Topographic representation based breast density segmentation for mammographic risk assessment

This paper presents a novel method for breast density segmentation in mammograms. The global structure of dense tissue is analysed based on a topographic map of the whole breast, which is a hierarchical representation, obtained from the upper level sets of the image. A shape tree is constructed to represent the topological and geometrical structure of the topographic map. The saliency and independency of shapes are analysed based on the shape tree to detect the candidate dense tissue regions. The geometric moments of the candidates are computed to remove incorrect dense regions. The segmentation results are evaluated based on the full MIAS database. Qualitative evaluation indicates realistic segmentation with respect to breast tissue density. For mammographic risk assessment, the obtained classification accuracy is 76% and 90% for BIRADS and low/high density classification.