A model that embedded synthetic echocardiographic motion into electrocardiogram analysis classified derived diastolic dysfunction risk phenotypes and stratified heart failure-related death risk across external cohorts, according to findings reported in an abstract published in the Journal of the American Society of Echocardiography.
The research is scheduled to be presented as a finalist in the 2026 Arthur E. Weyman Young Investigator’s Award Competition at the American Society of Echocardiography (ASE) 2026 Scientific Sessions in Aurora, Colorado.
Diastolic dysfunction reflects abnormalities in integrated left heart physiology, including ventricular relaxation, atrial function, and valvular dynamics. Although echocardiography is the reference standard for assessing diastolic dysfunction, the researchers noted that it is not feasible for widespread screening.
To address that limitation, Ankush Diwakar Jamthikar, of Rutgers Robert Wood Johnson Medical School, and colleagues developed an ensemble model that combined features from synthetic cardiac motion generated from 12-lead electrocardiograms using a generative adversarial network with embeddings from a foundation electrocardiogram model pretrained on more than 10 million recordings. The model was trained using echocardiography-derived, nonlinear, geometry-informed diastolic dysfunction risk phenotypes.
The model was developed using a multicenter electrocardiogram-echocardiography cohort of 1,012 patients and validated in an independent test cohort of 956 patients. Additional validation was performed in the EchoNext cohort of 100,000 patients from Columbia and the CODE-15% cohort of 233,770 patients from Brazil, which included longitudinal outcomes.
According to the abstract, the ensemble model classified derived diastolic dysfunction risk phenotypes with an area under the curve (AUC) of 0.86 in the development cohort and 0.85 in the external test cohort. The researchers also reported incremental predictive value over the electrocardiogram foundation model alone, with a net reclassification improvement of 0.54.
The authors reported that high-risk electrocardiogram phenotypes were associated with structural remodeling, including increased left ventricular mass index, left atrial volumes, elevated E/e′ ratio, and reduced e′ velocity. The phenotypes were also associated with clinical risk factors, including age, hypertension, and chronic kidney disease.
In the EchoNext cohort, the model identified structural heart diseases with AUCs ranging from 0.74 to 0.83 for aortic stenosis, valvular disease, and ventricular dysfunction.
In the CODE-15% cohort, the model predicted heart failure–related death. According to the abstract, patients classified as high risk had a 4-year heart failure–related death rate of 8.5%, compared with 3.0% among patients classified as low risk.
The researchers concluded that combining echo-derived risk states with synthetic cardiac motion in electrocardiogram models may improve detection of structural and functional cardiac abnormalities and may help optimize echocardiography use in high-risk, low-resource settings.
The abstract does not establish whether use of the model improves clinical outcomes, how it would perform as a prospective screening tool, or whether it could replace standard echocardiographic assessment.
Disclosures: Conflicts of interest were not specified in the abstract.
