Comparison of the Method with Existing Approaches

We include in Table  a comparison of the presented 2DPCA false positive reduction approach incorporating breast density information with the performance of the false positive reduction schemes mentioned in Chapter , and already compared in Section . For instance, without the breast density information, only the work of Sahiner et al. [157] at ratio $1/3$ obtained better performance that the 2DPCA approach. Note that using the 2DPCA approach with specific density learning, we now obtain better performances, even using more images. We want to clarify that, however, the methods do not use the same databases and therefore our aim is only to provide a general view of the performance of our approach with respect to those strategies.

Table 6.6: Approaches dealing with mammographic mass false positive reduction, detailing the number of RoIs and the ratio (number of RoIs with masses / number of normal tissue RoIs) used. Further, we include the results obtained with the proposed approach (2DPCA) at the same ratio.

	RoIs	Ratio	$A_z$
-||-- Chang [28]	$600$	1/1	$0.83$
-||-- Tourassi [179]	$1465$	$\cong1/1$	$0.89$
-||-- 2DPCA	$1024$	1/1	$0.96$
-||--
-||-- Sahiner [157]	$672$	1/3	$0.90$
-||-- Qian01 [144]	$800$	1/3	$0.86$
-||-- 2DPCA	$1024$	1/3	$0.91$
-||--