Estimación Semi-automática de Fracción Glandular Mamaria en Imágenes de Mastografía

Abstract

Breast cancer is a priority public health problem due to its global magnitude and importance, that develops mainly in the glandular tissue. On mammography imaging, the presence of a large amount of glandular tissue could conceal lesions. Due to this, the estimation of glandular fraction (FG) is a tool that allows evaluating the risk of developing breast cancer. Having knowledge of the different tissues that constitute the anatomy of the breast (glandular, connective and adipose tissues), on a mammography image there are structures that should not be considered for the estimation of the FG, such as skin or pectoral muscle. In the clinical practice, a proper differentiation between glandular and connective tissues is a challenging task, and a discrimination of extra-mammary structures from glandular tissue is particularly difficult due to an intensity similarity. In this work, a strategy to properly isolate the principal breast tissues from the extra-mammary structures, and to perform a robust semi-automatic segmentation of glandular, connective and adipose tissues by using the K-means algorithm in order to provide a quantitative estimation of the mammary glandular fraction is presented. Additionally, a comparison with the Density-based Spatial Clustering of Applications with Noise (DBSCAN) and an empirical glandular fraction estimated by a clinical expert, to demonstrate the convenience of the strategy is made.