Figure 3

Barplots of the DICE scores (mean and standard error) computed on each lesion and on overall lesions between automatic and manual segmentations on the test sub-dataset of Dataset1 (6 patients—17 CT-scans). Upper part shows the comparison of CNN1 and CNN2, as well as an illustration of the resulting lesions. Lower part shows the comparison of CNN3 and CNN4, as well as an illustration of the resulting lesions. Significance was assessed with non-parametric two-side Wilcoxon tests. *p < 0.05.