New AI Fuses X-Ray and Neutron Images to Spot Hidden Flaws

Scientists at Los Alamos National Laboratory have unveiled PiMiX 2.02, a piece of software that acts like a master translator for imaging machines. X-rays, neutron scanners and computer simulations all "see" the world differently, and each produces its own messy pile of pictures. PiMiX blends those piles into one clearer, more trustworthy image — and it does so partly on the spot, right next to the sensor, rather than shipping data off to a distant supercomputer.

The numbers are striking. When hunting for neutron events (tiny signals that reveal what's inside dense materials), the system hit better than 96% accuracy, and it could pinpoint each event to a spot smaller than a micron — roughly 100 times thinner than a human hair. That level of detail matters when you're inspecting a metal part for a crack you can't see.

The most interesting part isn't the speed, though. It's the teamwork. Rather than simply spitting out an answer, the AI generates competing explanations, ranks them using physics rules, rates its own confidence, and hands the options to a human for a final call. Researchers tested this on X-ray and neutron images from fusion energy experiments, a field where mistakes are expensive. The AI is less a replacement for a scientist than a tireless assistant who never gets bored of sorting through images.

There's a practical twist for manufacturing. The team compared the original digital blueprint of a 3D-printed metal lattice with an X-ray scan of the actual printed object, automatically flagging where reality drifted from the design. That's the kind of check that could catch a flawed part before it ends up in a car or a plane. For now, this is laboratory research, not a product you can buy, and every result still depends on human judgment. The team's stated goal is a broader "foundation model" for imaging — essentially a general-purpose AI brain for seeing inside things.