Mass Segmentation Using Shape and Size Lesion Information

Different approaches for mass segmentation have been proposed in recent years, although, as we have shown in Chapter , none of them obtain the best performances for all the studied cases. In this chapter, we develop a new algorithm for mammographic mass segmentation which takes shape and size information into account. The algorithm, which should be classified as model-based according to our survey, has been designed in two steps. Firstly, a set of real masses is used to obtain a mass prototype and its possible deformations. Secondly, a probabilistic template matching scheme is used to match the template to the masses present in a mammogram. The performance of the method, which is tested using two different databases and FROC and ROC analysis, demonstrates the validity of our approach.