AI improves breast cancer screening and detection rates in real-world research

A real-world study found that using artificial intelligence (AI) for breast cancer (BC) screening improves detection rates without increasing false-positive rates, according to research published in . natural medicine.¹

Woman undergoing mammography | Photo Woman undergoing mammography Image credit: © I YearDesign – Stock.adobe.com

Mammography and early screening have significantly reduced BC and BC-related mortality. However, sensitivity and specificity need to be increased to minimize false-positive results. Multiple retrospective studies have shown that AI has accuracy comparable to, and in some cases superior to, that of radiologists.¹

In the observational, multicenter, real-world, non-inferiority implementation PRAIM study (DRKS00027322), researchers enrolled more than 460,000 women who underwent systematic mammography screening at 12 centers in Germany starting in July 2021. We compared the performance of AI-assisted double reading with standard double reading. Until February 2023.²

The screening test was collected from Germany's mammography screening program, which screens more than 3 million women between the ages of 50 and 75 each year for the early detection of BC. Four 2-dimensional mammograms are taken from each woman and read independently by two radiologists. If anything suspicious is found, a consensus meeting is held with at least two initial readers and the lead radiologist. If concerns are confirmed, the woman is recalled for further tests such as ultrasound, tomosynthesis, magnification imaging, contrast-enhanced mammography, or MRI.¹

Of the screenings, 260,739 samples were screened by AI and 119 samples were screened by radiologists. Data show that radiologists in the AI ​​screening group achieved a detection rate that was 17.6% (95% confidence interval: +5.7%, +30.8%) higher than that achieved in the control group. Recall rates were 37.4/1000 and 38.3/1000 in the AI ​​and control groups, respectively (percentage difference: -2.5% (-6.5%, +1.7%)).¹

Furthermore, the positive predictive value (PPV) of recall was 17.9% in the AI ​​group compared to 14.9% in the control group. PPV for biopsy was 64.5% in the AI ​​group versus 59.2% in the control group.¹

“Our initial aim was to demonstrate that AI-based assessments are equivalent to human assessments,” says Professor, Principal Investigator and Director of the Institute of Social Medicine and Epidemiology at the University of Lübeck and UKSH Lübeck Campus. explains Alexander Katalin, MD. In a press release. “But the results exceeded our expectations. The AI ​​has improved significantly.” [BC] Detection rate. ”³

The study results highlight a vital advance in BC screening and demonstrate the potential of AI to improve detection rates while maintaining low false-positive rates. By demonstrating superior accuracy and efficiency compared to standard methods, AI-assisted screening represents a transformative step towards improving breast cancer early detection and outcomes at scale. These results strengthen the value of integrating innovative technologies into established screening programs to improve standards of care and save more lives.

References

1. Eisemann N, Mukama T, Baltas H, et al. National real-world implementation of AI for cancer detection in population-based mammography screening. natural medicine. January 7, 2025. doi:10.1038/s41591-024-03408-6

2. A prospective multicenter observational study of an integrated artificial intelligence system with live monitoring. Updated on December 1, 2023. Accessed January 9, 2025. https://drks.de/search/en/trial/DRKS00027322

3. AI enhances breast cancer detection in Germany's mammography screening program: results from the world's largest prospective AI study. news release. January 8, 2025. Accessed January 9, 2025. https://www.eurekalert.org/news-releases/1069807