Evaluation of the False Positive Approaches

In this section, the results using the PCA and 2DPCA approaches over a different set of manually selected RoIs are explained. The evaluation is done by using the MIAS [169] and the DDSM [67] databases and ROC analysis [120]. In contrast to the previous chapters where this analysis was performed counting pixel by pixel, we now proceed image by image, i.e. dealing with all the image as a RoI with mass or as a RoI with normal tissue (more details of this kind of evaluation are given in Appendix ). To avoid confusions, we will express the obtained $A_z$ values using this image per image strategy in the interval $[0-1]$ , while when using the pixel per pixel strategy the results will be given as a percentage.

In order to perform a more global evaluation of our results we propose to compute the $A_z$ value for different ratios of number of RoIs depicting masses and number of RoIs depicting normal tissue (from ratio $1/1$ to ratio $1/6$ ). The idea of analyzing these different ratios is twofold: firstly, to evaluate the performance of our method on different levels of difficulty (a ratio $1/1$ will obtain more optimistic results than $1/6$ ), and secondly, to compare our proposal with existing methods (the ones presented in Table ). It is important to notice that previous works only provide results for specific (and usually different) ratios. Hence, analyzing all these ratios will enable the comparison with them.

Figure: Performance of the system using the MIAS database.