Accurate object localization facilitates automatic esophagus segmentation in deep learning

Radiation Oncology

Table 1 Summary of evaluate metrics and p-values for all models

models	Dice	95%HD	volume ratio
model1: 3D U-net	0.771	6.55	2.18
model2: 2D U-net_coarse	0.807	3.57	1.42
model3: 2D U-net_fine	0.817	3.76	1.43
p (model1 vs. model2)	< 0.001	< 0.001	0.002
p (model2 vs. model3)	0.010	0.433	0.752

ISSN: 1748-717X