Question for train phase

[MaoFuyou]

I want to ask about the input of training phase only x is regarded as the input? what is the colimap, mask and target？In the whole phase when they are used?

[bilalkabas]

Hi,

For MRI reconstruction:

• mask: undersampling pattern in k-space
• x_us: undersampled input image
• x_fs: fully-sampled image (ground truth, target)
• coilmap: coil sensitivity maps in multi-coil MRI

x_us is the input image to the network. mask and coilmap are used in data-consistency layers. x_fs is the reconstruction target in supervised training.

MambaRoll/main.py

Lines 80 to 93 in 763f140

	def training_step_mri(self, batch):
	""" Supervised MRI reconstruction training step """
	x_fs, x_us, mask, coilmap, _ = batch
	# Supervised training
	x_fs = torch.cat((x_fs.real, x_fs.imag), dim=1)
	x_us_cc = (torch.conj(coilmap) * x_us).sum(axis=1, keepdim=True)
	x_recon = self.net(
	x=torch.cat([x_us_cc.real, x_us_cc.imag], dim=1),
	target=x_us,
	mask=mask,
	coilmap=coilmap
	)

[MaoFuyou]

Ok, thank your for your explanation! And I want to ask if I have the BraTS2021 dataset and including (T1, T1Gd, T2, T2-FLAIR) and seg mask. How can I get it coresponding mask and coilmap?