EchoDino improves heart ultrasound analysis from babies to adults
EchoDino: A pediatric foundation model for transferable echocardiographic analysis across the lifespan
Computer Vision and Pattern Recognition
Summary
Reading heart ultrasound videos is hard because the heart moves and looks different at different ages, especially for kids. The authors created EchoDino, a smart computer program trained on millions of unlabeled videos of kids' hearts, to understand key details without needing lots of labeled examples. EchoDino does better than older models at recognizing views, spotting heart problems, guessing age from images, and measuring heart function in both children and adults. This makes it easier to analyze heart health using ultrasound images across a person’s lifetime.
What this means in practice
- •For cardiac imaging specialists: Improve accuracy and speed of echocardiogram interpretation for pediatric and adult patients using a transferable analysis model.
- •For medical ai developers: Develop robust echocardiography diagnostic tools that require fewer labeled pediatric data by using EchoDino’s self-supervised learned representations.$Commercial implications: Enables creation of commercial AI software products for echocardiogram analysis that perform well across age groups with reduced data labeling costs.
Authors
Sheng Cheng, Donnchadh M. O'Sullivan, Daniel J. Penny, Craig G. Rusin, Minh B. Nguyen, Devika Subramanian
Abstract
Echocardiography is the most widely used cardiac imaging modality, yet interpretation demands integrating visual evidence across global anatomy, localized structures and dynamic cardiac motion. Machine-learning models have automated individual tasks, but they are typically built for a single purpose and depend on expensively labeled datasets - a barrier particularly acute in pediatric care, where data are scarce and anatomy changes with age. Here we present EchoDino, a self-supervised foundation model for echocardiography, created by adapting the DINOv3 framework to 3.7 million frames from 1.7 million unlabeled pediatric echocardiography videos. With its encoder frozen, EchoDino produces representations that capture global context, local anatomy, and dense spatial detail. We introduce Motion-biased Entropy Maximization Sampling (MEMS) to select the most informative frames for video-level analysis. Across nine pediatric and adult datasets, EchoDino outperformed strong baseline models, raising view-classification accuracy from 0.609 to 0.889 and the area under the receiver operating characteristic curve for structural-heart-disease detection from 0.811 to 0.872, while also cutting age-estimation error from 3.857 to 1.389 years, achieving the best segmentation accuracy and lowering ejection-fraction errors. By generalizing from label-free pediatric data to adult echocardiography, EchoDino offers a versatile foundation for cardiac image analysis across the lifespan.