Segmentação automática dos hipocampos em imagens de ressonância magnética usando pontos salientes 3D

• Ano de defesa: 2018
• Autor(a) principal: Korb, Matheus Müller
• Orientador(a): Ferrari, Ricardo José
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Imagens de ressonância magnética; Pontos salientes; SIFT; Modelos geométricos deformáveis; Segmentação dos hipocampos
• Palavras-chave em Inglês: Magnetic resonance images; 3D salient points; SIFT; Geometric deformable models; Hippocampus segmentation
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/10845

The segmentation of the hippocampus in Magnetic Resonance images is an important procedure in the Alzheimer’s disease early diagnostic aid. The neuroradiologist frequently needs, in addition to the atrophy analysis, to know the volume of the hippocampus for an accurate diagnosis or even perform the monitoring of some treatment. However, the segmentation of the hippocampus performed manually by a specialist is time-consuming and subject to the inter- and intra-operator variability of the measures, for this reason, methods for automatic segmentation has been an object of study for the scientific community. Among the several proposed methods, those using anatomical atlases and deformable models present better results. These two types of techniques easily embedding the format of the models in the segmentation process, but are highly dependent on the initial positioning of the models. In this work we used 3D salient points, detected in MR images using the 3D Scale-Invariant Feature Transform (3D-SIFT), for the positioning of deformable geometric models, representative of the hippocampus. The results indicate an 11% improvement over the exclusive use of affine transformation, 30% over 3D-SIFT without any modifications and 7% over non-weighted positioning.