MR Fingerprinting is a new multi-contrast imaging method that is appealing for its ability to acquire multiple quantitative maps efficiently when compared to other repeated acquisitions varying a single parameter at a time. This work shows our first demonstration of MRF in vivo repeatability through test/re-test imaging of seven healthy human volunteers at 1.5T and 3.0T. We found the group root-mean-square-difference (NRMSD) of MRF-SSFP T1 measurements to be 5-6% in grey and white matter at 1.5T and 4-5% at 3.0T, T2 NRMSD to be 6-9% at 1.5T and 11-13% at 3.0T, and PD NRMSD to be 2-3% at 1.5T and 2-3% at 3.0T.
MRF data were acquired on a 1.5T and a 3.0T MRI system (respectively HDxt and MR750 GE Healthcare, Waukesha, WI, USA), each equipped with an 8-channel dedicated brain coil. With local ethical approval, 7 healthy human subjects were imaged in two identical sessions per scanner, each including a 1mm-isotropic 3D fast spoiled gradient echo (FSPGR), and 2D steady-state free precession MRF2. Subjects were removed from the scanner between imaging sessions. The MRF acquisition trajectories used 89 undersampled golden-angle spiral interleaves with FOV = 256mm, matrix = 128x128, sampling bandwidth = ±250kHz, TE = 2.2ms. The scan parameters matched the repetition time and flip angle list from Jiang et al 2. We used 979 frames and 1 NEX per slice. The spirals were rewound and followed by a spoiler z-gradient achieving 8 dephasing through a 2mm slice. Three-dimensional volumes were reconstructed at 2mm isotropic resolution. The maximum gradient amplitude was 20mT/m and slew rate was 50T/m/s. MRF maps were obtained by inner-product pattern matching and the MRF dictionary was computed using the extended phase graphs formalism3; B0 and B1 were not included in the dictionary but slice profile was included to improve T2 accuracy4.
Analysis: M0 values were reported both un-normalised and self-normalised to the average inside the brain mask (here, we called normalised M0 data PDn). MRF maps for every test were rigidly co-registered to their respective re-test using Statistical Parametric Mapping (SPM 12) Toolbox. After registration, each subject’s volumetric image was warped to match a custom DARTEL atlas based on the subjects studied (the 3D FSPGR from the first test at 3T was used to estimate the transformation). After applying 6mm gaussian smoothing, the root mean square differences (RMSD) were calculated between each voxel in test and retest measurements, and voxel-wise normalized to the group average (NRMSD). For each subject, the segmented tissue probability maps were calculated and used to extract mean values of parameters in grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF). A threshold of 50% probability was used as a mask per each tissue
1. Yun Jiang, Dan Ma, Kathryn E. Keenan, Karl F. Stupic, Vikas Gulani, Mark A. Griswold. Repeatability of magnetic resonance fingerprinting T1 and T2 estimates assessed using the ISMRM/NIST MRI system phantom. Magn. Reson. Med. 2016. Doi: 10.1002/mrm.26509;
2. Jiang Y, Ma D, Seiberlich N, Gulani V, Griswold MA. MR fingerprinting using fast imaging with steady state precession (FISP) with spiral readout. Magn Reson Med 2015;74:1621–1631
3. Weigel M. Extended phase graphs: dephasing, RF pulses, and echoes - pure and simple. J Magn Reson Imaging2015;41:266–295.
4. Buonincontri, G., Sawiak, S.J., 2016. MR fingerprinting with simultaneous B1 estimation. Magn. Reson. Med. 76, 1127–1135. http://dx.doi.org/10.1002/mrm.26009