Patient-Specific AI Tool Aligns X-Rays With 3D Scans For Surgical Navigation

This uncertainty can prolong procedures and increase complication risks, particularly during time-critical interventions. Matching these images manually with preoperative scans is slow and operator-dependent, further limiting consistency. To improve navigation, researchers have developed a patient-specific AI technique that rapidly aligns intraoperative X-rays with preoperative three-dimensional scans.

The system, called X-ray volume registration (xvr), was developed by scientists and clinicians at the Massachusetts Institute of Technology (MIT; Cambridge, MA, USA) and collaborating institutions for surgical navigation in fields such as orthopedics and neurosurgery. The method adapts to each patient in about five minutes and then matches that patient’s X-rays to their three-dimensional scans within seconds. The process is designed to support precise, minimally invasive tool guidance.

Xvr uses a patient’s preoperative computed tomography (CT) or magnetic resonance imaging (MRI) study to generate thousands of synthetic X-rays per second via a physics-based simulation. These patient-specific images train an AI model that computes the alignment between two-dimensional fluoroscopic views and the three-dimensional anatomy with sub-millimeter precision. A pretrained foundation model enables rapid per-patient fine-tuning so the system can operate within clinical timeframes.

The developers report that xvr was pretrained using whole-body three-dimensional scans from more than 2,000 patients spanning ages, modalities, and anatomical regions. They evaluated performance on the largest available dataset of real two-dimensional/three-dimensional registrations, incorporating data from five hospitals and covering dozens of bones and organ systems in adult and pediatric patients. Xvr outperformed existing AI approaches by an order of magnitude in accuracy and robustness and ran fast enough for emergency surgeries, with potential to enhance robotic surgery systems.

Details appear in Nature on September 16, 2026. The team plans to make xvr faster for real-time deployment, conduct additional studies to verify reliability in more settings, and extend the framework to handle complex scenarios such as moving body parts.

“It takes decades of training for a clinician to become skilled enough to see grainy, 2D images and understand how everything is oriented. We want to make these 2D X-rays more informative, so it becomes safer and easier to do these life-saving procedures,” said Vivek Gopalakrishnan, a postdoc in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and lead author of the paper.

“A majority of Americans live more than an hour away from a center that can perform noninvasive procedures, like emergency stroke interventions. An hour in stroke time is incredibly substantial. Making these procedures easier by combining 2D and 3D information enables these types of highly specialized life-saving procedures to be more accessible to much broader parts of the population,” said Gopalakrishnan.

Related Links
MIT CSAIL