In a study released on June 23, researchers from Yale and UT collaborated to create an artificial intelligence tool that can perform and analyze echocardiogram on the heart.
The PanEcho app uses ultrasounds to interpret echocardiograms, or images of the heart. Almost a million echocardiography videos were used to design and train the tool. It can correctly identify situations like severe aortic stenosis and systolic dysfunction and perform 39 echocardiographic tasks.
According to Gregory Holste, a study author and doctoral candidate in the Department of Electrical and Computer Engineering, “our teammates helped identify a total of 39 key measurements and labels that are part of a complete echocardiographic report—basically what a cardiologist would be expected to report on when they are reinterpreting an exam.” To anticipate those 39 labels, we train the model. We use this rigorous multi-site validation to assess the model’s performance across those 39 tasks after it has been trained.
According to Holste, one of PanEcho’s most remarkable features is its capacity to estimate the left ventricular ejection fraction—the percentage of blood the heart’s left ventricle pumps out—much more precisely than human experts. Furthermore, according to Holste, PanEcho is able to examine the heart holistically, but humans are only able to assess the heart from a single perspective.
According to Holste, the most distinctive feature of PanEcho is its ability to accomplish this by combining data from all accessible perspectives, not just carefully chosen ones. PanEcho makes a more comprehensive, well-informed choice regarding metrics like ejection fraction by combining data from the full examination from several cardiac perspectives.
PanEcho is open-source software that enables researchers to use and test the instrument for upcoming research projects. According to Holste, the team has already gotten emails from users attempting to modify the application for various purposes.
It is not something that PanEcho was designed to accomplish right out of the box, and we are aware that other researchers are trying to modify it to function on pediatric scans, Holste stated. However, it can swiftly adjust and fine-tune to that domain because it has seen so much data. Future research opportunities are quite interesting.
