Segmentação e quantificação de tecidos da coxa e abdômen em imagens de tomografia computadorizada

• Ano de defesa: 2018
• Autor(a) principal: Felinto, Jonas de Carvalho
• 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: Tecido adiposo visceral; Tecido adiposo subcutâneo; Tecido adiposo intermuscular; Tomografia computadorizada; Segmentação automática
• Palavras-chave em Inglês: Visceral adipose tissue; Subcutaneous adipose tissue; Intermuscular adipose tissue; Computed tomography; Automatic segmentation
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/10401

The quantification of adipose and muscular tissues is of great importance in clinical practice and assists in the diagnosis of diseases such as type 2 diabetes, insulin resistance and, osteoarthritis. The use of Computed Tomography (CT) imaging has become indispensable for the quantification of such tissues and is currently considered the gold standard in clinical practice. CT imaging is able to produce results with high accuracy, with high soft tissue contrast, lower financial cost and time-wise efficiency when compared to MRI. Currently, the quantification of tissues is performed manually by a specialized radiologist with limited help of the computer and therefore has several limitations. For this reason, several researchers have turned their efforts to the development of automatic techniques of tissue segmentation. Some automatic techniques presented in the literature show high correlation values between automatic segmentation results and manual specialist markings, but the results were obtained using few images and to date none of these techniques are available to the medical community. In this context, this research aims to create a computational technique that allows automatic segmentation and quantification of thigh and abdomen tissues in CT images. This technique combines morphological operations, thresholding, a Gaussian mixture model and the use of an accumulator matrix with the projection of digital lines to classify each fat voxel in relation to its position to other tissues. Subsequently, a quantitative evaluation was done using 144 thigh images extracted from 72 thighs CT scans (left and right) and 15 CT images from the abdomen. All images were manually segmented by a specialist using image editing software and used for further comparative analysis. The precision results for the masks that delimit the thigh and fascia and that delimit the abdomenm and the visceral region are 0.99 and 0.98, respectively. Also, the coefficient of similarity of dice for these areas was close to 0.98.