Berlin, Germany - 24 October 2024

A new study published in The Lancet Digital Health today highlights the potential of different approaches to using AI to enhance cancer detection and improve efficiency in mammography screening programs.

Conducted across the UK, Sweden, and Germany, the research used data from the national screening programs of the three countries. The comparative study, titled "Strategies for integrating artificial intelligence into mammography screening programmes: a retrospective simulation analysis," is a collaborative effort by Vara, the Department of Radiology at the University of Cambridge School of Clinical Medicine, and the Department of Radiology at The Royal London Hospital.

Using data from nearly 2 million women, the pioneering study revealed the significant impact of AI for breast cancer detection. The research evaluated a range of different AI integration approaches, from offering independent opinions on all screenings, to managing straightforward cases (while reserving complex ones for radiologists), and introducing AI only when radiologists disagree. Each strategy was measured by its influence on detection rates, patient recall for further testing, and its ability to alleviate radiologists’ workloads. These insights provide several pathways for healthcare policymakers around the world to strategically implement AI at scale, ensuring more efficient and effective breast cancer screening, and potentially saving millions of lives.

Owing to strong standalone AI performance, most of the approaches evaluated in the study present viable methods of improving cancer detection and screening program efficiency. In particular, approaches using AI in a manner that combined the strength of AI and radiologists expertise offered the most significant potential to improve screening metrics (up to 8.6% increase in cancer detection) and substantially reduce workload (by up to 84%). These hybrid “decision referral” strategies use AI to make only confident predictions: those for which AI is confident are negative, and those for which it is confident are suspicious, leaving screening mammograms for which it is not assured to be interpreted by radiologists.

Stefan Bunk, CTO at Vara, commented on the findings: "This investigation marks a significant milestone in breast cancer screening. By harnessing the power of AI to complement radiologists' expertise and showing where its effect is greatest, we're opening new avenues for more accurate, efficient, and accessible screening programs. The potential impact on women's health globally is immense."

The AI model utilized in this analysis is commercially available through Vara. Today, Vara screens nearly 40% of women in Germany and is focused on improving access to more accurate, data-driven breast cancer screening in emerging markets such as India.