Gastao Cruz^{1}, Aurelien Bustin^{1}, Olivier Jaubert^{1}, Torben Schneider^{2}, René M Botnar^{1}, and Claudia Prieto^{1}

Magnetic Resonance Fingerprinting (MRF) estimates simultaneous, multi-parametric maps from a dynamic series of highly undersampled time-point images. At very high undersampling factors, some of these artefacts may propagate into the parametric maps leading to errors. Here we propose the use of locally low rank regularization for a low rank approximation reconstruction to enable highly accelerated MRF. The proposed approach was evaluated in simulations and in-vivo brain acquisitions. Results show that the proposed approach enables accurate MRF reconstructions from ~600 time-point images with one radial spoke per time-point.

1. Ma D et al, Nature 2013; 495:187-192

2. McGivney et al, IEEE Trans Med Imaging 2014;33(12):2311-22

3. Doneva et al, MRI 2017; doi:10.1016/j.mri.2017.02.007

4. Zhao et al, MRM 2017; doi:10.1002/mrm.26701

5. Trzasko J et al, ISMRM, 2011; abstract 4371

6. Zhang et al, MRM 2015; 73(2):655-61

7. Tamir et al, MRM 2017; 77(1):180-95

8. Jiang et al, MRM 2015; 74:1621-1631

Figure 1: Two representative
time-point reconstructions (#109 and #1633) using low rank approximation
(LrMRF) and locally low rank regularization (LLR-LrMRF) with four different
amounts of data (Nt), for a simulation. Blurring and streaking artefacts (residual
aliasing) are present with LrMRF and are considerably reduced with the proposed
LLR-LrMRF. Some of these artefacts create errors in the parametric maps, as can
be seen in Figure 2.

Figure 2: T1 and T2
maps using the low rank approximation (LrMRF) and locally low rank
regularization (LLR-LrMRF) with four different amounts of data (Nt), for a
simulation. Artefacts from time-point images (primarily blurring) propagate
into errors in the parametric maps with LrMRF; these are considerably reduced
with the proposed LLR-MRF. At reduced amounts of data, higher relative root
mean square errors are observed for LrMRF than for LLR-LrMRF.

Figure 3: Two representative time-point
reconstructions (#109 and #1633) using low rank approximation (LrMRF) and
locally low rank regularized low rank approximation (LLR-LrMRF) with four
different amounts of data (Nt), for an in-vivo brain MRF acquisition. Blurring and
streaking artefacts (residual aliasing) are present with LrMRF, even for 1750
time-points, and are considerably reduced with the proposed LLR-LrMRF. Some of
these artefacts create errors in the parametric maps, as can be seen in Figure
4.

Figure 4: T1 and T2
maps using the low rank approximation (LrMRF) and locally low rank regularized
low rank approximation (LLR-LrMRF) with four different amounts of data (Nt), for an in-vivo brain acquisition.
Artefacts from time-point images (primarily blurring) propagate into errors in
the parametric maps with LrMRF; these are considerably reduced with the
proposed LLR-LrMRF. LLR-LrMRF achieved higher quality than LrMRF even for 1750
time-points.