Study tests AI analysis of mammograms for cardiovascular disease

A mammogram taken to look for breast cancer may one day be read for more than breast tissue. At ESC Congress 2026, Dr. Viana Copeland of Chaim Sheba Medical Center and Tel Aviv University presented a deep-learning model that detected patterns associated with hypertension, ischemic heart disease and stroke in women’s mammograms.

The retrospective study covered 29,921 women and 97,364 mammography examinations, with a median age of 54. The researchers matched the images with clinical information drawn from electronic medical records, medication prescriptions, procedures and imaging findings. Hypertension affected 16% of the cohort; ischemic heart disease and stroke each affected 2.5%.

The model’s performance was measured with AUROC, a score that shows how well a system separates people with a condition from those without it: 0.5 is random guessing and 1.0 is perfect discrimination. It reached 0.79 for hypertension, 0.78 for ischemic heart disease and 0.86 for stroke. The results remained consistent when the researchers considered cancer status and age.

The practical appeal is straightforward. Many women attend routine breast-cancer screening even when they have not sought help for cardiovascular symptoms, while cardiovascular disease remains underdiagnosed and undertreated in women, according to Dr. Copeland. Reading the same images for cardiovascular clues could therefore flag risk during an appointment that already exists, without an additional imaging examination.

So what changes, concretely? Not medical care today: the model is still experimental, based on retrospective data, and the researchers are working to reduce both false positives and false negatives. The next step is to establish accuracy and reliability in clinical implementation; the team also plans to test whether mammograms can reveal other cardiovascular conditions.