MRS fitting challenge data setup by ISMRM MRS study group



Fitting of the magnetic resonance spectroscopy (MRS) data plays an important role in the quantification of metabolite concentrations. A number of commercial and home-built packages are available and used by the MRS community to fit spectra. The question arose whether any one of these packages was superior to the others or whether they all perform similarly if appropriately used. Hence, in preparation for a workshop of the ISMRM MRS study group on MR Spectroscopy: from Current Best Practice to Latest Frontiers, which took place in August 2016, it was decided by the organizing committee, that this question should be tackled by a fitting challenge open to everybody, where a set of spectra would be evaluated. For this purpose, synthetic MRS data were generated for 28 datasets. Short-echo time PRESS spectra were simulated using ideal pulses for the common metabolites at mostly near-normal brain concentrations. A macromolecular contribution was also included. Modulations of signal-to-noise ratio (SNR), lineshape type and width, concentrations of γ-aminobutyric acid, glutathione and macromolecules, and inclusion of artifacts and lipid signals to mimic tumor spectra were included as challenges to be coped with.

The deposited data contains 28 datasets, basis sets, Cr dataset for testing scaling, and ground truth values for all datasets. The datasets and basis sets are provided in various formats described below. Here are details of what is inside: 28 datasets are provided in three formats: 1. text file (column 1: real part of water suppressed FID; column 2: imaginary part of water suppressed FID; column 3: real part of water FID; column 4: imaginary part of water FID) 2. LCModel (.RAW and .h2o) 3. jMRUI (ASCII txt formats for both water suppressed [WS] and water spectra) The metabolite basis sets with macromolecular baseline are provided in these four formats: 1. text file (column 1: real part of FID; column 2: imaginary part of FID) for each metabolite and macromolecular baseline (MMBL) 2. text file with no reference peak at 0 ppm (column 1: real part of FID; column 2: imaginary part of FID) for each metabolite and MMBL 3. LCModel (.RAW) for each metabolite and MMBL and .BASIS 4. jMRUI (ASCII txt file format) for each metabolite and MMBL Cr scaling. An additional dataset, Cr_10mM_test_water_scaling (text, LCModel, jMRUI), is provided for testing water scaling. This dataset contains a Cr spectrum which - with correct water scaling - should come out to be 10.83 mM. metabolite concentration = (relaxation corrected water concentration) * (metabolite relaxation correction) * metabolite area / water area relaxation corrected water concentration = 1 mol / 18.015 g * 0.6 * 0.78 g/ml * exp(-30 ms/110 ms) + 1 mol/18.015 g * 0.4 *0.65 g/ml * exp(-30 ms/80 ms) = 29697 mM metabolite relaxation correction = 1/(exp(-30 ms/160 ms)) = 1.206 The following parameters describe the simulated data: Frequency = 123.2 MHz Sequence: PRESS TE = 30 ms, TE1 = 11 ms, TE2 = 19 ms TR >> T1’s Spectral width = 4000 Hz Number of points = 2048 Tissue content: GM = 60%, WM = 40% Water content: GM = 0.78 g/ml, WM = 0.65 g/ml T2 of water: GM = 110 ms, WM = 80 ms T2 of metabolites = 160 ms All metabolites datasets were generated using ideal pulses except for MMBL which was obtained experimentally. Glucose was simulated using both anomers, 0.36 alpha-glucose and 0.64 beta-glucose. The datasets may contain some or all of the metabolites listed among the basis spectra. In addition, lipid resonances and artifacts may be present.

Funding information
Sponsorship: NIH
Date made availableApr 16 2021
PublisherData Repository for the University of Minnesota
Date of data productionMar 1 2016

Cite this