Figure 10: Outliers plots: (A) boxplot with the results of the best three methods (ENET, FSMKL and SVM-RFE) with and without the outliers. *, Statistically significant difference with a p-value < 2.2 × 10−16 according to a pairwise Wilcoxon test; **, Statistically significant difference according to a non-parametric Friedman test with Iman and Davenport correction with a p-value < 1.92 × 10−41. At the bottom of the panel, a median-based contrast estimation heatmap. In (B) the residues are plotted by coloring each of them according to the category to which the image belongs.