Researchers at Mass General Brigham and the Broad Institute have developed an AI model that can predict the risk of a stroke up to ten years in advance based on a standard ECG. The model, called ECG2Stroke, uses only data from an electrocardiogram (ECG), combined with the patient’s age and gender.
According to the researchers, the technology could represent a significant step toward early stroke prevention, without the need for complex risk scores or additional tests.
Subtle patterns in ECGs
An ECG is a widely used, inexpensive, and non-invasive test that measures the heart’s electrical activity via electrodes placed on the skin. The new deep learning model analyzes subtle wave patterns in these ECGs that may indicate an increased risk of stroke. To develop ECG2Stroke, the researchers used data from more than 200,000 patients at Massachusetts General Hospital, Brigham and Women’s Hospital, and Beth Israel Deaconess Medical Center.
According to neurologist Rahul Mahajan, existing methods for determining stroke risk are often complex and difficult to scale. “Current tools for identifying which patients are at highest risk of stroke often require cumbersome calculations of clinical scores, are not easily scalable, and are therefore not widely used in daily practice.”
The results, published in the Journal of the American College of Cardiology (JACC), show that ECG2Stroke performs comparably to existing clinical risk scores but is easier to apply in daily clinical practice.
Cardioembolic strokes
The model proved particularly accurate in predicting so-called cardioembolic strokes. In these cases, blood clots form in the heart and subsequently travel to the brain. This type of stroke can often be prevented with blood thinners, provided that at-risk patients are identified in a timely manner.
Notably, abnormalities in the atria, the upper chambers of the heart, played a significant role in the AI model’s predictions. According to the researchers, this could yield new insights into the relationship between cardiac arrhythmias and strokes.
Cardiologist Shaan Khurshid emphasizes that further research is needed before the technology can be widely implemented in clinical practice. “If this is confirmed by prospective, real-world studies, such tools could help determine which patients should be prioritized for intensive preventive measures.”
According to the researchers, ECG2Stroke could ultimately contribute to more targeted prevention programs and new forms of cardiovascular research, where artificial intelligence helps identify hidden risks earlier.
AI Innovation in Stroke Care
In 2024, researchers from Imperial College London, the Technical University of Munich, and the University of Edinburgh developed AI software capable of analyzing brain scans of stroke patients with twice the accuracy.
The technology helps doctors better determine exactly when a stroke began and whether the damage is still treatable. This is crucial, as standard treatments are most effective in the first few hours after a stroke. Delayed treatment can actually lead to additional complications.
The researchers noted that patients treated within 4.5 hours are often still eligible for medical and surgical interventions. Surgical treatments are sometimes still possible up to six hours after the stroke. The AI solution can therefore contribute to faster and more accurate emergency care, especially in situations where the exact time of onset of the stroke is difficult to determine.
