Evaluation of Mammographic Segmentation and Risk Classification Based on Tabár Tissue Modelling

This chapter evaluates recently proposed computer vision methodologies for mammographic segmentation and risk classification based on Tabár tissue modelling (Tabár et al., 2004). The overall objective is to bridge the gap between the theoretical work and their clinical usefulness, with respect to mammographic risk assessment for early breast cancer detection. In the computer vision literature, there are two lines of scientific investigation into automated mammographic risk assessment. The first focuses on the correlation between mammographic risk and the parenchymal; and the second focuses on the correlation with the overall breast density. In addition, the sensitivity of mammography is significantly reduced by increased breast density which can mask some tumours due to dense fibroglandular tissue (Sickles, 2007). Whilst computer aided mammographic density estimation has been widely investigated as a two-class (ie dense and non-dense breast tissue) problem, Tabár breast components composition estimation as a four-class (ie nodular, linear, homogeneous densities and radiolucent) problem seems to be overlooked in computer aided mammography. Despite the current standard which is favouring breast density based risk estimation, clinical evidence has shown that the mammographic parenchymal appearances (eg distribution of various breast tissues) may provide more information than percent density alone (Astley, 2004).