Construção automatizada de modelos tridimensionais da região das mamas a partir de imagens térmicas

• Autor(a) principal: COSTA, Gleidson Mendes
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFMA
• Texto Completo: https://tedebc.ufma.br/jspui/handle/tede/tede/3717
• topic: câncer de mama;; imagens termográficas;; modelos tridimensionais.; breast cancer;; thermographic images;; three-dimensional models.; Cancerologia; Processamento Gráfico; Ciência da Computação

Thermographic breast exams are physiological exams that capture the body temperature by infrared rays. They have the characteristic of detecting abnormalities in the temperature distribution over the body surface, indicating the presence of lumps before they even begin to be palpable. In addition, other advantages associated with this exam are: non- invasiveness; possibility of performing preventive exams in young women; low cost. This work aims to develop a tool for the automatic construction of three-dimensional bresat models using thermographic examinations of the breasts that have five images at different angles obtained from the Visual Lab DMR’s database: frontal view, oblique on the right lateral view, oblique on the left lateral view, right lateral view and left lateral view. The work has two key points: segmentation and geometric transformation. In segmentation, u-net, that is a variant of convolutional neural networks, is used. Five networks are created, each of them is responsible for segmenting one of the five exam images. The geometric transformation, on the other hand, transforms the dimensionality of the points of the curves of the breasts from the two-dimensional to the three-dimensional plane. These new 3D plane curves are used to build the model’s surface using Rational Non-uniform B-splines Surface (NURBS). Finally, texture mapping is performed to design the frontal thermographic examination on the model. In order to evaluate the three-dimensional models constructed and to evaluate the comparison between the proposed methodology and the literature, a research based on the Likert scale was carried out, with five points, for 15 people, among them: health specialists, 3D modelers and experienced academics to the theme. The result was evaluated as satisfactory, as there was a high approval rate for the models agreeing with the statements made. However, there are still many opportunities for improvement, evidenced during development and also pointed out during the survey.