Classificação mamográfica por densidade mamária utilizando atributos de intensidade e textura

• Ano de defesa: 2015
• Autor(a) principal: Carneiro, Pedro Cunha
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Engenharia Biomédica; Engenharias; UFU
• Palavras-chave em Português: Câncer de mama; Classificação; Densidade mamária; Extração de atributos; Redes neurais artificiais; Breast Cancer; Classification; Breast density; Features extraction; Artificial neural network; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/14101; https://doi.org/10.14393/ufu.di.2015.438

Breast cancer is a global problem, being the most frequent kind of cancer among Brazilian women. The most common tissue on the breast, that is, the breast density, is strongly associated with the risk of developing breast cancer, once dense breasts may hinder the visualization of some tumors. The best way to prevent such illness is through the mammography exam, which has a certain degree of subjectivity and depends mainly on the professional experience of the one who analyses the images. Computer-aided systems have frequently been used by radiology doctors as a tool to detect breast cancer at an early stage. The approach of this paper is that different mammography exams that show different breast density are represented by different tissues and, thus, are of different characteristics, which can be differed by their features. This way, the aim of this paper is to propose a method of classification of mammography images in density patterns from the extraction of features from the histogram and texture. The first tests were done in screen-film mammography images, and later, a digital database was used in order to verify the influence of features in different images. On the first test, with 75 screen-film mammography images, the k-means clustering technique was used and the classification was 96% accurate. When the 307 images were tested, the use of artificial neural networks was proposed and the classification of 99,26% mammography images was accurate in four classes of the pattern of breast density.