Ox nerf improves 3d x ray images from very few views
OX-NeRF: 3D X-ray Tomography Reconstruction from Sparse Views Using Implicit Neural Representation
Computer Vision and Pattern Recognition
Summary
Getting a clear 3D image from only a few X-ray pictures is very hard using usual methods. The authors created OX-NeRF, a new technique that learns patterns from many similar scenes and then fine-tunes for each specific case. It uses combined neural networks to better guess what the inside of an object looks like in 3D from very limited X-rays. Their tests showed it makes more accurate 3D reconstructions than earlier methods.
What this means in practice
- •For medical imaging engineers: Improve 3D X-ray reconstructions when very few views are available, enabling clearer images from low-dose scans.
- •For industrial nondestructive testing teams: Generate accurate 3D models of objects from a minimal number of X-ray views to detect defects without full-angle scanning.
Authors
Thomas Welsch, Min-Hsin Tu, David J. Chapman, Daniel E. Eakins
Abstract
NeRF and Gaussian splatting methods have been successfully applied on X-ray scenes where the views are too sparse for 3D reconstruction via classical methods. Ultra-sparse scenes with 10 or fewer views such as those with high-rate or low-dose acquisition still, however, present a significant challenge. To address this problem we present a new framework, Optimised X-ray Neural Radiance Fields (OX-NeRF), that combines cross-scene feature learning with scene-specific optimisation to reconstruct sets of related scenes. OX-NeRF employs a convolutional neural network (CNN) to identify cross-scene features while maintaining scene-specific multi-resolution hash grids of spatial features. The paired representations are fused and passed to a multilayer perceptron (MLP); the CNN, hash grids and MLP are then jointly optimised end-to-end. Benchmarking on parallel-beam and cone-beam X-ray datasets shows OX-NeRF provides significantly higher reconstruction accuracy on ultra-sparse scenes compared to existing radiance field methods.