Skip to main content
. Author manuscript; available in PMC: 2018 Aug 5.
Published in final edited form as: IEEE Trans Med Imaging. 2018 Feb 14;37(8):1822–1834. doi: 10.1109/TMI.2018.2806309

Table III.

Median segmentation metrics from the algorithm comparison (whisker plots in Figure 3). Boldface denotes statistically significant findings where the median metric differed from DenseVNet (Confidence intervals are given in Table VI).

Spl. L. Kid. Gallb. Esoph. Liv. Stom. Panc. Duod.
Dice coefficient (%)

DEEDS+JLF 0.89 0.92 0.67 0.66 0.94 0.83 0.74 0.62
VoxResNet 0.91 0.91 0.79 0.69 0.95 0.87 0.74 0.59
VNet 0.94 0.92 0.72 0.68 0.94 0.85 0.71 0.58
DenseVNet 0.96 0.95 0.84 0.76 0.96 0.90 0.78 0.63

Mean boundary distance (mm)

DEEDS+JLF 2.1 1.3 3.0 2.3 2.1 3.6 2.3 3.9
VoxResNet 1.8 1.6 1.9 2.2 2.0 3.0 2.2 4.1
VNet 1.2 1.3 2.8 2.4 2.2 3.8 2.9 4.7
DenseVNet 0.8 0.9 1.6 1.7 1.6 2.5 1.9 4.1

95% Hausdorff distance (mm)

DEEDS+JLF 5.1 3.4 8.5 6.4 6.2 12.1 7.0 15.3
VoxResNet 3.8 4.2 5.0 6.0 5.2 9.1 6.3 13.9
VNet 3.6 3.7 7.5 6.5 6.4 12.6 9.5 16.0
DenseVNet 2.4 3.1 4.6 5.6 4.9 9.1 5.9 15.0