In the largest real-world study to date, researchers found that incorporating AI into mammogram screenings helped doctors identify one additional case of breast cancer per 1,000 patients compared to screenings without AI.
The study, published January 7 in Nature Medicine, involved nearly 500,000 women in Germany and showed that AI improved detection rates without increasing false positives.
“AI in mammography screening is at least as good as a human reader, and our study shows it’s even better,” said Alexander Katalinic, a cancer epidemiologist at the University of Lübeck, Germany.
Germany’s breast cancer screening program typically requires two radiologists to independently review each mammogram. If their assessments differ, a third radiologist steps in. This process is intensive, with 24 million images analyzed annually from 3 million participants.
To explore AI’s potential and reduce radiologists’ workload, researchers introduced decision-referral software at 12 screening sites across Germany. Over 460,000 women aged 50 to 69 participated in the study between July 2021 and February 2023. AI software categorized mammograms as normal, suspicious, or unclassified. For half the screenings, radiologists used an AI-supported viewer that displayed the software’s assessment.
AI-assisted screenings resulted in a 17.6% increase in breast cancer detection rates, with doctors identifying seven cases per 1,000 patients compared to six without AI. Additionally, AI slightly reduced false positives, improving overall accuracy.
The study highlights AI’s potential to streamline breast cancer screening, though questions remain about its integration into clinical workflows. One possibility is replacing one of the two initial radiologist reviews, suggested Stefan Bunk, chief technology officer and cofounder of Vara, the Berlin-based company behind the AI.
“This discussion should now start,” Bunk stated, emphasizing the importance of determining how AI can best complement radiologists’ expertise.
