ECG-Based AI Detects Left Ventricular Dysfunction and Heart Failure

Left ventricular dysfunction and heart failure with preserved ejection fraction are important cardiac conditions that require accurate detection to support clinical decision-making and care planning. Researchers have now developed electrocardiogram (ECG)-based artificial intelligence models to classify these conditions, including subtypes of left ventricular dysfunction. A newly published study reports the models’ performance using both 12-lead and single-lead inputs.

Researchers at Wake Forest University (Winston-Salem, NC, USA) developed two electrocardiogram (ECG)-based artificial intelligence (AI) models, one using 12-lead input and the other using single-lead input. The models were designed to classify subtypes of left ventricular dysfunction and detect heart failure with preserved ejection fraction, and were assessed using large-scale digital ECG datasets.

The team used nearly 1.08 million digital ECGs from 165,243 patients to develop and validate the models. External pediatric testing included 72,832 ECGs from 42,880 patients, with a subset of 8,418 pediatric ECGs comprising 142 cases and 8,276 controls for detailed evaluation.

In adults, the 12-lead model achieved areas under the curve of 0.90 for reduced ejection fraction (EF < 40%), 0.81 for midrange ejection fraction, and 0.80 for heart failure with preserved ejection fraction. The single-lead model reached corresponding values of 0.89, 0.78, and 0.75. In a validation cohort, the 12-lead model recorded values of 0.92, 0.76, and 0.73, while the single-lead model achieved 0.90, 0.75, and 0.74.

In the pediatric dataset, the 12-lead model attained areas under the curve of 0.97 for reduced ejection fraction, 0.71 for midrange ejection fraction, and 0.64 for heart failure with preserved ejection fraction. The single-lead model recorded corresponding values of 0.94, 0.77, and 0.67. Pediatric analyses were performed using both 12-lead and single-lead inputs.

Machine-learning models built solely from clinical data lacked generalizability, and combining clinical variables with ECG-AI outputs did not significantly improve performance over ECG-only models. The study reported both findings during model comparisons.

The study was published online on July 28, 2026, in the Journal of the American Heart Association. According to the report, the approach accurately detects left ventricular dysfunction and heart failure with preserved ejection fraction even from single-lead ECGs. The authors noted potential use in low-cost screening.

Related Links
Wake Forest University