AI finds cardiovascular disease on mammograms

Clinical takeaway: Mammography may eventually offer a routine, scalable way to catch cardiovascular disease earlier in women, who are often diagnosed late. 

Cardiovascular disease is the leading cause of death in women, yet it is consistently underdiagnosed and undertreated. Many women first get medical attention when the disease is already advanced, after years in which risk went unassessed. The gap is less a failure of tools than a matter of opportunity, since cardiovascular evaluation tends to happen only once symptoms force the issue. 

Mammography is one of the most widely performed imaging tests in women, and midlife women often complete breast screening even when they have never sought cardiovascular care. Researchers have suspected those images hold more than breast tissue, capturing latent vascular and cardiometabolic information that oncologic reads never use. Whether that signal is strong enough to matter clinically remains to be seen. A recent retrospective deep learning study suggests it may be, showing that routine mammograms carried enough information to flag women with several common cardiovascular conditions. 

Across nearly 30,000 women, the model separated those who had each of three cardiovascular conditions from those who did not, and it did so well above chance for all three. It performed best for stroke. Shown one woman who had a stroke and one who had not, the model picked out the affected woman 86% of the time. For hypertension the figure was 79%, and for ischemic heart disease 78%, against the 50% a coin flip would manage. 

Accuracy was similar whether or not a woman had breast cancer, and similar across ages. It edged higher on one standard view, the angled image taken in routine screening, reaching 88% for stroke and 80% for both hypertension and ischemic heart disease. 

The findings come from a retrospective cohort study at a single tertiary referral center, drawing on 29,921 women who underwent at least one mammogram between 2011 and 2025 and contributed 97,364 examinations in all. Median age was 54, and 18% of the women had breast cancer. Hypertension was recorded in 16% of women, ischemic heart disease and stroke in 2.5% each. The three conditions were defined from structured diagnoses in electronic medical records, supplemented by prescriptions, procedural and imaging findings, and in-hospital measurements. A convolutional neural network was trained to predict from the images alone whether each condition was present. The work was presented at the European Society of Cardiology (ESC) Congress 2026. 

The researchers are now working to improve the model's accuracy, with particular attention to reducing false positives and false negatives. They also plan to test whether mammograms can flag other cardiovascular conditions beyond the three studied. Whether the signal holds prospectively and outside a single referral center stands between a retrospective finding and a screening workflow that routes women from the mammography suite to cardiovascular assessment. 

"As both a cardiologist and a woman, I find this concept compelling: a mammogram may one day do more than look for breast cancer − it may also offer a window onto cardiovascular health. That matters because CVD in women is still too often recognised late," said Elena Arbelo, MD, PhD, cardiologist at Hospital Clínic de Barcelona and member of the ESC Communication Committee. "The challenge now is to establish accuracy and reliability − to move from experimentation to clinical implementation." 

Source: Ben-Ari L, et al. (2026 Aug 30) ESC Congress 2026. Detecting cardiovascular diseases from mammography using deep learning