RT info:eu-repo/semantics/doctoralThesis
T1 Enhancing brain tumor diagnosis with Synthetic MRI
A1 Moya Sáez, Elisa
A2 Universidad de Valladolid. Escuela de Doctorado
K1 Radiología
K1 Magnetic Resonance Imaging
K1 Imágen de Resonancia Magnética
K1 Synthetic MRI
K1 IRM sintética
K1 Brain Tumors
K1 Tumores Cerebrales
K1 3201.11 Radiología
AB Malignant gliomas are the most common primary brain tumors in adults. This family of brain tumors includes different types that differ in their genetic characteristics and prognostic outcomes, the latter being generally unfavorable. Survival is especially poor in high-grade gliomas such as glioblastomas, so in those cases predicting the expected survival is crucial for efficient surgery and treatment planning.As for diagnosis, in clinical practice this is commonly performed by magnetic resonance imaging (MRI) and, in particular, by visual inspection of the weighted images. The standard MRI protocol for brain tumor assessment includes (at least) four different weighted images: T1-weighted, T2-weighted, FLAIR and T1-weighted after injection of a gadolinium-based contrast agent (GBCA). The latter is used to evaluate blood brain barrier breakdown, a condition displayed on the image as signal enhancement caused by the contrast agent extravasation into the perivascular space. This diagnostic procedure has two main limitations; on the one hand, the qualitative nature of the weighted images hinders the usage of quantitative methods. On the other hand, the usage of GBCAs can trigger adverse effects that under certain circumstances can be severe, in addition to increasing the scan time and cost.In contrast, quantitative MRI is based on the computation of the tissue magnetic properties themselves, collectively known as parametric maps. These properties are the longitudinal relaxation time (T1), transverse relaxation time (T2) and proton density (PD). Parametric maps present an absolute scale and are generally considered more robust than weighted images. Recently, a new paradigm, Synthetic MRI, has gained popularity; it is based on the T1, T2, and PD parametric maps computation, followed by the synthesis of several weighted images from these maps. As a result, this procedure can enhance efficiency and diagnosis.Parametric maps computation can be performed by means of traditional relaxometry sequences. However, the long scan time of these sequences hampers their usage in clinical practice. Alternatively, fast multiparametric mapping techniques have been recently proposed. Although these techniques are faster than relaxometry sequences, they still imply a not negligible acquisition time. In addition, they usually require specific sequences or commercial products that are scarcely available on clinical scanners. Deep learning approaches could also be an alternative for the computation of parametric maps from conventional and, therefore, widely available weighted images. Thus, parametric maps could be easily computed both on pre-existing databases as well as on new acquisitions without increasing scan time. However, the lack of public datasets containing weighted images and their corresponding parametric maps could be one of the main limitations that hinders the usage of deep learning. In this Thesis we propose to enhance the diagnosis of brain tumors following a Synthetic MRI paradigm. The computation of T1, T2, and PD parametric maps have been performed with deep learning from conventional weighted images acquired with routine protocols. After that, different types of weighted images have been successfully synthesized out of the parametric maps. Several solutions have been proposed to overcome the lack of public datasets with parametric maps. In particular, training based on synthetic data and self-supervised learning strategies, which enable the training from only weighted images. In addition, both the synthesized weighted images and the computed parametric maps have been employed in different applications to improve brain tumor diagnosis. Specifically, predicting both the expected survival of glioblastoma patients and the post-contrast T1-weighted-enhanced tissues without the injection of a GBCA.
YR 2024
FD 2024
LK https://uvadoc.uva.es/handle/10324/69491
UL https://uvadoc.uva.es/handle/10324/69491
LA eng
NO Escuela de Doctorado
DS UVaDOC
RD 28-nov-2024