Dealing with false positive reduction in mammographic mass detection
Xavier Lladó, Arnau Oliver, Joan Martı́ · 2007
Abstract. In this paper we analyze a set of false positive reduction methods in the field of mammographic mass detection. The main goal of this false positive reduction process is the discrimination between the true recognized masses and the ones which actually are normal parenchyma. We describe three different approaches to extract breast mass image features. The first approach is based on modeling the tissue variation of both kinds of regions of interest (RoIs) by extracting the principal components (PCA) of a set of already classified RoIs. The second approach is based on an extension of the PCA approach by using the recently proposed 2DPCA algorithm. Finally, the third approach is based on Local Binary Patterns (LBP) for representing texture information and preserving at the same time the spatial structure of the masses. Once those image descriptors are extracted, the system is trained and used for classifying the unknown RoIs. We evaluate our false positive reduction approaches using a set of 1792 suspicious RoIs extracted from the DDSM database, providing a comparison when using different ratios of number of RoIs depicting masses and number of RoIs depicting normal tissue, and also when using different mass sizes, a critical aspect in mass detection systems. 1