Discussion

We have presented a new algorithm for mass detection based on the eigenfaces approach, which has been reported to be very useful for face detection and classification problems. The approach learns to detect masses using a database of RoIs only containing masses, and a probabilistic template is created representing the most probable contours (shapes) of masses. This template forms the basis of an algorithm for looking for masses in a mammogram using a probabilistic scheme. The result of this algorithm is a set of RoIs containing suspicious regions.

The performance of our approach has been evaluated using a leave-one-out methodology and FROC and ROC analysis, and two different databases. In addition a comparison with similar approaches from the state of the art has been given, obtaining slightly improved results. Although, in general, the obtained results are considered promising, the number of false positive obtained at high sensitivity levels is still significant. Moreover, one of the characteristics of the algorithm is that it performs better when dealing with smaller masses than for the larger ones. In fact, this behaviour could be expected as the algorithm is template-based and, as we have shown in Chapter , these algorithms demonstrate that behaviour. Moreover we have seen that the algorithm is independent of the RoIs view.