Patient-specific neural network framework achieves rapid, accurate 2D/3D X-ray to volume alignment

A self-supervised framework, xvr, fine-tunes patient-specific neural networks in 5 min from preoperative scans to align 3D volumes with 2D fluoroscopy. In real-world testing across anatomies, modalities, and hospitals, it improved registration accuracy by an order of magnitude o…

• Uses physics-based simulation to generate training data from the patient's own CT or MRI, eliminating the need for manual labels.
• Pretrained on thousands of whole-body scans, enabling pan-anatomical application with only 5 min of patient-specific fine-tuning.
• Open-source software available, evaluated on the largest real-fluoroscopy 2D/3D registration dataset to date.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.