Reconstrução e segmentação de angiografias via técnicas variacionais e métodos de Level Set

• Autor(a) principal: Bulant, Carlos Alberto
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do LNCC
• Texto Completo: https://tede.lncc.br/handle/tede/138
• topic: Processamento de imagem; Sistema cardiovascular; Aneurisma cerebral; Cardiovascular system; Image processing; Intracranial aneurysms; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO::PROCESSAMENTO GRAFICO (GRAPHICS)

This work focuses on tomographic reconstruction and three-dimensional arterial structures segmentation problems, understanding both as fundamentals modules of patient specific-based cardiovascular modeling. Is in this context where the possibility of improving information quality contained in medical images, by means of more accurate reconstruction methods, is of interest because of its impact on segmentation algorithm precision, allowing the construction of more realistic vessel geometrical models. Specifically, in this work, we study the reconstruction from projection inverse problem based on the classic simplified model for x-ray interaction, using the Radon transform, in two and three dimensions with different acquisition geometries. Then, we present the standard methods, which are based on analytic inversion of the Radon transform in the form of back-projection. An iterative reconstruction method based on sensitivity analysis, was studied and adapted to work in a parallel and matrix-free form. In addition, a novel reconstruction method, with foundations given by the theoretical framework of variational formulations, is developed. In view of the characteristics of the methodology, the computational implementation is carried out following a distributed computing paradigm on top of a matrix-free form version of the method. We perform parameter sensitivity analysis for each method, and a quantitative comparative analysis among them, using the well known Shepp-Logan head phantom, is also presented. Some preliminary reconstruction studies with real data are presented. In the context of medical image segmentation, we explore two alternatives for arterial structures segmentation from three dimension images acquired on three-Dimensional Rotational Angiography and Computed Tomography Angiography clinical studies. The first methodology is based on a simple seeded region growing algorithm, combined with thresholding and anisotropic diffusion techniques. The second pipeline uses a Level Set approach, initialized with the colliding fronts algorithm. Now days, Level Set methods variants are gaining popularity for vessel segmentation problems. In this work, we present a qualitative comparison by segmenting a numerical phantom, as well as segmenting arterial structures on different anatomical regions of the head and neck. At last, we present some recommendations for method and parameter choosing, as a function of the anatomical region.