Segmentação rápida e totalmente automática da gordura cardíaca em conjunto de dados de tomografia computadorizada

• Ano de defesa: 2021
• Autor(a) principal: Rodrigues, Douglas de Araújo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Segmentação de gordura cardíaca; Floor of Log; Tomografia Computadorizada
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/59386

Heart disease affects a large part of the world population. Studies show that these diseases are correlated with several chronic and acute conditions that affect the heart, such as atherosclerosis, carotid stiffness, coronary artery calcification, atrial fibrillation, and many others. All of these diseases are related to heart fat. An incorrect diagnosis made by a doctor can be the result of errors in segmenting heart fat. The process of manual segmentation involves a great deal of intra-individual and inter-individual variation in doctors. Also, this practice takes a long time, as there are many images to be analyzed by exam. In this context, this work focuses on the segmentation of cardiac fat from the Computed Tomography (CT) data set to assist the physician who is a specialist in the diagnosis. In addition to precision, the approach is dedicated to the segmentation time to enable its application in clinical routines. The proposed approach consists of using an algorithm called Floor of Log. The advantage of this method is the significant reduction in the segmentation time. This work uses Support Vector Machines to learn the best parameter of the base of the logarithm. Spatial filters and morphological processing methods are like pre-processing and post-segmentation techniques. With this approach, the average time to segment cardiac fat on a complete CT scan was just 2.01 seconds. Given the architecture review, this approach is the fastest to segment a complete exam. The precision achieved in the segmentation was 93.45%, and the specificity was 95.52%. The proposed approach is automatic and requires less computational effort. With these results, its use for cardiac fat segmentation proves to be effective and with good application times. Therefore, it can be a tool to aid medical diagnosis and, consequently, it can help specialists to obtain faster and more accurate diagnoses.