DeliCS Preprocessed Data

This data set consists of pre-processed MRI data as presented in Deep Learning Initialized Compressed Sensing (Deli-CS) in Volumetric Spatio-Temporal Subspace Reconstruction [1]. By downloading this dataset you will be able to re-create the figures presented in the paper using the code available on: https://github.com/SetsompopLab/deli-cs .

Each tarball named caseXXX_preprocessed.tar.gz contains data related to that subject:

deli_2min.npy is the DL genrated initial reconstruction.

init_adj_2min.npy is the inital gridding reconstructions.

ref_2min.npy is the reference LLR reconstruction (not initialized with deliCS).

ref_6min.npy is the reference LLR reconstruction using 6 min of MRF acquisition. This is considered gold standard - NOT AVAILABLE FOR TEST CASES 002-004, which are acquired in the clinic.

refine_2min_iters_20.npy is the reconstruction from the full proposed deliCS pipeline.

T1... .npy are T1 maps from various matching reconstructions

T2... .npy are T2 maps from various matching reconstructions

Additionally, the tarball named bartcompare.tar.gz contains the ref_2min.npy density compensated Sigpy reconstruction along with bartrecon_2min.cfl and bartrecon_2min.hdr, which are the non-density compensated Bart reconstructions shown in figure 3 in [1].

Furthermore, meta-data needed to process the data as presented in [1] are included. Some of the figure generation code requires the subspace basis and dictionary to perform dictionary matching on the fly. The tarball shared.tar.gz contains:

the k-space trajectory for 2 min data (traj_grp16_inacc2.mat)

the k-space trajectory for 6 min data (traj_grp48_inacc1.mat)

the density compensation function for each trajectory (dcf_2min.npy and dcf_6min.npy)

the subspace basis (phi.mat)

the dictionary (dictionary.mat)

a scaling factor for the deli reconstruction (deli_scaling_2min.npy)

[1] Iyer S, Schauman S, Sandino C, et al. Deep Learning Initialized Compressed Sensing (Deli-CS) in Volumetric Spatio-Temporal Subspace Reconstruction. BioRxiv: https://www.biorxiv.org/content/10.1101/2023.03.28.534431v1