Table 1.

Quantitative comparison of pelvic CT segmentation performance of different methods ($\text{mean}\pm \mathrm{SD}$).

Metrics	Method	Prostate	Bladder	Rectum
DSC	Hierarchical UNet-GAN	$0.91\pm 0.05$	$0.95\pm 0.06$	$0.90\pm 0.09$
	UNet-GAN	$0.86\pm 0.07$ ($p=0.031$)	$0.92\pm 0.05$ ($p=0.028$)	$0.87\pm 0.13$ ($p=0.429$)
	UNet	$0.84\pm 0.05$ ($p=0.0009$)	$0.88\pm 0.06$ ($p=0.0002$)	$0.83\pm 0.16$ ($p=0.155$)
MSD (mm)	Hierarchical UNet-GAN	$1.56\pm 0.37$	$0.95\pm 0.15$	$1.78\pm 1.13$
	UNet-GAN	$2.28\pm 0.78$	$2.11\pm 0.45$	$3.45\pm 0.95$
	UNet	$2.89\pm 1.15$	$2.34\pm 0.91$	$3.91\pm 0.47$
HD95 (mm)	Hierarchical UNet-GAN	$5.21\pm 1.17$	$4.37\pm 0.56$	$6.11\pm 1.47$
	UNet-GAN	$6.55\pm 2.87$	$5.83\pm 1.53$	$7.11\pm 3.42$
	UNet	$7.20\pm 1.90$	$7.20\pm 1.90$	$9.20\pm 1.90$
PPV	Hierarchical UNet-GAN	$0.90\pm 0.06$	$0.94\pm 0.02$	$0.87\pm 0.09$
	UNet-GAN	$0.84\pm 0.50$	$0.86\pm 0.63$	$0.83\pm 0.25$
	UNet	$0.86\pm 0.04$	$0.88\pm 0.45$	$0.85\pm 0.31$
SEN	Hierarchical UNet-GAN	$0.84\pm 0.07$	$0.97\pm 0.13$	$0.88\pm 0.16$
	UNet-GAN	$0.90\pm 0.06$	$0.92\pm 0.08$	$0.89\pm 0.01$
	UNet	$0.89\pm 0.82$	$0.93\pm 0.63$	$0.90\pm 0.96$

Note: $p$-values were computed between UNet-GAN/UNet and hierarchical UNet-GAN through Wilcoxon signed rank test. $p<0.05$ is considered statistically significant.