Texture-preserving implicit neural representation for Cone beam CT truncated reconstruction
Researchers have developed a new self-supervised 3D reconstruction framework for Cone Beam CT (CBCT) that addresses data truncation artifacts. The method utilizes neural scene representations to map spatial coordinates to radiodensity, bypassing traditional reconstruction operations and enabling extrapolation beyond the original field of view. To preserve high-frequency textures often lost by neural networks, a physics-based iterative refinement module is integrated, combining the artifact suppression of neural networks with the detail preservation of iterative algorithms. AI
IMPACT This new method could improve the quality and utility of medical imaging by reducing artifacts and extending the usable field of view in CT scans.