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

TL;DR

ρ-NeRF introduces a self-supervised, attenuation-prior–driven 4D neural radiance field for 3D CT reconstruction from sparse-view X-ray data. By modeling a continuous attenuation field F_{Θ}:(x,y,z,ρ0)→ρ and leveraging forward projection along rays with attenuation priors initialized from traditional methods, the approach improves both novel view synthesis and CT reconstruction with modest computational overhead, validated on the X3D dataset. Key contributions include redefining INR inputs to include ρ0, efficient attenuation initialization and encoding, and comprehensive ablations showing the benefits of FDK-based priors and edge-preserving interpolation. The method advances low-dose, high-fidelity 3D X-ray imaging and can be integrated with other Nerf-based CT techniques for broader medical and industrial applications.

Abstract

This paper introduces $ρ$-NeRF, a self-supervised approach that sets a new standard in novel view synthesis (NVS) and computed tomography (CT) reconstruction by modeling a continuous volumetric radiance field enriched with physics-based attenuation priors. The $ρ$-NeRF represents a three-dimensional (3D) volume through a fully-connected neural network that takes a single continuous 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 is applied to integrate attenuation data across the 3D space. By matching and refining pre-initialized attenuation values derived from traditional reconstruction algorithms like Feldkamp-Davis-Kress algorithm (FDK) or conjugate gradient least squares (CGLS), the enriched schema delivers superior fidelity in both projection synthesis and image recognition.

Figures (7)

• Figure 1: Overview of the $\rho$-NeRF framework for CT reconstruction. Discrete X-ray sample points integrate attenuation priors $\rho_0$ with spatial coordinates $(x, y, z)$. Attenuation values are updated along the X-ray paths by applying forward projection technique with encoders and the mapping function $F_{\Theta}$.
• Figure 2: Geometric configuration of CBCT and X-ray sampling.
• Figure 3: Visualization of some representative CT reconstruction results for Jaw, Foot, Chest, Aneurism, and Bonsai (from top to bottom). Zoomed region-of-interests are also provided.
• Figure 4: Visualization of some representative novel view results for Jaw. Foot, Chest, Aneurism, and Bonsai (from top to bottom).
• Figure 5: Model convergence with respect to training steps.