Challenging the status quo
And answering with the future of radiology.
Dr. med. Renate Tewaag
About two years ago, we knew that the future of breast cancer screening would include artificial intelligence. But we didn’t quite know how AI would be integrated into the radiologist’s workflow. We needed to deliver both superior AI — and a superior experience for radiologists.
So, from day one, we partnered closely with screening units in Germany who believed in the transformative potential of AI. These development partners - or co-creators, as we like to call them - collaborated with the Vara team to not only design a breast cancer screening platform that would improve overall performance and quality metrics of screening radiologists, but most importantly, always keep screening radiologists in the loop.
The Decision Referral Pathway ensures that each mammogram is analyzed by AI and classified into one of three categories: "normal" (AI is confident there are no suspicious signs), "no classification" (AI is not confident about a classification), or "Safety Net" (AI is confident that it is highly suspicious).
In other words, Vara pre-screens normal mammograms with very high confidence (and pre-fills the structured report) so that screening radiologists can focus their attention on potentially suspicious exams. Vara also post-screens mammograms with very high confidence for potentially missed exams (but only triggered if the reader assigns BI-RADS <3).
Ideally, every classification our AI performs should be understood by three entities: the radiologist, the screening unit, and even the screening program organizers. Understanding how AI works is key to building trust. That’s why Vara provides screening performance metrics in real-time — and why we’re constantly observing the interactions between radiologists and our AI.
We regularly check in with the screening unit and review all of the radiologists’ interactions with Vara. We check how often they overrule the AI’s recommendation, how often the safety net helped them, and how often they disregarded the safety net — checking every screening parameter carefully against relevant clinical insights (e.g. biopsy information).
Consequently, we're able to show how Vara affects a radiologist’s sensitivity, specificity, and PPV. Normally, such performance metrics are only provided sporadically — and only if a country has an organized screening program. Countries without such programs are, quite literally, screening in the dark. Vara can shed some light on important quality indicators.
At the heart of our CE-certified (class IIb), web-based software is our advanced viewer. Vara integrates AI technology and automation directly into a highly-optimized breast-screening-compliant viewer. Not just any viewer. An advanced viewer designed specifically to make breast cancer screening more productive, more intuitive, and in our users’ own words, “delightful.”
Vara can be used across multiple workstations and automates the pre-filling of structured reports via seamless integration with IT systems. It also facilitates single-click reporting for normal exams. Breast density is automatically classified, prior exams are automatically fetched and ready to view, and image annotations in the viewer are all integrated into the simple and intuitive user interface.
Vara pre-screens normal mammograms with very high confidence and automatically fills the report.
Screening radiologists can then focus on potentially suspicious exams.
Vara post-screens mammograms for potentially missed cases, alerting the screening radiologist to take a second look at suspicious, unreported findings. We refer to this as the safety net, which is always running in the background, but only springs into action when the radiologist misses something.
In 2022, we will release a new feature: support for a digital consensus conference — including capabilities enabling radiologists to collaborate on Vara. Together, they'll be able to discuss cases where the double readers could not reach consensus, and decide whether these women should be recalled.
Radiologists gain critical insight by comparing the development of tissue and lesions over time. Likewise, leveraging temporal information will further improve the diagnostic accuracy of AI models. Not only will our AI run on the current exam, but it will check prior exams for signs of cancer — with the promise of further improving screening performance.
Vara is a zero-footprint, end-to-end, all-in-one workflow solution that can easily be accessed from any certified screening workstation — without installation.
Vara handles most of the repetitive, normal exams — empowering screening radiologists to focus on complex and suspicious cases. Currently, Vara can confidently triage around 50% of the entire caseload, automatically pre-filling the accompanying structured report.
Vara alerts radiologists to highly suspicious cases through the safety net. This increases the chance of finding cancers that could have otherwise been overlooked. Vara’s unique monitoring system allows to track each decision — continuously, keeping the screening radiologist in control.