AI Spots Hidden Heart Risk in Routine Mammograms

Mammograms are X-rays of the breast that most women over 40 get every year or two. Researchers asked a simple question: since those images also show blood vessels, could AI trained on breast cancer screening pick up on hidden heart problems too? They took "foundation models" — large AI systems trained on huge amounts of medical images, which can be reused for new jobs — and pointed them at heart risk instead.

They tested the idea on 22,497 women, tracking who had a heart attack, stroke, or similar event over five years. There were 500 such events. Using only the mammogram image, and none of the usual information like blood pressure or cholesterol, the AI ranked women's risk with a score of about 0.82. That compares with 0.77 for a model using only a woman's age. (A score of 0.5 is a coin flip; 1.0 is perfect.)

The most striking part was what the AI noticed on its own. Radiologists sometimes see calcium deposits in the breast's arteries, called breast arterial calcifications, which are linked to heart disease. The AI was never told these mattered — yet it flagged women who had them as higher risk, and its internal patterns matched blood vessel findings. In other words, it taught itself to read a heart warning sign hiding in plain sight.

There are real caveats. This is one study, not yet tested in everyday clinics, and a risk score is not a diagnosis — it could lead to false alarms and unnecessary worry or testing. But the appeal is obvious: heart disease is the leading cause of death in women, and risk checks often depend on information that's missing or outdated. If this holds up, a scan women already get could quietly double as a heart check, at no extra cost or radiation.