A 10-second heart test could soon tell you whether you're at risk of stroke a decade from now. Researchers at Mass General Brigham and the Broad Institute of MIT and Harvard have developed ECG2Stroke, an artificial intelligence model that predicts stroke risk up to 10 years into the future using a routine electrocardiogram—the same inexpensive, non-invasive test that captures your heart's electrical activity through sensors on your skin.
Stroke remains one of the leading causes of disability worldwide, and identifying which patients face the highest risk is crucial for prevention. But existing tools to spot vulnerable patients often require cumbersome clinical calculations, making them impractical for widespread use. Dr. Rahul Mahajan, a neurologist at Mass General Brigham Neuroscience Institute and co-lead of the research, saw an opportunity. "Existing tools to identify which patients are at the highest risk of stroke often require cumbersome clinical score calculations, are not easily scalable, and are therefore not used widely in routine practice," he explained.
The team trained their deep learning model on data from over 200,000 patients at Massachusetts General Hospital, Brigham and Women's Hospital, and Beth Israel Deaconess Medical Center. The model learned to detect subtle patterns in the electrical waveforms of the heart's rhythm. Remarkably, ECG2Stroke needs only three inputs: ECG data, plus the patient's age and sex. Despite this simplicity, it performs as accurately as established clinical risk scores across different hospitals and patient populations.
One particularly striking finding emerged: features related to dysfunction in the atria—the heart's upper chambers—had some of the largest influence on the model's predictions. Even more compelling, the model proved especially accurate at predicting cardioembolic strokes, the type caused when blood clots form in the heart, break loose, and travel to the brain. These strokes are preventable with blood thinners, meaning early identification could quite literally be lifesaving.
The research, published in JACC, suggests a transformative shift in how clinicians might approach stroke prevention. Instead of relying on time-consuming calculations, cardiologists could run an ECG—a test already routine in many clinical settings—and instantly generate a personalized 10-year stroke risk estimate. Dr. Shaan Khurshid, a cardiologist at Mass General Brigham Heart and Vascular Institute and co-senior author, captured the promise: "If confirmed after prospective, real-world studies, tools like this could identify which patients should be prioritized for intensive prevention efforts."
The work opens multiple doors. Beyond immediate clinical use, the model could help researchers understand why certain electrical patterns in the atria connect to future stroke risk, uncovering new pathways for prevention. The next step is real-world testing—seeing how ECG2Stroke performs when deployed in actual clinical practice, where patients and workflows are far messier than research datasets. That kind of validation will take time, but the foundation is solid. For patients and clinicians tired of guesswork, this simple 10-second test might soon offer clarity about what the next decade holds.