New AI Framework LiftXR Improves CT Reconstruction from Sparse X-ray Views
Researchers have introduced LiftXR, a geometry-guided AI framework for reconstructing CT volumes from bi-planar X-rays. The method addresses the severely ill-posed problem of recovering 3D anatomical structures from only a few projection views, where depth information is collapsed and the spatial layout and intensity distributions are entangled. LiftXR explicitly separates the task into two stages: a layout lifter first generates a 3D anatomical layout, providing spatial guidance that makes estimating CT intensities substantially more tractable; an intensity renderer then reconstructs the volume. The approach is motivated by the observation that once spatial organization is known, intensity estimation becomes easier. The research is detailed in a paper posted on arXiv with identifier 2608.17255.
Key facts
- LiftXR is a geometry-guided framework for CT reconstruction from bi-planar X-rays.
- The framework separates layout recovery from intensity estimation.
- Reconstructing CT volumes from few X-ray views is severely ill-posed.
- The layout lifter generates a 3D anatomical layout from the X-rays.
- The intensity renderer uses the layout to reconstruct the CT volume.
- The paper is available on arXiv with ID 2608.17255.
- X-ray imaging is modeled as projection of a volumetric attenuation field.
- Once spatial layout is established, CT intensity estimation becomes more tractable.
Entities
Institutions
- arXiv