A research team co-led by investigators at Mass General Brigham and the Broad Institute of MIT and Harvard has developed and validated an artificial intelligence model, known as ECG2Stroke, that can predict the risk of a stroke up to 10 years into the future using a single 10-second cardiology test. The work is published in JACC .
"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," said co-lead author Rahul Mahajan, MD, PhD, a neurologist with Mass General Brigham Neuroscience Institute and the Broad Cardiovascular Disease Initiative.
To find an alternative, Mahajan and colleagues developed a deep learning model using data from patients at Massachusetts General Hospital that captured subtle waveform patterns from electrocardiograms (ECGs), which are inexpensive, non-invasive, and commonly used tests that record the heart's electrical activity through electrode sensors adhered to the skin. The team also tested the ECG2Stroke model in patients from Brigham and Women's Hospital and Beth Israel Deaconess Medical Center. In total, information on more than 200,000 patients was used to train and validate the model.
Results showed that ECG2Stroke — using only ECG data plus a patient's age and sex — can consistently predict a stroke up to 10 years in the future with performance similar to a validated clinical risk score across hospitals and patient subgroups. Features related to detecting dysfunction of the heart's atria, the upper chambers that receive blood from the body, had some of the largest influence on the model's predictions. The model was particularly accurate at predicting strokes caused by blood clots that form in the heart, break off, and travel to the brain — called cardioembolic stroke, which is preventable with blood thinners.
"If confirmed after prospective, real-world studies, tools like this could identify which patients should be prioritized for intensive prevention efforts," said co-senior author Shaan Khurshid, MD, MPH, a cardiologist with Mass General Brigham Heart and Vascular Institute and the Broad Cardiovascular Disease Initiative. "The tool could also be helpful in driving future mechanistic research into abnormalities in the upper chambers of the heart and links to stroke."
Authorship: In addition to Mahajan and Khurshid, co-authors include Danielle F. Pace (co-lead), Samuel F. Friedman, Shinwan Kany, Valentina d'Souza, Athar Roshandelpoor, Tamara N. Kimball, Savvina Prapiadou, Benjamin Y.Q. Tan, Jonathan W. Waks, Jennifer E. Ho, Patrick T. Ellinor, Mahnaz Maddah, and Christopher D. Anderson (co-senior).
Disclosures: Ellinor has received research support from Bayer, IBM Health and Bristol- Myers Squibb/Pfizer, and has served on advisory boards or consulted for Bayer, MyoKardia, and Novartis. Ho has received sponsored research support from Bayerand and consulted for Eli Lilly. Waks has served on the advisory board for HeartcoR Solutions, is a consultant for HearcoR Solutions and Heartbeam, and has received research support from Anumana. Anderson has received sponsored research support from Bayer and the Massachusetts General Hospital McCance Center for Brain Health.
Funding: This work was funded by National Institutes of Health (NIH) grants (R01HL134893, R01HL160003, R01HL140224, K24HL153669 2R01HL092577, K24HL105780, K23HL169839, R01NS103924 and U01NS069673), American Heart Association (AHA) grants (18SFRN34110082 and 23CDA1050571); Fondation Leducq 2; and AHA grants (18SFRN34250007 and 21SFRN812095.
Paper cited: Mahajan R et al. "ECG Signatures and Long-Term Ischemic Stroke Risk" JACC DOI: 10.1016/j.jacc.2026.03.084
