AI in Mammography: Heart Disease Risk, Breast Cancer Screening, and a $16M Trial

Artificial intelligence is being applied to mammography in new ways, including predicting cardiovascular disease risk from breast arterial calcification, assessing breast cancer risk with the BRAIx tool, and a new U.S. clinical trial evaluating AI-assisted reading of screening mammograms.

Researchers analyzed data from 123,762 women who underwent screening mammography at two U.S. health care systems: 74,124 women from Emory Healthcare (average age 55.5 years) and 49,638 from Mayo Clinic (average age 59.5 years). They developed, trained, and validated a deep learning AI model to segment breast arterial calcification (BAC) and quantify risk severity. BAC severity was categorized as zero (0 mm2), mild (>0-10 mm2), moderate (>10-25 mm2), and severe (>25 mm2). BAC was detected in 16.1% of the Emory cohort and 20.6% of the Mayo cohort. Increased BAC predicted increased cardiovascular risk over and above PREVENT scores, confirming it as an independent risk factor. Every square millimeter of BAC increases the risk of CVD by 1%; for example, 20 mm2 of BAC confers a 20% increased hazard ratio. At Emory, a severe BAC score conferred a 2.2 times risk of CVD. Higher BAC severity was associated with a higher prevalence of antihypertensive and statin use, as well as higher prevalence of diabetes mellitus, higher systolic blood pressure, higher body mass index, and lower estimated glomerular filtration rate, but not with higher smoking prevalence. Mammography can potentially be used to screen women at risk of cardiovascular disease without additional radiation exposure, costs, doctor visits, or blood tests.

A new Australian study published in The Lancet Digital Health evaluated an AI tool called BRAIx, trained using BreastScreen Australia data, to help radiologists assess mammograms. Among 95,823 Australian women assessed, 1.1% (1,098) had developed breast cancer in the four years after receiving a clear mammogram. Among 4,430 Swedish women assessed, 6.9% had developed breast cancer within two years of a clear screen. The findings showed BRAIx scores were very useful for identifying women more likely to develop cancer one to two years after a clear screen, and the Australian dataset suggested BRAIx identified cancers found three to four years later, but with less accuracy. BRAIx could help identify women who might benefit from additional tests such as MRI or contrast-enhanced mammography. The findings reinforce a 2024 Swedish study that used AI-based risk assessment to select women for additional testing; researchers referred 7% of women to follow-up MRI, and 6.5% were found to have cancers missed by mammograms. Limitations of the study include difficulty comparing BRAIx to genetic testing, since BRAIx is trained to find missed or emerging cancers over a four-year period while genetic testing identifies lifetime risk, and the breast density data may not match the tool used by BreastScreen.

In the United States, UC Davis Health is co-leading a newly funded national clinical trial known as the PRISM trial (Pragmatic Randomized Trial of Artificial Intelligence for Screening Mammography), supported by a $16 million award from the Patient-Centered Outcomes Research Institute. Described as the first large-scale randomized trial in the U.S. to evaluate the effectiveness of AI in breast cancer screening interpretation, the study will involve hundreds of thousands of mammograms interpreted at academic medical centers and breast imaging facilities in California, Florida, Massachusetts, Washington, and Wisconsin. Mammograms will be randomly assigned to be interpreted either by a radiologist on their own or with assistance from an AI support tool; in all cases, a radiologist will read the mammogram and determine results. The trial aims to improve breast cancer detection and reduce unnecessary callbacks and anxiety for patients. The trial's results will inform clinical practice, coverage decisions, technology adoption, and how to communicate with patients about AI in screening. The trial brings together seven leading academic medical centers, with UCLA serving as the administrative coordinating site and UC Davis Health collecting the data and conducting the analyses.

Breast cancer remains the second leading cause of cancer death among women in the United States. Australia's BreastScreen program has offered free mammograms every two years to women aged 50 to 74 since 1992, and just over half of eligible women participate. Of the women found to have cancer, about 25% are diagnosed between the biennial screens; these interval cancers are often aggressive and more likely to be fatal. Researchers are exploring risk-adjusted screening, which tailors screening to women based on risk, as a way to detect more cancers earlier.