Gradient Step Plug-and-Play Model for Dental Cone-Beam CT Reconstruction
A novel computational model seeks to minimize photon noise in the reconstruction of dental cone-beam CT images. This method employs an inverse problem framework along with a data-driven prior. To train a gradient-step denoiser, simulated fan-beam acquisitions with introduced photon noise were utilized. This trained model is incorporated into a plug-and-play gradient-step algorithm for reconstructing images. Tests conducted on synthetic data reveal effective denoising, and qualitative assessments on actual images indicate successful generalization. The findings have been published on arXiv.
Key facts
- Goal: reduce photon noise in dental cone-beam CT reconstruction
- Uses inverse problem formulation and data-based prior
- Simulated fan-beam acquisitions with added photon noise
- Prior obtained by training gradient-step denoiser on reconstructed simulated acquisitions
- Model integrated into plug-and-play gradient-step algorithm
- Experiments on synthetic data show denoising capabilities
- Qualitative evaluations on real images show generalization ability
- Published on arXiv
Entities
Institutions
- arXiv