Modelagem baseada em grafo para detecção de regiões de anomalias em termografia de mama

• Ano de defesa: 2019
• Autor(a) principal: PONTES, Stefano Walker Pereira
• Orientador(a): BORCHARTT, Tiago Bonini
• Banca de defesa: BORCHARTT, Tiago Bonini , BRAZ JUNIOR, Geraldo , ARAÚJO, Marcus Costa de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM CIÊNCIA DA COMPUTAÇÃO/CCET
• Departamento: DEPARTAMENTO DE INFORMÁTICA/CCET
• País: Brasil
• Palavras-chave em Português: Termografia infravermelha; Câncer de mama; Grafo; Redes complexas; Superpixel; SLIC; Centralidade de autovetor
• Palavras-chave em Inglês: Infrared thermography; Breast cancer; Graph; Complex networks; Superpixel; SLIC; Eigenvector centrality
• Área do conhecimento CNPq: Ciência da Computação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/2795

The most common type of cancer, among women, occurs in the breasts. As the prevention is still difficult, due to several factors, early diagnosis is primordial in order to increase the survival rate of those affected. Therefore, much effort is dedicated to improving diagnostic methods by increasing of accuracy, and reduction of the imaging tests cost. In this context, thermography appears as a promising examination, with a series of advantages, but still in evolution. This work aims to provide a methodology for detection of abnormalities regions in breast thermography. In this methodology are combined superpixel segmentation, graph construction and analysis of complex network measurements. In the construction of the graph, each superpixel becomes a node of the graph and the adjacency relations between the regions becomes the edges, whose weights are determined by functions. The generated graphs are applied threshold variations to highlight the characteristics of small world networks. After several experiments involving graph measurements and community analysis, the detected regions are evaluated according to the proposed evaluation method. The evaluation of the detection performed in the methodology experiments provided results that can still be improved, in relation to the detection of the exact region of the anomalies, with a percentage of 59.1 %. However, it produced significant results, in order to indicate in which breast there is an anomaly, with a success rate of 86.4%.