AI Reads Mammograms for Heart Disease: 97,000-Scan Study Hits 86% Stroke Detection

One of the more quietly consequential AI papers of the summer dropped this week at the European Society of Cardiology’s annual congress in Munich — the world’s largest heart conference. A team of doctors in Israel has shown that a machine-learning model can read routine mammograms and reliably identify women who have suffered a stroke, have high blood pressure, or live with coronary heart disease. No new imaging. No extra radiation. Just a second, computational read of a scan that hundreds of millions of women already get every year.

The numbers behind the study are substantial. Researchers led by Dr. Viana Copeland of Tel Aviv University examined 97,364 breast scans from 29,921 women with an average age of 54. By cross-referencing the women’s medical records, the team established ground truth: 16% of the cohort had high blood pressure, 2.5% had coronary heart disease, and 2.5% had experienced a stroke. A machine-learning model was then trained to detect these conditions from mammogram images alone.

The results, presented at ESC Congress 2026, break down like this: the model identified women who had suffered a stroke with 86% reliability based solely on their mammogram. For high blood pressure and coronary heart disease, the figures were 79% and 78% respectively — the model distinguishing between women with the condition and those without. Crucially, the results held consistent regardless of the women’s age, and whether or not they also had cancer.

Why this matters: the underdiagnosis gap

Cardiovascular disease is the leading cause of death in women worldwide, and it is consistently underdiagnosed and undertreated. Part of the problem is timing: by the time a woman seeks medical help for cardiovascular symptoms, the disease is often already advanced. Part of it is a stubborn myth — as British Heart Foundation clinical director Dr. Sonya Babu-Narayan put it, the persistent belief that heart disease is “a man’s disease” means women are “disproportionally unaware, unheard, underdiagnosed, undertreated and typically underrepresented in clinical research.”

“Despite being the leading cause of death in women worldwide, CVD is consistently underdiagnosed and undertreated,” Dr. Copeland told delegates in Munich. “A common finding in our medical centre, and around the world, is that when women do seek medical help, their CVD is already advanced. On the other hand, many women do attend routine breast cancer screening, even when they haven’t sought care for cardiovascular symptoms.”

That asymmetry is the study’s core insight. Breast cancer screening programs already achieve what cardiology has struggled with: routine, midlife contact with asymptomatic women. Mammography, as Copeland noted, “reaches many women in midlife, an important period for recognising and addressing cardiovascular risk.” Attaching cardiovascular risk detection to that existing infrastructure “could potentially offer a scalable approach without requiring an additional imaging examination.”

The mechanism: breast arterial calcification

How can a breast X-ray reveal anything about the heart? The likely conduit is breast arterial calcification (BAC) — calcium deposits visible in the arteries of the breast on standard mammography. BAC has been recognized as a marker of systemic atherosclerosis for decades, and earlier work published this year in the Journal of Medical Internet Research demonstrated that AI-quantified BAC is an independent predictor of cardiovascular events. Radiologists, however, rarely report it: it sits outside the referral question, and manual grading of calcification across thousands of screens simply doesn’t scale.

This is precisely where machine learning earns its keep. A model that quantifies vascular calcification patterns automatically — and correlates them with clinical outcomes at scale — turns an incidental, ignored finding into a screening signal. It is the same pattern that has powered AI’s recent wins in radiology: the model doesn’t see things invisible to humans; it systematically extracts information humans see but don’t act on, at a volume no human workflow could process.

From experiment to clinic

Experts greeted the findings with enthusiasm tempered by the usual caveats. Elena Arbelo, an expert member of the ESC communication committee, called the results “compelling”: “A mammogram may one day do more than look for breast cancer – it may also offer a window on to cardiovascular health. That matters because CVD in women is still too often recognised late.” Her follow-up, though, is the real gating question: “The challenge now is to establish accuracy and reliability – to move from experimentation to clinical implementation.”

That path is not hypothetical. Commercial precursors already exist — Solis Mammography’s FDA-cleared Mammo+Heart service analyzes breast arterial calcium during routine mammography to produce a cardiovascular risk score, demonstrating that regulators will approve this class of tool. And the Israeli team is actively iterating: the researchers are now working to improve the model’s accuracy, reduce false positives and false negatives, and expand the set of heart conditions it can detect.

A global payoff would be large. If validated in prospective trials across diverse populations, dual-purpose mammography could convert an existing, well-attended screening channel into a cardiovascular early-warning system — flagging high-risk women years before symptoms force them into a cardiology ward. For health systems, the marginal cost approaches zero: the scans already exist; the compute to analyze them is cheap and falling.

The bigger picture

The study joins a rapidly growing body of work using AI to extract “bonus findings” from images ordered for other purposes — retinal scans predicting kidney disease, ECG algorithms flagging asymptomatic atrial fibrillation, and prior mammogram-based cardiovascular models such as the Emory-led work on AI-assessed BAC predicting heart attack, heart failure, and stroke risk. What distinguishes the ESC 2026 presentation is the combination of cohort size (nearly 30,000 women), multi-condition coverage (stroke, hypertension, CHD), and a headline accuracy (86% for stroke) high enough to force clinical attention.

It also lands at a moment when AI-in-medicine is moving from novelty to infrastructure. As one congress preview noted, a key focus of ESC 2026 is the integration of artificial intelligence across cardiology. Studies like this one are the concrete substance behind that theme — not AI replacing cardiologists, but AI repurposing the diagnostic data society already collects to close a deadly, long-standing gap in women’s health.

The caveats deserve the last word: 78–86% reliability is a screening signal, not a diagnosis, and the false-positive burden of population-wide deployment remains to be measured in prospective studies. But the direction is clear. The humble mammogram — a fixture of women’s preventive care for half a century — may be on its way to becoming one of the most powerful cardiovascular risk detectors we have.