Classificação automática da densidade da mama baseada em modelos gravitacionais simplificados

Abstract

Breast cancer remains the second most common cancer in the world. Currently, mammography is the main test for early diagnosis of the disease, a fundamental factor to the success of the treatment. However, the composition of the mammary tissue can make it difficult to detect lesions as the high density of the fibroglandular tissue greatly attenuates the effective energy of the X-ray beam of mammography. As a result, dense breasts present 4 to 6-fold greater risk of not detecting lesions in the initial stage. This master thesis presents a research of the simplified gravitational model applied to the analysis of mammographic images textures for later automatic density classification of the mammary tissue in computer-aided diagnostic systems. The simplified gravitational model under study was applied to 300 mammograms from the MIAS image database and, through lacunarity descriptors, combined with the Support Vector Machine (SVM) and K-nearest neighbors (Knn), obtained a performance of 76.7% of Accuracy with average Specificity of 87.27%, indicating that the method of simplified gravitational models can be used in the analysis of mammographic images texture, combined with other methods of classification and preprocessing.