Oscar van der Heide^{1}, Alessandro Sbrizzi^{1}, Anna Kruseman^{1}, Martijn Cloos^{2}, Peter Luijten^{1}, and Nico van den Berg^{1}

MR-STAT is a framework for obtaining quantitative parameter maps from a single short scan. It is based on a time domain model. Large numerical inversion problems are solved to simultaneously localize signal and estimate tissue parameters. In this work we demonstrate the first experimental in-vivo results obtained with a clinical MR system.

A 2D balanced gradient-echo sequence is used. Spatial encoding is performed using a Cartesian readout scheme. A total of 32 full k-spaces are acquired, each one with a different flip angle and no waiting times. The flip-angles are chosen randomly between $$$5^{\circ}$$$ and $$$75^{\circ}$$$ (Fig. 1). The excitation phases alternate between $$$0^{\circ}$$$ and $$$180^{\circ}$$$. The sequence is preceded by an inversion pulse.

MR-STAT was implemented on a 3T whole-body MR system (Philips-Ingenia). Single slices are acquired for gadolinium-doped agarose gel-samples and a brain of a volunteer with a 15 channel receive head-coil. Total sequence time was 16s.

Sequence parameters were imported into MATLAB. A 1D Fourier transform was applied along the readout direction to decouple the problem into smaller subproblems. These were solved in parallel on a local High Performance Computing (HPC) facility using a MATLAB implementation of VarPro^{4} and a Bloch simulator^{5}. The reconstruction time (= longest job) was 12 minutes in both cases.

Our model does not require knowledge of RF transmit fields, off-resonance and/or proton density. The slice profile is included in the model. In this work we reconstruct $$$T_{1}$$$ and $$$T_{2}$$$. A 5-minute interleaved inversion-recovery and multi spin-echo sequence that comes with our MR-system was used to obtain $$$T_{1}$$$ and $$$T_{2}$$$ values for the phantoms as a reference.

Previous work on MR-STAT was limited to computational proof-of-principle demonstrations. In this work we have successfully applied MR-STAT for the first time on a clinical MR-system. The results show that MR-STAT can simultaneously perform signal localization and parameter estimation using time-domain data from a simple 16s scan.

Unlike MRF, MR-STAT is based on a deterministic process that monitors the accuracy and precision. Instead of relying on a pre-computed dictionary, a large-scale inversion problem is solved. 2D scan's results show that the reconstruction can be performed in reasonable times on a cluster.

Since Fourier Transform is not applied over the whole k-space (MR-STAT works in the time-domain) there is increased flexibility in sequence design.

Parallel imaging and/or compressed sensing strategies could be implemented to further accelerate scan time and/or increase precision.

[1] Ma D. et al, Nature 2013.

[2] Sbrizzi A. et al, ISMRM 2015, p. 3712.

[3] Sbrizzi A. et al, ISMRM 2016, p. 4226.

[4] O’Leary, D.P. & Rust, B.W. Comput Optim Appl (2013) 54: 579.

[5] Hargreaves, B., http://www-mrsrl.stanford.edu/~brian/blochsim.