EchoPrime, featured in Nature in February 2026, surpasses specific AI tools and earlier models in 23 cardiac benchmarks, with its code, weights, and demo accessible publicly.
An echocardiogram, a common cardiology diagnostic tool, is an ultrasound of the heart showing its movement, chamber activity, and structural integrity. Reading it demands training, time, and spatial attention to interpret the beating heart into a clinical narrative.
Cedars-Sinai Medical Center researchers, with Kaiser Permanente Northern California, Stanford Health Care, Beth Israel Deaconess Medical Center in Boston, and Chang Gung Memorial Hospital in Taiwan, developed an AI system to perform this task.
EchoPrime, a video-based vision-language model, assesses echocardiogram footage and writes a report on cardiac form and function. Its findings, “Comprehensive echocardiogram evaluation with view primed vision language AI,” appeared in Nature (volume 650, pages 970-977) in February 2026.
EchoPrime’s training scale is unmatched, using over 12 million echocardiography videos and cardiologists’ interpretations from 275,442 studies of 108,913 Cedars-Sinai patients.
Tested across five global health systems, EchoPrime excelled in 23 diverse cardiac benchmarks, outperforming task-specific AI, single-task models, and broader-capacity foundation models.
Designed to assist, not replace clinicians, the model offers a summary for cardiologists to review, not an autonomous diagnosis.
The team shared the model’s code, weights, and demo publicly to encourage testing against other patient populations.
EchoPrime debuts during a year when AI misdiagnosis is a top patient safety threat, as identified by ECRI. The context highlights the consistent accuracy necessary to support rather than burden cardiologists.
Cardiology suits AI diagnostics due to structured, abundant data like ultrasound video and ECGs. Cedars-Sinai’s effort is the most thorough attempt to make a generalized tool from this data. EchoPrime’s clinical deployment depends on factors like regulatory approval and institutional adoption, beyond the scope of the Nature paper.
It sets a new benchmark in cardiac AI capabilities.