Generative Tomography Reconstruction

TL;DR

This paper introduces a novel end-to-end differentiable architecture for tomography reconstruction that improves accuracy and efficiency, and a generative model for realistic sample generation and artifact reduction.

Contribution

It presents a new end-to-end differentiable model for direct sinogram-to-reconstruction mapping and a generative prior to enhance reconstruction quality.

Findings

More accurate reconstructions with fewer parameters and less time.

The generative model produces realistic reconstructions from noisy sinograms.

The approach reduces artifacts and errors in tomography reconstructions.

Abstract

We propose an end-to-end differentiable architecture for tomography reconstruction that directly maps a noisy sinogram into a denoised reconstruction. Compared to existing approaches our end-to-end architecture produces more accurate reconstructions while using less parameters and time. We also propose a generative model that, given a noisy sinogram, can sample realistic reconstructions. This generative model can be used as prior inside an iterative process that, by taking into consideration the physical model, can reduce artifacts and errors in the reconstructions.