Published on
An AI model developed by researchers at New York-Presbyterian and Columbia University Irving Medical Center has received approval from the Food and Drug Administration for use in detecting structural heart disease through electrocardiograms (ECGs), according to a news release. Known as EchoNext, the deep learning model was trained to detect a broad range of structural abnormalities, as published in Nature:
- Left ventricular systolic dysfunction
- Ventricular hypertrophy/thickening
- Aortic stenosis
- Mitral regurgitation
- Tricuspid regurgitation
- Right ventricular dysfunction
- Pulmonary hypertension
- Other findings that warrant an echocardiogram
The model analyzes a routine 12-lead ECG and estimates whether a patient is likely to have significant structural heart disease, helping determine which patients should be recommended for echocardiography. In a controlled evaluation of 3,200 ECGs, EchoNext achieved 77.3% accuracy, and the 13 cardiologists participating in the study achieved 64.0% accuracy, according to the published results. However, this comparison was limited to ECGs alone—other information such as clinical history or physical exam that would typically be available was not accessible in the reviews.
Added diagnostic tool: For urgent care, the advantage could come from the opportunity to perform ECGs for patients with syncope, dyspnea, or palpitations, for example, who could potentially be identified for referral to cardiology or echocardiography. ECGs have traditionally been unable to detect structural heart disease, so this AI tool represents a new screening approach for clinicians.