Introduction

In the conclusions of Chapter  we showed that the sensitivity of most of the mass detection algorithms decreases as the density of the breast increases. In order to take advantage of such information, we introduce a breast density classification step with the goal of increasing the overall breast mass detection. This step consists in a first classification of the database of RoIs according to the breast density parameter. Therefore, we can divide our database of images, not only according to the shape but also based on their breast density.

Note that this training step needs a huge database of mammograms with annotated masses. However, the MIAS database does not provide this set. For this reason, in this chapter we will train the system using the DDSM database, while the MIAS will only be used for evaluation. Moreover, in order to evaluate the improvement of using this information, we will repeat our experiments of mass detection for both situations: the first one in which the original database of RoIs is directly used and the second one in which the database of RoIs is previously classified according to the breast density.

The rest of this chapter is structured as follows. Section  describes in more detail this pre-classification step. Afterwards, Section  and Section  explain the obtained results for the mass detection and the false positive reduction algorithms respectively. In Section  the results obtained using a new digital database are described. This will prove the effectiveness of our method also on this kind of images. The chapter ends with a discussion.