Artificial intelligence–based automated quantification of breast arterial calcification on screening mammography could independently predict major adverse cardiovascular events and mortality as well as provide prognostic value incremental to the PREVENT score in a cohort of 123,762 women.
In the retrospective study, researchers analyzed screening mammograms from two US health systems and found a clear dose–response relationship between breast arterial calcification (BAC) severity and cardiovascular outcomes, including acute myocardial infarction, stroke, heart failure, and all-cause mortality.
The researchers evaluated 74,124 women from Emory Healthcare and 49,638 women from the Mayo Clinic aged 40 to 79 years who underwent screening mammography and had follow-up clinical data. A transformer-based neural network automatically segmented and quantified BAC from mammography images, generating a calcification area measurement in square millimeters.
BAC severity was categorized as zero (0 mm²), mild (> 0–10 mm²), moderate (> 10–25 mm²), or severe (> 25 mm²).
BAC was detected in 16.1% of the women in the Emory cohort and 20.6% of those in the Mayo cohort. Increasing BAC severity was associated with progressively higher rates of major adverse cardiovascular events (MACE). In the internal cohort, event rates rose from 5.96 per 1,000 person-years among the patients with no BAC to 48.89 per 1,000 person-years among those with severe BAC.
When analyzed as a continuous variable, each 1 mm² increase in BAC area was associated with about a 1% to 2% increase in risk of cardiovascular outcomes.
After adjusting for the PREVENT risk score, BAC remained an independent predictor of MACE. In the internal cohort, the risk of MACE increased with calcification severity compared with zero BAC: mild BAC = 1.32, moderate BAC = 1.75, and severe BAC = 3.29. In the external cohort, severe BAC was associated with a greater likelihood of MACE. Adding BAC to the PREVENT model improved discrimination, increasing the concordance index from 0.71 to 0.73 in the internal cohort and from 0.62 to 0.64 in the external cohort.
In a subgroup analysis, patients younger than 50 years with moderate to severe BAC had lower event-free survival compared with those who had zero BAC.
The study’s artificial intelligence segmentation model demonstrated strong performance, with 0.91 sensitivity and 0.95 specificity for BAC detection.
The researchers noted that the study was retrospective and relied on electronic health record data for outcome identification. Important variables such as menopause status and reproductive history weren't available, and PREVENT scores could only be calculated in a subset of patients as a result of missing clinical data.
“Automated BAC quantification from routine mammography may provide an opportunistic and effective cardiovascular risk assessment method in women, without additional radiation exposure,” wrote lead study author Theodorus Dapamede, of the Department of Radiology at Emory University, and colleagues.
The researchers reported no conflicts of interest. Funding was provided in part by the National Heart, Lung, and Blood Institute and the National Institutes of Health.
Source: European Heart Journal
