New study published in European Radiology

Can an AI system improve the quality of breast cancer screening? Yes! 

The School of Medicine at the University of St.Gallen is pleased to present a new peer-reviewed, open-access study published in European Radiology, titled “Retrospective evaluation of interval breast cancer screening mammograms by radiologists and AI”.

In collaboration with the Cancer League of Eastern Switzerland and the Mammography Screening Center Paderborn, this research examined whether artificial intelligence (AI) can help detect breast cancers that are not detected during routine screenings but within the following two years —so-called interval breast cancers. Therefore, the study evaluated a large dataset of 151,233 mammograms collected between 2010 and 2019 as part of the organized “donna” screening program in the Swiss cantons of St.Gallen and Grisons. Among these, we identified 268 cases of interval breast cancer, of which 119 were suitable for a retrospective image review by both three experienced radiologists and the AI system ProFound AI. Our findings reveal that in 31.1% of these cases, radiologists retrospectively identified suspicious abnormalities. Importantly, 51.4% of these same cases also received a high-risk score from the AI (case score ≥ 50 out of 100). Even more remarkable: the AI system flagged 13.4% of cases as high-risk where radiologists—both during initial screening and upon retrospective review—had seen no visible signs of cancer. Overall, these results suggest that AI can serve as a valuable second/ third reader in population-wide screening programs, potentially reducing the number of interval breast cancers, which are typically more aggressive, harder to treat, and associated with worse outcomes. The full article is available open-access via European Radiology 

Our research continues: in the next step, we will assess an even larger dataset to determine which AI implementation scenario and thresholds offer the greatest value.