New AI Reads Heart X-Rays and Predicts Blocked Arteries Without Surgery

When someone has a blocked heart artery, doctors face a tricky question: is this blockage bad enough to need a stent (a tiny mesh tube that props the artery open)? Today they usually answer it in two ways. First, an angiogram — an X-ray with dye that shows the arteries, but only as flat 2D pictures, so depth and true narrowing are hard to judge. Second, a fractional flow reserve (FFR) test, where a thin wire is threaded into the artery to measure pressure on both sides of the blockage. That's accurate, but it's invasive, takes time, and adds cost and risk for the patient.

The new study, from a team of Chinese researchers, offers a third option: let AI do the math. Their system reads two standard angiogram views and rebuilds the artery's shape in 3D. Then a second network — one that has the laws of fluid physics built directly into it — predicts how fast blood moves and what pressure it's under at each point, without any extra measurements. The physics constraint matters: it means the AI can't produce answers that are physically impossible, which has been a weakness of earlier pure machine-learning attempts.

On 32 real patients tested under four different flow conditions, the pressure-drop error was about 2%, and the predicted results matched hospital-measured FFR 93.8% of the time — 30 of 32 patients. The whole pipeline, from X-ray images to blood-flow numbers, ran in under 20 minutes per patient. The researchers also showed they could model what would happen after a stenting procedure, which could help doctors plan.

The honest catch: this is an arXiv preprint, meaning it hasn't yet been reviewed and published in a medical journal, and 32 patients is a small sample from one setting. Large, multi-hospital trials are needed before any doctor could rely on it. But the direction is clear — cheaper, faster, needle-free answers to a question heart doctors ask every day.