Sinogram-based tomography densification and denoising based on a physics-informed deep learning approach
Biomedical Signal Processing and Control · 2026-10 · 126 · 110864
DOI: 10.1016/j.bspc.2026.110864 ↗Kurz gefasst
Fachartikel in der Zeitschrift Biomedical Signal Processing and Control, 2026-10. Verfasst von Carlo Pezzoli und 3 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Computed tomography (CT) is a widely used medical imaging technology that provides detailed information about the internal structures of the human body. Despite its effectiveness, CT relies on X-ray radiation, which can be harmful at high doses. To mitigate this risk, Low-Dose Computed Tomography has been developed as a safer alternative. This can be achieved either by reducing the radiation dose while maintaining a high number of projections, or by decreasing the number of projections without altering the X-ray power. Both strategies pose challenges during the reconstruction process: the former results in noisy sinograms, whereas the latter produces sparse sinograms. In this study, a novel CT reconstruction framework is proposed to address these challenges. A modular deep learning architecture, based on Generative Adversarial Networks (GANs) and Filtered Back Projection (FBP), is combined with a customized training procedure. The framework consists of a sinogram enhancement phase and a reconstruction phase. Specifically, the main novelty of the proposed approach lies in the implementation of FBP within the Physical Coherence Module (PCM), which acts as a layer of the network and serves as a bridge between the sinogram enhancement sub-network and the reconstruction sub-network. These components are trained both separately and jointly by means of a multitask loss function. Experimental results on synthetic and public datasets show improved reconstruction accuracy compared with traditional FBP and other advanced techniques for both denoising and sinogram densification tasks. The proposed framework enhances CT image detail while reducing radiation exposure.
Zitieren
APA
Pezzoli, C., Paracchini, M. B. M., Tubaro, S., & Marcon, M. (2026). Sinogram-based tomography densification and denoising based on a physics-informed deep learning approach. Biomedical Signal Processing and Control, 126, 110864. https://doi.org/10.1016/j.bspc.2026.110864
Vancouver
Pezzoli C, Paracchini MBM, Tubaro S, Marcon M. Sinogram-based tomography densification and denoising based on a physics-informed deep learning approach. Biomedical Signal Processing and Control. 2026;126:110864. doi:10.1016/j.bspc.2026.110864
BibTeX
@article{Pezzoli2026,
author = {Pezzoli, Carlo and Paracchini, Marco Brando Mario and Tubaro, Stefano and Marcon, Marco},
title = {Sinogram-based tomography densification and denoising based on a physics-informed deep learning approach},
journal = {Biomedical Signal Processing and Control},
year = {2026},
volume = {126},
pages = {110864},
doi = {10.1016/j.bspc.2026.110864}
}
EUVIMED