Including Breast Density Information in our Mass Detection Approach

The sensitivity of mass detection algorithms decreases as the density of the breast increases. In order to deal with this problem an initial pre-processing of the mammogram can be done estimating the density of the breast. This pre-processing may be performed manually or, as shown in Chapter  automatically. In both cases, this parameter is known before the detection algorithm begins.

Once the breast density is known, the question is how to introduce this information into the algorithm. Obviously, there is not an unique solution, and actually this will be algorithm-dependent. A way forward could be fine-tuning the parameters of the algorithm depending on the tissue type.

In our case, however, this information is easier to introduce. We cluster both RoIs databases (the first one containing only masses for template creation and the second one containing masses and normal tissue for false positive reduction) according to the breast density parameter. In order to avoid the need of manual intervention, this tissue classification step is automatically done using the automatic classification approach analyzed in Chapter . In Section  we compare the obtained results using this algorithm and the ones obtained using the experts manual classification. Note that this cannot be done with the MIAS database as there is not a significant difference between both classification (only one of the forty mammograms with masses was classified in different BIRADS categories using the manual annotations and the automatic algorithm).