DeliCS Testing Data + DL Checkpoints - SPI-TGAS-MRF+GRE

This data set consists of raw MRI k-space data from 3 healthy volunteers (train_case000, test_case000, and test_case001) and 3 patients (test_case002, test_case003, and test_case004). The data were acquired on a 3T Premier MRI scanners (GE Healthcare, Waukesha, WI) with 48-channel head receiver-coils. The raw data was saved as numpy-arrays to remove any potentially identifying meta-data, and to work in the reconstruction pipeline presented in [1]. Each case tarball contains three files: <strong>raw_mrf.npy, gre_mrf.npy, noise.npy</strong> SPI-TGAS-MRF (files named <strong>raw_mrf.npy</strong>): The acquisition consists of an initial adiabatic inversion pulse followed by a 500 TR long readout train (TI/TE/TR = 20/0.7/12ms) with varying flip angles (10 to 75 degrees) and a rotating 3D center-out spiral trajectory. 48 repeats of the TR train are used for a 6 min acquisition. Details available in [2]. The data shape is: (2000, 48, 24000) = (data along spiral readout, number of receive channels, number of spirals across 500 TR's and 48 repeats) GRE (files named <strong>raw_gre.npy</strong>): A 20 second, low resolution (6.9 mm isotropic) gradient echo (GRE) pre-scan with a large FOV of 440x440x440mm^3. The data shape is: (64, 48, 4096) = (data along readout, number of receive channels, number of phase encode lines (64x64)) Noise estimation (files named <strong>noise.npy</strong>): Data from a noise scan acquired using all receive channels to calculate the noise coherence matrix. The data shape is: (48, 4096) = (number of receive channels, noise measurement points) Finally, <strong>checkpoints.tar.gz</strong> contains the pre-trained weights used for the deliCS network. [1] Iyer S, Schauman S, Sandino C, et al. Deep Learning Initialized Compressed Sensing (Deli-CS) in Volumetric Spatio-Temporal Subspace Reconstruction. <em>BioRxiv: </em>https://www.biorxiv.org/content/10.1101/2023.03.28.534431v1 [2] Cao, X, Liao, C, Iyer, SS, et al. Optimized multi-axis spiral projection MR fingerprinting with subspace reconstruction for rapid whole-brain high-isotropic-resolution quantitative imaging. <em>Magn Reson Med</em>. 2022; 88: 133- 150. doi:10.1002/mrm.29194