RA-CT NeRF: Region Adaptive CT NeRF for Arbitrary-Scale Super-Resolution

In this work, we introduce Region Adaptive-Computed Tomography Neural Radiance Fields (RA-CT NeRF), a novel framework for arbitrary-scale super-resolution of cross-sectional CT images.The proposed method addresses the challenges of anisotropic imaging by incorporating a region-adaptive sampling strategy that dynamically adjusts cube sizes based on local feature complexity.This adaptive approach enables efficient representation of both smooth and detailed regions, leading to faster convergence and higher-quality reconstructions.Furthermore, we employ an alternating training scheme to jointly optimize the color and cube size networks, enhancing stability and performance.Extensive experiments indicate that RA-CT NeRF outperforms state-of-the-art methods, achieving superior results across various scaling factors while preserving fine details and mitigating over smoothing.