Skip to main content

Table 1 Mean relative volume of segmentations produced by the different models as compared to the ground truth segmentations, both in the 70/30 and the 50/50 dataset split

From: Comparison of automated segmentation techniques for magnetic resonance images of the prostate

Model

70/30

50/50

Rel.vol.

p value

Rel.vol.

p value

Atlas

1.01

\(>0.05\)

1.00

\(>0.05\)

Siemens

1.04

\(>0.05\)

1.10

\(3\cdot 10^{-4}\)

V-net

0.99

\(>0.05\)

0.99

\(>0.05\)

TL

0.96

\(6\cdot 10^{-4}\)

0.93

\(3\cdot 10^{-7}\)

GAN

0.96

\(1\cdot 10^{-3}\)

0.98

\(>0.05\)

Eff.Det3D

0.98

\(>0.05\)

1.01

\(>0.05\)

  1. A value above one means that the model tends to overestimate the volume and vice versa. The p-values indicate how significant the (signed) volume difference is (Wilcoxon signed-rank test)