Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2404.18182 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910426528219136 |
|---|---|
| author | Molina, Juan Bousse, Alexandre Catalán, Tabita Wang, Zhihan Petrache, Mircea Sahli, Francisco Prieto, Claudia Courdurier, Matìas |
| author_facet | Molina, Juan Bousse, Alexandre Catalán, Tabita Wang, Zhihan Petrache, Mircea Sahli, Francisco Prieto, Claudia Courdurier, Matìas |
| contents | Magnetic resonance imaging (MRI) is fundamental for the assessment of many diseases, due to its excellent tissue contrast characterization. This is based on quantitative techniques, such as T1 , T2 , and T2* mapping. Quantitative MRI requires the acquisition of several contrast-weighed images followed by a fitting to an exponential model or dictionary matching, which results in undesirably long acquisition times. Undersampling reconstruction techniques are commonly employed to speed up the scan, with the drawback of introducing aliasing artifacts. However, most undersampling reconstruction techniques require long computational times or do not exploit redundancies across the different contrast-weighted images. This study introduces a new regularization technique to overcome aliasing artifacts, namely CConnect, which uses an innovative regularization term that leverages several trained convolutional neural networks (CNNs) to connect and exploit information across image contrasts in a latent space. We validate our method using in-vivo T2* mapping of the brain, with retrospective undersampling factors of 4, 5 and 6, demonstrating its effectiveness in improving reconstruction in comparison to state-of-the-art techniques. Comparisons against joint total variation, nuclear low rank and a deep learning (DL) de-aliasing post-processing method, with respect to structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) metrics are presented. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_18182 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | CConnect: Synergistic Convolutional Regularization for Cartesian T2* Mapping Molina, Juan Bousse, Alexandre Catalán, Tabita Wang, Zhihan Petrache, Mircea Sahli, Francisco Prieto, Claudia Courdurier, Matìas Medical Physics Magnetic resonance imaging (MRI) is fundamental for the assessment of many diseases, due to its excellent tissue contrast characterization. This is based on quantitative techniques, such as T1 , T2 , and T2* mapping. Quantitative MRI requires the acquisition of several contrast-weighed images followed by a fitting to an exponential model or dictionary matching, which results in undesirably long acquisition times. Undersampling reconstruction techniques are commonly employed to speed up the scan, with the drawback of introducing aliasing artifacts. However, most undersampling reconstruction techniques require long computational times or do not exploit redundancies across the different contrast-weighted images. This study introduces a new regularization technique to overcome aliasing artifacts, namely CConnect, which uses an innovative regularization term that leverages several trained convolutional neural networks (CNNs) to connect and exploit information across image contrasts in a latent space. We validate our method using in-vivo T2* mapping of the brain, with retrospective undersampling factors of 4, 5 and 6, demonstrating its effectiveness in improving reconstruction in comparison to state-of-the-art techniques. Comparisons against joint total variation, nuclear low rank and a deep learning (DL) de-aliasing post-processing method, with respect to structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) metrics are presented. |
| title | CConnect: Synergistic Convolutional Regularization for Cartesian T2* Mapping |
| topic | Medical Physics |
| url | https://arxiv.org/abs/2404.18182 |