An echocardiogram is like a window into the engine room: you can see the chambers contracting and relaxing. The key question is how much blood the heart pumps out with each beat. Doctors calculate this by hand, and two specialists can differ by 10–15%. A computer could be more precise, but typical neural networks are like black boxes: they give an answer but don't say how.
The new approach works differently. Instead of churning through millions of coefficients, it first identifies important details in the image on its own, like an experienced mechanic who can tell where the problem is from the sound of the engine. Then these details are assembled into a simple, understandable scheme. It turns out that this scheme predicts the ejection fraction (the proportion of blood the heart pumps out per beat) with 94.3% accuracy — better than complex 3D models. And it uses ten times fewer parameters.
Why does this matter? A cheap, transparent algorithm could run right in the ultrasound room. The doctor sees not just a number but also which parts of the heart the computer focused on. It's like a signature on an X-ray: trust grows when you can see where the conclusion came from.
A healthy human heart pumps about 60–70% of the blood out of the left ventricle with each contraction. If it drops below 40%, that's heart failure.
In the movie Minority Report, machines predicted crimes, but no one understood how they did it. Here, the computer explains which parts of the heart it looked at.