Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging

Magnetic Resonance Imaging (MRI) is a medical imaging technique that produces detailed images of the body organs and tissues without the use of ionizing radiation. In the field of cardiology, cardiac magnetic resonance (CMR), also known as cardiac MRI, is a valuable tool for the evaluation of patien...

Descripción completa

Detalles Bibliográficos
Autor: Martín González, Elena
Tipo de recurso: tesis doctoral
Estado:Versión publicada
Fecha de publicación:2023
País:España
Institución:Universidad de Valladolid
Repositorio:UVaDOC. Repositorio Documental de la Universidad de Valladolid
OAI Identifier:oai:uvadoc.uva.es:10324/62630
Acceso en línea:https://doi.org/10.35376/10324/62630
https://uvadoc.uva.es/handle/10324/62630
Access Level:acceso abierto
Palabra clave:Resonancia magnética
Magnetic resonance
Deep learning
Aprendizaje profundo
Reconstruction
Reconstrucción
Image registration
Registro de imagen
2211.18 Resonancia Magnética
id ES_da2db6b1b31a63a751ea5fbd07042fa3
oai_identifier_str oai:uvadoc.uva.es:10324/62630
network_acronym_str ES
network_name_str España
repository_id_str
dc.title.none.fl_str_mv Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
title Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
spellingShingle Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
Martín González, Elena
Resonancia magnética
Magnetic resonance
Resonancia magnética
Deep learning
Aprendizaje profundo
Reconstruction
Reconstrucción
Image registration
Registro de imagen
2211.18 Resonancia Magnética
title_short Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
title_full Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
title_fullStr Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
title_full_unstemmed Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
title_sort Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imaging
dc.creator.none.fl_str_mv Martín González, Elena
author Martín González, Elena
author_facet Martín González, Elena
author_role author
dc.contributor.none.fl_str_mv Alberola López, Carlos
Casaseca de la Higuera, Juan Pablo
Universidad de Valladolid. Escuela de Doctorado
dc.subject.none.fl_str_mv Resonancia magnética
Magnetic resonance
Resonancia magnética
Deep learning
Aprendizaje profundo
Reconstruction
Reconstrucción
Image registration
Registro de imagen
2211.18 Resonancia Magnética
topic Resonancia magnética
Magnetic resonance
Resonancia magnética
Deep learning
Aprendizaje profundo
Reconstruction
Reconstrucción
Image registration
Registro de imagen
2211.18 Resonancia Magnética
description Magnetic Resonance Imaging (MRI) is a medical imaging technique that produces detailed images of the body organs and tissues without the use of ionizing radiation. In the field of cardiology, cardiac magnetic resonance (CMR), also known as cardiac MRI, is a valuable tool for the evaluation of patients with different pathologies. CMR modalities such as cardiac cine and first-pass perfusion allow practitioners to visualize the myocardium dynamic behavior and to evaluate blood perfusion in the heart tissue respectively, but they require breath-holds during acquisition and have technical difficulties due to the dynamics of the heart. In addition, there are other drawbacks to CMR, including the protocol duration, high costs, and the likelihood of image degradation due to irregular cardiac and respiratory motion. There is a high demand to improve the efficiency of data collection to reduce scan times or to improve spatio-temporal resolutions. To increase patient comfort, the amount of information required for image reconstruction must be reduced, i.e. collecting only a portion of k-space (undersampling), which results in increased complexity in processing the acquisitions. The issue arises because the information in the reconstruction problem is insufficient to determine a solution, i.e. the problem is ill-conditioned, so additional information and constraints must be added. This results in a cost function that is optimized to provide a regularized solution. The focus of this Thesis is on the challenge of efficiently reconstructing dynamic MRI images from highly undersampled data, taking advantage of the motion present in the dynamic sequences to utilize the redundant information in both time and space. For this purpose, we start from the groupwise compressed sensing (GWCS) solution, previously reported by our research group. On the one hand, a parallel framework for GWCS is explored. GWCS is a highly parallelizable problem, so we will make use of the OpenCLIPER framework, another previous work of ours, for its implementation. This solution is motion compensated to increase the sparse character of the solution, so we have to parallelize: 1) the groupwise registration algorithm to estimate the motion, which will be done using FFDs with cubic B-splines, and 2) the optimization algorithm of the reconstruction itself, for which we will use NESTA. Results obtained with and without motion estimation and compensation are analyzed to conclude that the solution is clinically viable in terms of execution times, and suitable for any computing device which has an OpenCL implementation. On the other hand, we propose a GWCS-like approach that leverages deep learning to enhance the reconstruction process. Our approach eliminates the need of optimization steps and utilizes deep learning techniques instead to speed up reconstructions and reduce computational complexity. We first create a fast solution for registration with unsupervised DL, called dGW, and then a self-supervised DL solution for motion-compensated reconstruction (SSMoComp) that relies on the previously trained registration. Regarding dGW, we found that it achieved comparable accuracy to traditional optimization-based approaches, but with significantly reduced registration runtimes. As for SSMoComp, we conducted a comparative analysis with a state-of-the-art solution and observed that our design outperformed it, yielding superior results. A modified version of the cine DL solution was additionally adapted for first-pass perfusion, called SECRET. Compared with state-of-the-art approaches, the SECRET method maintains good quality reconstructions for higher acceleration rates, with low training and very fast reconstruction times.
