an independent read.
What’s behind the CE certification
The new CE mark, issued under Europe’s Medical Device Regulation (MDR), authorizes Vara’s AI to act as an independent second reader in mammography.
That means:
In a double-reading program, one of the two reads can now be performed autonomously by AI.
In diagnostic workflows, where a second human read is often no longer economically feasible, AI can now restore that safety net.
This is not “decision support.” It is an independent diagnostic interpretation, regulated as a high-risk Class IIb medical device under MDR.
Equally important, the certification acknowledges Vara’s real-time performance-monitoring system, which tracks how the AI behaves across populations, scanners, and time.
That transparency is what makes autonomy possible.
What’s next
Regulatory approval is the beginning, not the end.
For AI to play its full role in screening programs, national guidelines and reimbursement frameworks must evolve.
In Germany alone, 3,000 additional cancers would be found early enough every year. The survival prognosis for early-stage breast cancer is very high (>90%) but drops significantly if found too late. Additionally, 30,000 unnecessary recalls (with enormous emotional pain associated with them) could be prevented.
Germany is uniquely positioned to lead.
Its screening program has just completed a 10-year evaluation confirming significant mortality reduction. Its health-technology-assessment (HTA) framework provides a clear, evidence-based process for evaluating new methods.
The opportunity now is to adapt that process for AI.
If successful, Germany could become the first country in the world to formally integrate AI as a standalone reader in a national cancer-screening program.
A decade ago, AI in radiology was overhyped. Today, it’s finally proven.
And with this CE certification, the field moves from speculation to implementation – responsibly, transparently, and at scale.
 | A guest post by
|