Home | Publications | DFT+26

Streak Artifact Reduction in Human-Scale Dark-Field CT Using 3D Gaussian Splatting

MCML Authors

Abstract

The clinical dark-field CT scanner (DFCT) enables attenuation- and small-angle scatter-based imaging of humanscale phantoms. Due to the scanner’s vibrating Talbot-Lau grating interferometer, images suffer from inherent streak artifacts. In this work, DFCT data is processed by re-purposing the 3D Gaussian Splatting (3DGS) method, designed for rendering sparsely sampled X-ray attenuation data, and is compared against conventional filtering. A ventilated ex vivo porcine lung placed inside an anthropomorphic phantom was measured at 20 randomly selected table positions, and reconstructed with the filtered back-projection (FBP) algorithm at the clinical DFCT setup. The re-purposed 3DGS model was compared with Gaussian, bilateral, and total variation (TV) filters for post-processing DFCT data. For each conventional filter, the hyperparameters were optimized via grid search, maximizing the structural similarity index between 3DGS and the filtered image at the global image level. The contrast-to-noise ratio (CNR) and the full-width-at-half-maximum (FWHM) of manually-selected 20×20 pixel regions of lung tissue and air were calculated. The resulting mean and standard deviation values indicate that 3DGS outperforms conventional filtering in DFCT streak reduction, preserving edge sharpness and a high CNR, a necessity for diagnostic accuracy in future clinical trials.

inproceedings DFT+26

CT Meeting 2026

9th International Conference on Image Formation in X-Ray Computed Tomography. Salt Lake City, USA, Jun 01-03, 2026. To be published. Preprint available.

Authors

T. Dorosti • D. Frey • J. B. Thalhammer • J. F. Hilmer • P. Bleuel • S. Peterhansl • J. McGinnis • D. Pfeiffer • F. Pfeiffer • D. Rückert • F. Schaff

Links

URL

Research Area

 C1 | Medicine

BibTeXKey: DFT+26

#p-rueckert #p-wiestler
Back to Top