CardioBench

CardioBench is a benchmark for echocardiography foundation models, accepted at IJCAI-ECAI 2026. It covers 9 tasks across 8 public datasets, including ejection fraction estimation, aortic stenosis grading, STEMI detection, and view classification, and evaluates foundation models such as EchoCLIP, EchoPrime, PanEcho, BioMedCLIP, SigLIP2, and DINOv3.

By releasing the preprocessing, splits, and evaluation pipelines, CardioBench establishes a reproducible reference point for how well these models generalize beyond individual datasets or isolated tasks, and offers actionable insights for designing future echocardiography foundation models.

Code and evaluation pipelines are available on GitHub, and the paper is on arXiv.