AI Improves X-Ray Vision for Safer Surgery

A surgeon steering a catheter through a blood vessel during a stroke intervention relies on live X-rays to see where the tool is going. The problem is that an X-ray is flat, and a blood vessel or a lesion isn't. Matching that 2D image to the 3D map a CT or MRI scan already provided means clinicians often have to align the two manually, a slow process that eats into procedures where minutes matter. Existing AI tools built to automate that alignment have a habit of failing outside the narrow set of cases they were trained on, which has kept them out of real operating rooms.

Researchers at MIT and several partner institutions built a new system called xvr, short for X-ray volume registration, to close that gap. The model adapts to a new patient in about five minutes, then matches that patient's X-rays to their preoperative 3D scan in seconds, with precision measured in fractions of a millimeter. Tested across different body parts and procedures, the team says it beat prior AI methods by a wide margin. The stakes behind that speed and accuracy became clear when one of the researchers explained why access to these procedures is so uneven across the country.