publishDate 2023
dc.date.none.fl_str_mv 2023
dc.type.none.fl_str_mv info:eu-repo/semantics/doctoralThesis
info:eu-repo/semantics/publishedVersion
format doctoralThesis
status_str publishedVersion
dc.identifier.none.fl_str_mv https://doi.org/10.35376/10324/62630
https://uvadoc.uva.es/handle/10324/62630
url https://doi.org/10.35376/10324/62630
https://uvadoc.uva.es/handle/10324/62630
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
http://creativecommons.org/licenses/by-nc-nd/4.0/
eu_rights_str_mv openAccess
rights_invalid_str_mv http://creativecommons.org/licenses/by-nc-nd/4.0/
dc.format.none.fl_str_mv application/pdf
dc.source.none.fl_str_mv reponame:UVaDOC. Repositorio Documental de la Universidad de Valladolid
instname:Universidad de Valladolid
instname_str Universidad de Valladolid
reponame_str UVaDOC. Repositorio Documental de la Universidad de Valladolid
collection UVaDOC. Repositorio Documental de la Universidad de Valladolid
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869421528875008000
spelling Parallelization and deep learning techniques for the registration and reconstruction of dynamic cardiac magnetic resonance imagingMartín González, ElenaResonancia magnéticaMagnetic resonanceResonancia magnéticaDeep learningAprendizaje profundoReconstructionReconstrucciónImage registrationRegistro de imagen2211.18 Resonancia MagnéticaMagnetic Resonance Imaging (MRI) is a medical imaging technique that produces detailed images of the body organs and tissues without the use of ionizing radiation. In the field of cardiology, cardiac magnetic resonance (CMR), also known as cardiac MRI, is a valuable tool for the evaluation of patients with different pathologies. CMR modalities such as cardiac cine and first-pass perfusion allow practitioners to visualize the myocardium dynamic behavior and to evaluate blood perfusion in the heart tissue respectively, but they require breath-holds during acquisition and have technical difficulties due to the dynamics of the heart. In addition, there are other drawbacks to CMR, including the protocol duration, high costs, and the likelihood of image degradation due to irregular cardiac and respiratory motion. There is a high demand to improve the efficiency of data collection to reduce scan times or to improve spatio-temporal resolutions. To increase patient comfort, the amount of information required for image reconstruction must be reduced, i.e. collecting only a portion of k-space (undersampling), which results in increased complexity in processing the acquisitions. The issue arises because the information in the reconstruction problem is insufficient to determine a solution, i.e. the problem is ill-conditioned, so additional information and constraints must be added. This results in a cost function that is optimized to provide a regularized solution. The focus of this Thesis is on the challenge of efficiently reconstructing dynamic MRI images from highly undersampled data, taking advantage of the motion present in the dynamic sequences to utilize the redundant information in both time and space. For this purpose, we start from the groupwise compressed sensing (GWCS) solution, previously reported by our research group. On the one hand, a parallel framework for GWCS is explored. GWCS is a highly parallelizable problem, so we will make use of the OpenCLIPER framework, another previous work of ours, for its implementation. This solution is motion compensated to increase the sparse character of the solution, so we have to parallelize: 1) the groupwise registration algorithm to estimate the motion, which will be done using FFDs with cubic B-splines, and 2) the optimization algorithm of the reconstruction itself, for which we will use NESTA. Results obtained with and without motion estimation and compensation are analyzed to conclude that the solution is clinically viable in terms of execution times, and suitable for any computing device which has an OpenCL implementation. On the other hand, we propose a GWCS-like approach that leverages deep learning to enhance the reconstruction process. Our approach eliminates the need of optimization steps and utilizes deep learning techniques instead to speed up reconstructions and reduce computational complexity. We first create a fast solution for registration with unsupervised DL, called dGW, and then a self-supervised DL solution for motion-compensated reconstruction (SSMoComp) that relies on the previously trained registration. Regarding dGW, we found that it achieved comparable accuracy to traditional optimization-based approaches, but with significantly reduced registration runtimes. As for SSMoComp, we conducted a comparative analysis with a state-of-the-art solution and observed that our design outperformed it, yielding superior results. A modified version of the cine DL solution was additionally adapted for first-pass