Introduction

A segmentation algorithm, in a mammographic context, is an algorithm used to detect something, usually the whole breast or a specific kind of abnormalities, like micro-calcifications or masses. It is generally accepted that the detection of masses is technically more difficult than the detection of micro-calcifications, because masses can be simulated or obscured by normal breast parenchyma [5,163]. Moreover, there is a large variability in these lesions, which is reflected in the morphology variation (shape and size of the lesions), and also in the large number of features that have been used to detect and classify them.

The objective of this chapter is to review and classify the different mass detection algorithms found in the literature. The initial classification of the algorithms is done according to the number of images that the algorithms deal with. Thus, in Section  the algorithms which look for masses using a single mammogram are described and classified from a computer vision viewpoint, reflecting the strategy and the features used. It is important to notice that most of the algorithms are only able to detect a specific kind of mass, usually circular or spiculated masses. On the other hand, Section  briefly describes the algorithms which used more than one image to detect masses. They are classified according to the kind of images they used: comparison of both CC and MLO views of the same breast, comparison of the left and right breasts of the same woman, or also the temporal comparison of the same mammogram.

In Section  two key methods have been selected from each strategy and re-implemented in order to evaluate and compare their performance over the same set of MLO mammographic images, which can be clearly divided into two subsets: one containing masses and the other one with normal cases. The resulting conclusions are explained in Section .

Before starting the review of the mass segmentation strategies, however, the next section briefly describes different proposals for segmenting the breast profile. This is a necessary step in order to correctly focus the mass detection algorithms, otherwise, the annotations and background noise will introduce a huge number of outliers.