Model-based deep learning for inverse problems in imaging
Iowa Research Online (The University of Iowa) · 25.05.2023
DOI: 10.25820/etd.007111 ↗Kurz gefasst
Fachartikel in der Zeitschrift Iowa Research Online (The University of Iowa), erschienen am 25.05.2023. Verfasst von Aniket Pramanik.
Abstract (Originalsprache)
Magnetic Resonance Imaging (MRI) systems provide incredible soft tissue contrast for the organ of interest in a non-invasive fashion. However, it is a notoriously slow imaging modality. A challenge with the acquisition of MRI data is the long scan time, especially for kids and older adults. Long acquisition times are also associated with extensive motion artifacts. A typical approach is to accelerate the data acquisition process by collecting fewer samples and interpolate the unknown ones using an image reconstruction algorithm. The state-of-the-art are model-based deep learning (MoDL) algorithms. Although MoDL approaches are fast and provide impressive performance, those require extensive memory which makes their applicability to higher dimensional problems challenging. MoDL approaches are inspired from compressed sensing based iterative optimization algorithms. However, these methods lack convergence guarantees to the global minima and are also fragile to adversarial attacks. MoDL is fragile to distribution shifts and therefore cannot be deployed in clinic since a wide variety of MRI acquisition settings are encountered. In the context of parallel MRI reconstruction, MoDL suffers from motion artifacts arising from mismatch in calibration and main scans mainly due to the explicit requirement of the MRI scanner’s coil sensitivity information.
Schlagwörter
Zitieren
APA
Pramanik, A. (2023). Model-based deep learning for inverse problems in imaging. Iowa Research Online (The University of Iowa). https://doi.org/10.25820/etd.007111
Vancouver
Pramanik A. Model-based deep learning for inverse problems in imaging. Iowa Research Online (The University of Iowa). 2023. doi:10.25820/etd.007111
BibTeX
@article{Pramanik2023,
author = {Pramanik, Aniket},
title = {Model-based deep learning for inverse problems in imaging},
journal = {Iowa Research Online (The University of Iowa)},
year = {2023},
doi = {10.25820/etd.007111}
}
EUVIMED