perfusion, called SECRET. Compared with state-of-the-art approaches, the SECRET method maintains good quality reconstructions for higher acceleration rates, with low training and very fast reconstruction times.La imagen por resonancia magnética (MRI) es una técnica de imagen médica que produce imágenes detalladas de los órganos y tejidos corporales sin utilizar radiaciones ionizantes. En el campo de la cardiología, la resonancia magnética cardiaca (CMR), también conocida como MRI cardiaca, es una valiosa herramienta para la evaluación de pacientes con distintas patologías. Las modalidades de CMR como cine cardiaco y perfusión de primer paso permiten visualizar el comportamiento dinámico del miocardio y evaluar la perfusión sanguínea en el tejido cardiaco respectivamente, pero requieren contención de la respiración durante la adquisición y presentan dificultades técnicas debido a la dinámica del corazón. Además, la CMR presenta otros inconvenientes, como la duración del protocolo, los elevados costes y la probabilidad de degradación de la imagen debido a movimientos cardíaco y respiratorio irregulares. Existe una gran demanda de mejora de la eficiencia de la adquisición de datos para reducir los tiempos de las pruebas o mejorar las resoluciones espacio-temporales. Para aumentar la comodidad del paciente, es necesario reducir la cantidad de información necesaria para la reconstrucción de las imágenes, es decir, recoger sólo una parte del espacio k (submuestreo), lo que se traduce en una mayor complejidad en el procesamiento de las adquisiciones. El problema surge porque la información del problema de reconstrucción es insuficiente para determinar una solución, es decir, el problema está mal condicionado, por lo que hay que añadir información y restricciones adicionales. Esto da lugar a una función de coste que se optimiza para proporcionar una solución regularizada. Esta Tesis se centra en el reto de reconstruir eficientemente imágenes de resonancia magnética dinámicas a partir de datos altamente submuestreados, aprovechando el movimiento presente en las secuencias dinámicas para utilizar la información redundante tanto en tiempo como en espacio. Para ello, partimos de la solución groupwise compressed sensing (GWCS), previamente desarrollada en el grupo de investigación. Por un lado, se explora un marco paralelo para GWCS. GWCS es un problema altamente paralelizable, por lo que haremos uso del framework OpenCLIPER, también desarrollado previamente en el grupo, para su implementación. Esta solución está compensada en movimiento para aumentar el carácter sparse de la solución, por lo que tenemos que paralelizar tanto el algoritmo de registro grupal para estimar el movimiento, que se hará usando FFDs con B-splines cúbicos, como el algoritmo de optimización de la propia reconstrucción, para lo que usaremos NESTA. Se analizan los resultados obtenidos con y sin estimación y compensación de movimiento, lo que nos permite afirmar que la solución es clínicamente viable en términos de tiempos de ejecución, y apta para cualquier dispositivo informático que disponga de una implementación OpenCL. Por otro lado, se propone una solución tipo GWCS basada en deep learning. Proponemos sustituir los pasos de optimización por aprendizaje profundo para que las reconstrucciones sean más rápidas y se reduzca la complejidad computacional. Primero creamos una solución rápida para el registro con DL no supervisado, llamada dGW, y luego una solución DL autosupervisada para la reconstrucción compensada en movimiento que hace uso del registro previamente entrenado, llamada SSMoComp. En cuanto a dGW, los resultados fueron comparables a los de una solución basada en la optimización, mientras que los tiempos de ejecución del registro se redujeron claramente. En cuanto a SSMoComp, se comparó con una solución del estado del arte y los resultados favorecieron nuestro diseño. Se adaptó además una versión modificada de la solución cine DL para la perfusión de primer paso, denominada SECRET. En comparación con los enfoques más avanzados, el método SECRET mantiene reconstrucciones de buena calidad para tasas de aceleración más altas, con tiempos de entrenamiento bajos y tiempos de reconstrucción reducidos.Escuela de DoctoradoDoctorado en Tecnologías de la Información y las TelecomunicacionesAlberola López, CarlosCasaseca de la Higuera, Juan PabloUniversidad de Valladolid. Escuela de Doctorado2023info:eu-repo/semantics/doctoralThesisinfo:eu-repo/semantics/publishedVersionapplication/pdfhttps://doi.org/10.35376/10324/62630https://uvadoc.uva.es/handle/10324/62630reponame:UVaDOC. Repositorio Documental de la Universidad de Valladolidinstname:Universidad de ValladolidInglésinfo:eu-repo/semantics/openAccesshttp://creativecommons.org/licenses/by-nc-nd/4.0/oai:uvadoc.uva.es:10324/626302026-06-13T12:44:47Z
score 15,301603

Similares en LA Referencia