The vessels inside the skull resemble a mountain road's hairpin turns — sharp bends, narrowings, unexpected curves. When a traffic jam forms on such a serpentine, blood stops reaching the brain properly. To predict trouble, doctors need to see the entire flow: where it accelerates, where it swirls into eddies. The full picture comes from a special MRI, like slow-motion cinematography of fluid. But not every hospital has such a machine, and the scan is long and expensive.
A quick measurement is far simpler, like tossing a few floats into a river and recording their speed at individual points. That's exactly how ultrasound probes and CT work: they yield sparse, fragmentary numbers. The trouble is that reconstructing the entire flow in a winding channel from these crumbs is nearly hopeless. Conventional neural networks stumble here: they guess right in one spot, err by a third in another, and produce complete nonsense at a sharp bend.
The new approach resembles the work of a restorer who reconstructs a fresco from a few surviving fragments. First, they sketch the overall contours — smooth lines, large figures. This is done by the main network, shaped like a diamond lattice: it captures slow, gradual changes in flow. Then a second network steps in — broad and deep, like a set of fine brushes. It fills in the fast eddies and sharp gradients near the walls that arise precisely from the vessel's complex geometry. Both networks work in parallel and correct each other — hence the name "dual correction."
To keep the neural network from hallucinating where data are scarce, the authors added a special smoothness constraint. They made it ensure that neighboring points do not diverge too sharply — much as a restorer ensures that the color transition at the junction of brushstrokes is imperceptible. This mathematical condition is based on the Taylor expansion — a way to predict a curve's behavior near a known point. With it, the network reconstructs the entire three-dimensional dance of blood in a real vessel, copied from a patient's CT scan, from just three or four measurements. The error was several times lower than with previous methods, and the flow pattern is physically plausible — no eddy violates the laws of hydraulics.
The internal carotid artery makes nearly a full loop inside the skull: this segment is called the "carotid siphon." It is thought that this shape cushions the pulse wave from the heart, protecting the delicate vessels of the brain.