CC Views with MLO Training

Finally, we also evaluated our algorithm using the set of $35$ CC viewed mammograms but using as a training set the RoIs extracted from MLO mammograms. Thus, the leave-one-out methodology is now not necessary as the mass of the same mammogram represented by both views is morphologically different.

The overall performance when using such approach was $A_z = 91.8\pm 3.4$ , while when using RoIs of the same view was $Az = 91.9\pm 4.7$ . Table  shows the performance of the algorithms depending on the mass size. Note that the results are very similar for the overall mean as well as for each specific size. Thus, we can conclude that the performance of the algorithm is independent on the RoIs view of the training set.

Table 4.4: Comparison of the performance when using as a training set RoIs from the same mammogaphic view (CC) or using RoIs from different views (MLO), detailed for lesion size (in $cm^2$ ). The results show mean and standard deviation of the $A_z$ values.

		Lesion Size (in $cm^2$ )
		$<$ 0.70	0.70-1.20	1.20-2.00	$>$ 2.00
-||--	Eig (CC)	$91.7\pm 5.3$	$93.2\pm 5.8$	$92.9\pm 4.0$	$91.9\pm 4.7$
	-||--	Eig (MLO)	$92.1\pm 2.5$	$93.4\pm 4.8$	$92.1\pm 2.6$
	-||--