EchoJEPA: a foundation model for heart ultrasound videos
Researchers describe EchoJEPA, an artificial intelligence model trained on 18 million echocardiograms from 300,000 patients, in a preprint that has not yet been clinically validated.
Source: EchoJEPA: a foundation model for heart ultrasound videos (arxiv.org). Summary written with AI from the source and reviewed by rules; read the original.
What the research shows
The preprint describes EchoJEPA, a foundation model of artificial intelligence trained on 18 million echocardiograms, which are ultrasound scans of the heart, from 300,000 patients. Instead of learning from the pixels of the images, the model was guided to learn the structure of the heart. According to the researchers, it produced robust anatomical representations, evaluated on a multiview benchmark, a test that uses videos captured from different angles.
This work is still at an early stage. A preprint is a text released before peer review, so the results have not yet passed formal evaluation by other experts. The research also has not been clinically validated, meaning the model has not been tested as a diagnostic tool in patients. Models like this could one day help interpret scans, but it is too early to know whether they will be useful in medical practice.
What it means
The finding suggests AI can learn anatomical patterns of the heart from many ultrasound scans. It is not yet known whether the model is accurate across different hospitals, varied populations or routine use, or whether it improves diagnosis.
How to read it with care
This result does not replace a medical consultation. Anyone with questions about their heart health or about scans should talk to a doctor or another qualified health professional, who can assess each case.