| Campo Dublin Core | Valor | Língua |
|---|
| dc.contributor.advisor | Miosso, Cristiano Jacques | - |
| dc.contributor.author | Ziegler, Gabriel Gomes | - |
| dc.identifier.citation | ZIEGLER, Gabriel Gomes. Context-Dependent Probabilistic Prior Information Strategy for MRI Reconstruction. 2021. 66 f. Trabalho de Conclusão de Curso (Bacharelado em Engenharia de Software)—Universidade de Brasília, Brasília, 2021. | pt_BR |
| dc.description | Trabalho de Conclusão de Curso (graduação)—Universidade de Brasília, Faculdade UnB Gama, 2021. | pt_BR |
| dc.rights | Acesso Aberto | pt_BR |
| dc.subject.keyword | Compressed Sensing | pt_BR |
| dc.subject.keyword | Prior Information | pt_BR |
| dc.title | Context-Dependent Probabilistic Prior Information Strategy for MRI Reconstruction | pt_BR |
| dc.type | Trabalho de Conclusão de Curso - Graduação - Bacharelado | pt_BR |
| dc.date.accessioned | 2021-12-07T01:16:05Z | - |
| dc.date.available | 2021-12-07T01:16:05Z | - |
| dc.date.submitted | 2021-05 | - |
| dc.identifier.uri | https://bdm.unb.br/handle/10483/29340 | - |
| dc.language.iso | Inglês | pt_BR |
| dc.rights.license | A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor que autoriza a Biblioteca Digital da Produção Intelectual Discente da Universidade de Brasília (BDM) a disponibilizar o trabalho de conclusão de curso por meio do sítio bdm.unb.br, com as seguintes condições: disponível sob Licença Creative Commons 4.0 International, que permite copiar, distribuir e transmitir o trabalho, desde que seja citado o autor e licenciante. Não permite o uso para fins comerciais nem a adaptação desta. | pt_BR |
| dc.contributor.advisorco | Gusmão, Davi Benevides | - |
| dc.description.abstract1 | Obtaining images from a Magnetic Resonance Imaging (MRI) scan is a challenging
task due to the arduous process of obtaining the measurements from the machine and it
is practically impossible to collect all the signal of a subject for a given scan. To mitigate
this issue, Compressed Sensing (CS) based algorithms have been widely used in academia
to achieve high-quality images with much fewer measurements needed. CS is capable of
reconstructing MRI images at a sampling rate much lower than the Nyquist rate whilst
maintaining sufficient quality.
Since its introduction, CS has been significantly improved by the usage of
preprocessing techniques like sparsifying filters and prior information, that are focused
on improving the quality of the input data used in the CS algorithm. With that in mind,
we have improved the prior information theory by utilizing non-deterministic support
positions as well as multiple variances for the regions in the image that contain different
levels of motion. This is the intuition behind our proposed method Context-Dependent
Probabilistic Prior Information (CoDePPI) which parts from an image segmentation
based on the motion of an image to address the different levels of confidence that a
particular region in the image is part of a support position in other frames of a dynamic
MRI. This makes our method more robust by minimizing the introduced error and by
maximizing the probability to accurately use values from support regions.
Our proposed method has shown better results in MRI reconstruction when compared
to the classical prior information algorithm and non-prior information usage. Our method
was evaluated in a dynamic cardiac MRI where we had four different motion levels
regarding the movement in internal organs throughout the frames in the exam.
Additionally, this research also produced Deep Learning (DL) content intended to
be used in the improvement of CoDePPI by either utilizing Generative Adversarial
Network (GAN)s for support positions generation from an image or by automatizing
the segmentation step with a motion-detection model. A generation experiment was
done to validate the usage of GANs for signal generation for future experimentation with
MRI signal. | pt_BR |
| Aparece na Coleção: | Engenharia de Software
|
