AI Algorithm on Smartwatch ECG Sensors Accurately Detects Structural Heart Disease

An artificial intelligence algorithm paired with the single-lead electrocardiogram sensors on a smartwatch accurately diagnosed structural heart diseases, such as weakened pumping ability, damaged valves or thickened heart muscle, according to a preliminary study to be presented at the American Heart Association's Scientific Sessions 2025. Researchers said this is the first prospective study to show that an AI algorithm can detect multiple structural heart diseases based on measures taken from a single-lead ECG sensor on the back and digital crown of a smartwatch.

Millions of people wear smartwatches, and they are currently mainly used to detect heart rhythm problems such as atrial fibrillation. Structural heart diseases, on the other hand, are usually found with an echocardiogram, an advanced ultrasound imaging test of the heart that requires special equipment and isn't widely available for routine screening. In our study, we explored whether the same smartwatches people wear every day could also help find these hidden structural heart diseases earlier, before they progress to serious complications or cardiac events, said study author Arya Aminorroaya, M.D., M.P.H., an internal medicine resident at Yale New Haven Hospital and a research affiliate at the Cardiovascular Data Science Lab at Yale School of Medicine in New Haven, Connecticut.

Researchers developed the AI algorithm using more than 266,000 12-lead ECG recordings from more than 110,000 adults. Based on this library of data, they developed an algorithm to identify structural heart disease from a single-lead ECG that can be obtained using smartwatch sensors. For this purpose, researchers isolated only one of the 12 leads of the ECG, which resembles the single-lead ECG on smartwatches. They also accounted for random interference in ECG signaling or noise that could arise during the recording of a single-lead ECG using real-world smartwatches. The AI model was then externally validated using data from people seeking care at community hospitals, as well as data from a population-based study from Brazil.

Then, they prospectively recruited 600 participants who underwent 30-second, single-lead ECGs using a smartwatch to gauge the algorithm's accuracy in a real-world setting. The analysis found that using single-lead ECGs obtained from hospital equipment, the AI model was very effective at distinguishing people with and without structural heart disease, scoring 92% on a standard performance scale. Among the 600 participants with the single-lead ECGs obtained from a smartwatch, the AI model maintained high performance at 88% for detecting structural heart disease. The AI algorithm accurately identified most people with heart disease with 86% sensitivity and was highly accurate in ruling out heart disease with 99% negative predictive value.

On its own, a single-lead ECG is limited; it can't replace a 12-lead ECG test available in health care settings. However, with AI, it becomes powerful enough to screen for important heart conditions, said Rohan Khera, M.D., M.S., the senior author of the study, and the director of the CarDS Lab. This could make early screening for structural heart disease possible on a large scale, using devices many people already own. To get the AI model ready for interpreting signals from real-world, single-lead ECGs, researchers added some noise into the mix for model training. This little tweak helped the AI become resilient and more reliable when dealing with less-than-perfect signals, making it better at spotting structural heart disease even when the data isn't crystal clear.

During the real-world prospective study, 600 patients wore the same type of smartwatch with a single-lead ECG sensor for 30 seconds on the same day they were getting a heart ultrasound. The median age of the participants was 62 years, and about half were women, 44% were non-Hispanic white, 15% non-Hispanic Black, 7% Hispanic, 1% Asian and 33% others. About 5% were found to have structural heart disease on the heart ultrasound. Study limitations include a small number of patients with the actual disease in the prospective study and the number of false positive results. We plan to evaluate the AI tool in broader settings and explore how it could be integrated into community-based heart disease screening programs to assess its potential impact on improving preventive care, Aminorroaya said. Additional information about the study can be found in the abstract and the American Heart Association's Scientific Sessions 2025 Online Program Planner.