Differentiable Few-view CT-Reconstruction for Arbitrary CT-Trajectories including Prior Knowledge

Schneider LS, Waldyra A, Sun Y, Maier A (2025)

Publication Type: Conference contribution

Publication year: 2025

Journal

e-Journal of Nondestructive Testing

Book Volume: 30

Conference Proceedings Title: 14th Conference on Industrial Computed Tomography (iCT), 4 - 7 February 2025, Antwerp, Belgium (iCT 2025)

Event location: Antwerp BE

Issue: 2

DOI: 10.58286/30724

Abstract

 Computed tomography (CT) is widely used in non-destructive testing (NDT), but the increasing flexibility of robot-based CT systems often results in more sparse and unevenly distributed projection data. This sparsity introduces significant challenges in reconstructing high-quality images. This paper presents a novel two-step pipeline for few-view CT reconstruction that combines discrete prior generation with differentiable optimization. First, the Discrete Algebraic Reconstruction Technique generates a binary volume that provides robust prior information about the object’s structure. This prior is then integrated into a fully differentiable reconstruction framework through two distinct strategies: gradient update cropping, which focuses optimization on regions identified by the prior, and prior-informed initialization, which uses the binary volume to create an informed starting point. Together, these approaches guide the iterative refinement of the reconstruction using known operator learning. Experiments on real-world datasets demonstrate the efficacy of the approach. Compared to conventional methods, the proposed framework achieves significant improvements in reconstruction quality. The results highlight the method’s ability to leverage sparse projection data, providing high-quality reconstructions even in challenging industrial scenarios.

Authors with CRIS profile

Linda-Sophie Schneider Lehrstuhl für Informatik 5 (Mustererkennung) Yipeng Sun Lehrstuhl für Informatik 14 (Bild- und Sprachverarbeitung) (LME) Andreas Maier Lehrstuhl für Informatik 5 (Mustererkennung)

Involved external institutions

Fraunhofer-Institut für Integrierte Schaltungen (IIS) DE Germany (DE)

How to cite

APA:

Schneider, L.-S., Waldyra, A., Sun, Y., & Maier, A. (2025). Differentiable Few-view CT-Reconstruction for Arbitrary CT-Trajectories including Prior Knowledge. In 14th Conference on Industrial Computed Tomography (iCT), 4 - 7 February 2025, Antwerp, Belgium (iCT 2025). Antwerp, BE.

MLA:

Schneider, Linda-Sophie, et al. "Differentiable Few-view CT-Reconstruction for Arbitrary CT-Trajectories including Prior Knowledge." Proceedings of the 14th Conference on Industrial Computed Tomography (iCT), 4 - 7 February 2025, Antwerp, Belgium (iCT 2025), Antwerp 2025.

BibTeX: Download