Simulator-Based Self-Supervision for Learned 3D Tomography Reconstruction

TL;DR

This paper introduces a self-supervised deep learning approach for 3D CT reconstruction that uses a differentiable simulator and noisy data, achieving high-quality results without reference images.

Contribution

The method trains a 3D reconstruction model solely on noisy data using a differentiable simulator, eliminating the need for reference reconstructions and improving fidelity.

Findings

Higher visual fidelity and PSNR compared to traditional methods.

Significantly faster than iterative reconstruction techniques.

Effective on both real and simulated data.

Abstract

We propose a deep learning method for 3D volumetric reconstruction in low-dose helical cone-beam computed tomography. Prior machine learning approaches require reference reconstructions computed by another algorithm for training. In contrast, we train our model in a fully self-supervised manner using only noisy 2D X-ray data. This is enabled by incorporating a fast differentiable CT simulator in the training loop. As we do not rely on reference reconstructions, the fidelity of our results is not limited by their potential shortcomings. We evaluate our method on real helical cone-beam projections and simulated phantoms. Our results show significantly higher visual fidelity and better PSNR over techniques that rely on existing reconstructions. When applied to full-dose data, our method produces high-quality results orders of magnitude faster than iterative techniques.