Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

Multi-source stationary computed tomography (CT) enables rapid image reconstruction and is well suited for time-sensitive clinical and industrial applications. However, practical systems are often constrained by ultra-sparse-view sampling, which significantly degrades reconstruction quality. Conventional methods struggle in this setting because interpolation becomes inaccurate and the reconstructions contain severe artifacts. To address this challenge, this study proposes Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework tailored for multi-source stationary CT under ultra-sparse-view conditions. Diff-NAF combines a Neural Attenuation Field representation with a dual-branch conditional diffusion model. The process begins by training an initial NAF using ultra-sparse-view projections. New projections are then generated by an Angle-Prior Guided Projection Synthesis strategy that exploits inter-view priors and are refined by a Diffusion-driven Reuse Projection Refinement Module. The refined projections are incorporated as pseudo-labels into the training set for the next iteration. Through iterative refinement, Diff-NAF improves projection completeness and reconstruction fidelity, yielding high-quality CT reconstructions from ultra-sparse views. Experiments on multiple simulated 3D CT volumes and real projection data show that Diff-NAF consistently outperforms existing sparse-view reconstruction methods.