How to undersample knee singlecoil_train k-space?

[juliadietlmeier]

I am trying to build a simple reconstruction pipeline that is trained on fully-sampled and undersampled images. I have downloaded the NYU FastMRI knee singlecoil train data. I presume this is fully-sampled. How do I create the undersampled e.g. by factor x8 k-space? Thank you!

[mmuckley]

Hello @juliadietlmeier, the standard technique is by applying a mask, which simulates downsampling while ignoring some physics effects. We do this in quite a few areas of the repository for training models, mostly by creating a mask_func and incorporating it into the data transform pipeline.

fastMRI/fastmri_examples/varnet/train_varnet_demo.py

Lines 27 to 31 in edb5737

	mask = create_mask_for_mask_type(
	args.mask_type, args.center_fractions, args.accelerations
	)
	# use random masks for train transform, fixed masks for val transform
	train_transform = VarNetDataTransform(mask_func=mask, use_seed=False)