CMMCSegNet: Cross-Modality Multicascade Indirect LGE Segmentation on Multimodal Cardiac MR

Table 2

Indirect segmentation performance comparisons between CMMCSegNet models based on U-net and ResNet generator blocks using different training losses, where only one cascade generation block is used, “P-” means to add cosine similarity perceptual loss into training loss, and “P-” means to add perceptual loss into training loss.

Block	Loss	LV	MYO	RV
U-net	Vanilla GAN loss
	Vanilla GAN
	Vanilla GAN
	WGAN-GP loss
	WGAN-GP
	WGAN-GP
	LSGAN loss
	LSGAN
	LSGAN
ResNet	Vanilla GAN loss
	Vanilla GAN
	Vanilla GAN
	WGAN-GP loss
	WGAN-GP
	WGAN-GP
	LSGAN loss
	LSGAN
	LSGAN