ρ-NeRF: Leveraging Attenuation Priors in Neural Radiance Field for 3d Computed Tomography Reconstruction

This paper introduces ρ-NeRF, a novel approach that integrates attenuation priors into neural radiance fields, setting a new benchmark in novel view synthesis (NVS) and computed tomography (CT) reconstruction. ρ-NeRF models a continuous volumetric radiance field enriched with physics-based attenuation priors, representing a three-dimensional (3D) volume through a fully connected neural network. The model takes a single four-dimensional (4D) coordinate—spatial location (x, y, z) and an initialized attenuation value (ρ)—and outputs the attenuation coefficient at that position. By querying these 4D coordinates along X-ray paths, the classic forward projection technique integrates attenuation data across the 3D space. Matching and refining pre-initialized attenuation values from traditional reconstruction algorithms, like Feldkamp-Davis-Kress algorithm (FDK) or conjugate gradient least squares (CGLS), ρ-NeRF enhances both projection synthesis and image reconstruction with minimal computational overhead. This paper details the optimization of ρ-NeRF for accurate NVS and high-quality CT reconstruction from limited projections, establishing a new standard for sparse-view CT applications.