A Box Is Not the Bird
In a dim repair room, a crushed paper bird lantern comes out of a cardboard box. The box shows where the bird sat, but not its thin neck, curved wings, or the empty ribs inside. A rough box on a body scan works the same way: it says where an organ is, not what shape it truly has.
If the repairer trusts the box alone, the bird turns into a fat block of guesswork. Wing edges disappear. Empty spaces get filled in. Even bits of packing cloth can look like part of the bird in the same flat light. That is the trouble with box-only organ finding too, especially when nearby tissue looks almost the same.
So the first new help is shape memory. The software keeps a reference organ as a cloud of tiny dots, like a bead version of the lantern with its outer skin and inner ribs. It compares that bead shape with its current guess and pulls them closer, but only when the scan seems to show the whole organ, not just part of it.
The second new help is a better feel for what belongs together. It starts with rough yes and no guesses from the box, then builds an inner map of likeness. In the lantern room, that is like sorting paper by weave and stiffness instead of by how dark it looks under one weak bulb. This helps separate organ from background when the shades are muddy.
At first, shape memory and likeness feel like two different pushes. They are not. Together they work better than either one alone. On kidney scans, keeping both gives the strongest overlap. Remove the likeness cue and the fit drops. Remove the shape cue and it falls much more. Even the inner ribs matter.
That is the practical part. People can still give quick box marks instead of tracing every edge by hand, and the system can clip onto tools already in use. Across liver, kidney, and hippocampus scans, rough guesses move much closer to clean outlines. The box did not get smarter on its own. It finally got a memory of shape and a better eye for belonging.