archiveSeptember 17, 20261 min read

New AI tool aligns surgical X-rays to 3D scans in seconds

Figure 1: Vivek Gopalakrishnan et al. Rapid patient-specific neural networks for X-ray to volume registration. Nature (2026).Full-size image

A self-supervised framework called xvr can match two-dimensional intraoperative X-rays to a patient's own three-dimensional CT or MRI scan in seconds, without manual labelling. Described in Nature on 17 September 2026, the system trains a small neural network directly from each patient's preoperative scan using physics-based simulation. According to the study, xvr achieved high accuracy across diverse anatomical structures, imaging modalities and hospitals, in what the authors call the largest evaluation of this kind of registration on real fluoroscopy to date. A foundation model pretrained on thousands of whole-body scans can be fine-tuned to any body region in roughly five minutes.

Surgeons navigating catheters and robots in real time will have more accurate, faster image guidance during procedures on any part of the body.

Source: Rapid patient-specific neural networks for X-ray to volume registration - Nature (nature.com).

Written by the HappyGood desk from the primary source linked above and checked against it. Every story links its source. Corrections: [email protected].

New AI tool aligns surgical X-rays to 3D scans in seconds

More from the archive