Mammographic Image Segmentation and Risk Classification Using a Novel Texture Signature Based Methodology

Clinical research has shown that breast cancer risk is strongly related to characteristic mixture of breast tissues as seen on mammographic images. We present an automatic mammographic image segmentation approach, which uses a novel texture signature based methodology, the resultant segmentation can be found useful as a means of aiding radiologists’ estimation in mammographic risk assessment. The developed approach consists of four distinct steps: 1) feature extraction use a stack of small detail annotated mammographic patches, 2) Tabár mammographic building blocks are modelled as texture signatures, 3) model selection and reduction is used to remove noise and possible outliers, and 4) mammographic image segmentation. Visual assessment indicates good and consistent segmentation results. The MIAS database was used in a quantitative and qualitative evaluation with respect to mammographic risk assessment based on both Tabár and Birads risk categories. We found substantial agreement (κ= 0.7 and 0.75 based on Tabár and Birads risk categories, respectively) between classification results and ground truth data. Classification accuracy were 78% and 75% in Tabár and Birads categories, respectively; 86% and 87% in the corresponding low and high categories for Tabár and Birads, respectively.