Study on Optimal Generative Network for Synthesizing Brain Tumor-Segmented MR Images

Table 2

Our discriminator’s and generator’s losses for each generative model: ResNet, and U-Net.

Loss	ResNet	U-Net	Remarks
Dis	0.430	4.890e‐3	Sum of discriminator’s losses for A, B
DisA	0.206	2.358e‐3	Sum of discriminator’s losses for both real A and fake A
DisfakeA	0.282	8.767e‐4	Discriminator’s loss for fake A
DisrealA	0.131	3.840e‐3	Discriminator’s loss for real A
DisB	0.223	2.532e‐3	Sum of discriminator’s losses for both real B and fake B
DisfakeB	0.252	3.000e‐3	Discriminator’s loss for fake B
DisrealB	0.195	1.027e‐3	Discriminator’s loss for real B
Gen	1.338	2.572	Sum of generator’s losses for A ⟶ B, B ⟶ A
GenA ⟶ B	1.185	1.591	Generator’s loss for A ⟶ B
GenB ⟶ A	1.125	1.549	Generator’s loss for B ⟶ A

A is the brain tumor domain and B is the segmentation mask domain.