Figure 9.

Accuracy for background/lesion classification with respect to the number of features. The gray box highlights the accuracy obtained for $n=12$, significantly different to the model with only two features (p-value of Wilcoxon–Mann–Whitney test $<0.05$).