Artificial intelligence (AI)-enhanced echocardiograms successfully detected structural heart disease (SHD) and outperformed cardiologist analysis in real time with accuracy and speed in a new study published in Nature.1
Currently, the cost and accessibility of imaging tools like echocardiograms (ECG) pose a barrier to early detection of SHD and associated conditions like valvular heart disease and heart failure. SHD also constitutes $100 billion in annual direct and indirect costs in the US, which experts predict will continue to rise, thus increasing the disease burden.2 However, the recent study aimed to develop a deep learning ECG model that can accurately detect a broad range of SHDs and, in that, assess the model’s performance across institutions, patient demographics, and clinical context.1
New AI-enhanced echocardiogram model, EchoNext, outperformed cardiologists for accuracy when identifying structural heart disease. | Image credit: Image of an ECG.jpeg
“All forms of SHD can be definitively diagnosed with echocardiography, but cost, required expertise, and appropriate patient selection limit its total use,” the study authors wrote. “Thus, there remains a critical need to better risk-stratify patients and determine who should be referred for echocardiography to improve rates of SHD diagnosis and early treatment.”
The model, named EchoNext, was set to train on a database composed of 1,245,273 ECG-echocardiogram pairs from 230,318 unique patients aged 18 or older collected between December 2008 and 2022 at a New York Presbyterian–affiliated hospital. While the study did document patient covariates (age, sex, race/ethnicity), there was no clinically relevant difference in the model’s performance. Furthermore, in preparation to train the model, the data set was split at a patient level into categories for the model to train with (149,819 unique patients with 796,816 ECG–echocardiogram pairs), validate (35,780 unique patients with 35,780 ECG–echocardiogram pairs), and test (44,719 unique patients with 44,719 ECG–echocardiogram pairs).
The presence of SHD was based on clinical ECG reports and guidelines, which included low LVEF less than or equal to 45%, maximum low left ventricular wall thickness greater than or equal to 1.3 cm, moderate or severe right ventricular dysfunction, and pulmonary hypertension (pulmonary artery systolic pressure greater than or equal to 45 mm Hg, or tricuspid regurgitation jet velocity greater than or equal to 3.2 m s⁻¹). The model was also trained to make multiple predictions at once to determine which specific diseases are present or not, whether the multiple disease presence is correlated, and to learn from its predictions. The model performed the best with right ventricular (AUROC 91%) and low left ventricular systolic dysfunction (90%). The lowest performance was seen for low left ventricular wall thickness (AUROC 77%), aortic regurgitation (78%), pulmonary regurgitation (79%), and pericardial effusion (80%).
Furthermore, when compared with cardiologist non-AI–assisted reviews of ECGs, the EchoNext significantly outperformed cardiologists in a review of clinically normal and abnormal ECGs, maintaining a 77% accuracy, while cardiologists’ accuracy was 69% versus 62%, respectively. However, with AI assistance, cardiologist review accuracy improved to 69.2% (CI, 95%, 66.9–71.4%) with sensitivity 64.7% (CI, 95%, 60.9–68.3%) and specificity 72.4% (CI, 95%, 69.4%–75.3%) was still less than the EchoNext model’s accuracy with 77.3%, sensitivity 72.6%, and specificity of 80.7%.
The study was limited by race in that 89.4% of the population was White, thus limiting generalizability; however, amongst the small diverse sub-populations, there was no clinically significant difference in the model’s performance. Furthermore, the disease labels were based on ECG reports based on clinician interpretation, which may implement unknown bias.
“Together, these data demonstrate the potential for AI to help further expand the clinical and diagnostic use for an already broadly used and broadly accessible test,” the study authors wrote. “The fact that EchoNext alone performed significantly better than cardiologists, even when given the AI results, warrants further exploration.”
References
Poterucha, T.J., Jing, L., Ricart, R.P. et al. Detecting structural heart disease from electrocardiograms using AI. Nature (2025). https://doi.org/10.1038/s41586-025-09227-0
Tsao CW, Aday AW, Almarzooq ZI, Anderson CAM, et.al; American Heart Association Council on Epidemiology and Prevention Statistics Committee and Stroke Statistics Subcommittee. Heart disease and stroke statistics-2023 update: a report from the American Heart Association. Circulation. 2023 Feb 21;147(8):e93-e621. doi:10.1161/CIR.0000000000001123