Synopsis

Keywords: Low-Field MRI, Relaxometry

We demonstrated quantitative T₂ mapping at 0.064 T using a multi-echo spin-echo sequence and CALIPR subspace constrained image reconstruction. CALIPR reconstruction results had minimal evidence of undersampling, providing better artifact suppression and preservation of fine anatomical detail than compressed sensing reconstruction.

Phantom validation work was performed using 14 vials doped with known concentrations of MnCl₂ and showed a relationship between measured T₂ and MnCl₂ concentration in agreement with expectations from a mathematical model. In-vivo, we demonstrated short acquisition time (8m:40s or 5m:59s), strong agreement with a highly-sampled reference acquisition, and T₂ maps with good contrast between tissue types.

Introduction

With reduced cost, simplicity of operation and negligible installation requirements, ultra-low-field systems are poised to drastically increase the accessibility of MRI. Previous work has pioneered the development of relaxation time mapping at ultra-low field^1,2. However, inherently low SNR can lead to long acquisition times and limit the utility and accuracy of T₂ mapping in-vivo.
In this work we develop a rapid T₂ mapping protocol on a point-of-care, ultra-low field MRI system that does not compromise on SNR or T₂ map quality. This is accomplished by combining a multi-echo spin-echo pulse sequence with a low-rank reconstruction approach previously introduced for multi-component T₂ mapping at high field³.

Methods

All data was acquired on a 64mT Hyperfine Swoop System (hardware version 1.7, software rc8.5.0).

Acquisition
A T₂ mapping acquisition was developed with three versions:
1) “Reference”: for comparison (3D multi-echo fast spin echo, FOV=180x220x180 mm³, resolution=3x3x5 mm³, 120 echoes, echo spacing=4.1 ms, TR=1500 ms, k-space undersampling factor=1.0, partial Fourier=0.7 in both the phase and slice directions, acquisition time=33m:35s),
2) “Accelerated 120 Echo”: identical to the Reference but with k-space undersampling factor=4.0 (acquisition time=8m:40s)
3) “Accelerated 80 Echo”: identical to the Accelerated 120 Echo version but with 80 echoes instead of 120 and TR=1250ms instead of 1500ms (acquisition time=5m:59s).

Reconstruction
Image reconstructions were performed offline in Matlab (R2019b) using BART⁴. Reconstructions were performed using non-uniform fast Fourier transform (nuFFT), compressed sensing (CS), as well as the constrained, adaptive, low-dimensional, intrinsically precise reconstruction (CALIPR) framework³, a recently introduced approach for subspace constrained reconstructions. ESPIRiT was used for coil sensitivity map estimation⁵. CS and CALIPR reconstructions were solved using FISTA⁶ on GPU (NVIDIA Titan RTX) with L1 wavelet regularization factors of λ=0.002 and λ=0.005 for phantom and in-vivo data, respectively. All CALIPR reconstructions used a fixed subspace size of 8.

Phantom Experiment
The Accelerated 120 Echo sequence was acquired in a phantom (Mini Hybrid Phantom Model 137, serial number CMRI 137-0002, CaliberMRI, Inc., Boulder, CO USA), containing 14 separate vials with known concentrations of MnCl₂. The behaviour of T₂ relaxation times as a function of MnCl₂ concentration $$$ c $$$ should follow:
$$ \frac{1}{T_{2}(c)} = \frac{1}{T_{2}(c=0)} + Rc $$
for some unknown T2 of distilled water at 0.064 T, $$$ T_{2}(c=0) $$$, and nuclear relaxivity $$$ R $$$.
Regions of interest were manually delineated in the centre of each vial, from which mean and standard deviation T₂ times were extracted. Because the true T₂ times of the vials at 0.064 T are unknown, $$$ T_{2}(c) $$$ data was fit to this model to assess whether our results were consistent with this known model.

In-Vivo Experiments
All three versions of the T₂ mapping sequence were acquired for a single healthy volunteer. Default T₂w and FLAIR anatomical images were acquired for the same subject in another session.

Note that apart from Figure 1, all T₂ mapping results used CALIPR reconstructions.

Results

Compared to nuFFT and CS reconstructions with the same Accelerated 80 Echo dataset (Figure 1), the CALIPR images and T₂ maps showed minimal evidence of undersampling, with better artifact suppression and preservation of fine anatomical detail.

For the phantom experiment in Figure 2, the T₂ map showed the expected trend in T₂ times between vials, based on the known concentrations of MnCl₂. With appropriate scaling, all the vials could be distinguished visually by their T₂ times. Our measured T₂ as a function of MnCl₂ concentration agreed with the expected model (Figure 3).

Figure 4 showed excellent agreement between the Reference, Accelerated 120 Echo, and Accelerated 80 Echo in-vivo datasets reconstructed using the CALIPR framework. