Low-Rank Angular Prior Guided Multi-Diffusion Model for Few-Shot Low-Dose CT Reconstruction

Abstract

Low-dose computed tomography (LDCT) is essential in clinical settings to minimize radiation exposure; however, reducing the dose often leads to a significant decline in image quality. Additionally, conventional deep learning approaches typically require large datasets, raising concerns about privacy, costs, and time constraints. To address these challenges, a few-shot low-dose CT reconstruction method is proposed, utilizing low-Rank Angular Prior (RAP) multi-diffusion model. In the prior learning phase, projection data is transformed into multiple consecutive views organized by angular segmentation, allowing for the extraction of rich prior information through low-rank processing. This structured approach enhances the learning capacity of the multi-diffusion model. During the iterative reconstruction phase, a stochastic differential equation solver is employed alongside data consistency constraints to iteratively refine the acquired projection data. Furthermore, penalized weighted least-squares and total variation techniques are integrated to improve image quality. Results demonstrate that the reconstructed images closely resemble those obtained from normal-dose CT, validating the RAP model as an effective and practical solution for artifact and noise reduction while preserving image fidelity in low-dose situation.