Proximal Neural Networks based reconstruction for few-view CT applications
This paper addresses the challenge of tomographic reconstruction from a limited number of views by using learning-based approaches. Recent advancements in Plug-and-Play (PnP) algorithms have shown promise for solving imaging inverse problems by utilizing the capabilities of Gaussian denoising algor...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | deu |
| Published: |
NDT.net
2025-02-01
|
| Series: | e-Journal of Nondestructive Testing |
| Online Access: | https://www.ndt.net/search/docs.php3?id=30770 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832086580002553856 |
|---|---|
| author | Hoang Trieu Vy Le Caroline Bossuyt Julie Escoda Marius Costin Jan De Beenhouwer Jan Sijbers |
| author_facet | Hoang Trieu Vy Le Caroline Bossuyt Julie Escoda Marius Costin Jan De Beenhouwer Jan Sijbers |
| author_sort | Hoang Trieu Vy Le |
| collection | DOAJ |
| description |
This paper addresses the challenge of tomographic reconstruction from a limited number of views by using learning-based approaches. Recent advancements in Plug-and-Play (PnP) algorithms have shown promise for solving imaging inverse problems by utilizing the capabilities of Gaussian denoising algorithms to handle complex optimization tasks. Traditional denoising hand-crafted methods produce images with predictable features but require intricate parameter tuning and suffer from slow convergence. In contrast, learning-based models offer faster performance and higher reconstruction quality, although they lack interpretability. In this work, we propose training proximal neural networks (PNNs) to eliminate arbitrary artifacts and improve the performances of PnP algorithms. These networks are obtained by unrolling a proximal algorithm designed to find a maximum a posteriori (MAP) estimate, but within a fixed number of iterations, using learned linear operators. PNNs, grounded in optimization theory, offer flexibility and can be adapted to any image restoration task solvable by a proximal algorithm. Additionally, they feature much simpler architectures compared to traditional neural networks.
|
| format | Article |
| id | doaj-art-7cf756457f554da1a013959f72f8ab37 |
| institution | Kabale University |
| issn | 1435-4934 |
| language | deu |
| publishDate | 2025-02-01 |
| publisher | NDT.net |
| record_format | Article |
| series | e-Journal of Nondestructive Testing |
| spelling | doaj-art-7cf756457f554da1a013959f72f8ab372025-02-06T10:48:19ZdeuNDT.nete-Journal of Nondestructive Testing1435-49342025-02-0130210.58286/30770Proximal Neural Networks based reconstruction for few-view CT applicationsHoang Trieu Vy LeCaroline BossuytJulie Escodahttps://orcid.org/0009-0005-4587-5294Marius CostinJan De Beenhouwerhttps://orcid.org/0000-0001-5253-1274Jan Sijbershttps://orcid.org/0000-0003-4225-2487 This paper addresses the challenge of tomographic reconstruction from a limited number of views by using learning-based approaches. Recent advancements in Plug-and-Play (PnP) algorithms have shown promise for solving imaging inverse problems by utilizing the capabilities of Gaussian denoising algorithms to handle complex optimization tasks. Traditional denoising hand-crafted methods produce images with predictable features but require intricate parameter tuning and suffer from slow convergence. In contrast, learning-based models offer faster performance and higher reconstruction quality, although they lack interpretability. In this work, we propose training proximal neural networks (PNNs) to eliminate arbitrary artifacts and improve the performances of PnP algorithms. These networks are obtained by unrolling a proximal algorithm designed to find a maximum a posteriori (MAP) estimate, but within a fixed number of iterations, using learned linear operators. PNNs, grounded in optimization theory, offer flexibility and can be adapted to any image restoration task solvable by a proximal algorithm. Additionally, they feature much simpler architectures compared to traditional neural networks. https://www.ndt.net/search/docs.php3?id=30770 |
| spellingShingle | Hoang Trieu Vy Le Caroline Bossuyt Julie Escoda Marius Costin Jan De Beenhouwer Jan Sijbers Proximal Neural Networks based reconstruction for few-view CT applications e-Journal of Nondestructive Testing |
| title | Proximal Neural Networks based reconstruction for few-view CT applications |
| title_full | Proximal Neural Networks based reconstruction for few-view CT applications |
| title_fullStr | Proximal Neural Networks based reconstruction for few-view CT applications |
| title_full_unstemmed | Proximal Neural Networks based reconstruction for few-view CT applications |
| title_short | Proximal Neural Networks based reconstruction for few-view CT applications |
| title_sort | proximal neural networks based reconstruction for few view ct applications |
| url | https://www.ndt.net/search/docs.php3?id=30770 |
| work_keys_str_mv | AT hoangtrieuvyle proximalneuralnetworksbasedreconstructionforfewviewctapplications AT carolinebossuyt proximalneuralnetworksbasedreconstructionforfewviewctapplications AT julieescoda proximalneuralnetworksbasedreconstructionforfewviewctapplications AT mariuscostin proximalneuralnetworksbasedreconstructionforfewviewctapplications AT jandebeenhouwer proximalneuralnetworksbasedreconstructionforfewviewctapplications AT jansijbers proximalneuralnetworksbasedreconstructionforfewviewctapplications |