Table 3.

Dice score comparison of different segmentation networks trained on “Train” and “Train + TumorGAN Augmentation”. All networks were trained without usual data augmentation techniques (e.g., flip, crop, rotation, and so on). The best mean dice scores are highlighted bold.

Networks		Whole	Core	en	Mean
Cascaded Net	Without augmentation	0.848	0.748	0.643	0.746
	With TumorGAN augmentation (ours)	0.853	0.791	0.692	0.778
U-Net	Without augmentation	0.783	0.672	0.609	0.687
	With TumorGAN augmentation (ours)	0.806	0.704	0.611	0.706
Deeplab-v3	Without augmentation	0.820	0.700	0.571	0.697
	With TumorGAN augmentation (ours)	0.831	0.762	0.584	0.725