Rapid patient-specific neural networks for X-ray to volume registration.

xvr achieves high-accuracy, patient-specific 2D/3D X-ray to volume registration within seconds, outperforming existing methods by an order of magnitude across diverse anatomies.

• Why it matters: Precise and rapid alignment of preoperative 3D images to intraoperative 2D X-rays is crucial for image-guided interventions, but current methods lack generalization and require extensive manual tuning or labeling.
• What they did: The authors developed xvr, a self-supervised framework combining physics-based simulation, a foundation model pretrained on thousands of scans, and minimal fine-tuning to adapt to individual patients in just 5 minutes.
• The result: This approach enables fast, accurate, and broadly applicable 2D/3D registration, making advanced navigation techniques more accessible in clinical and research settings through open-source software.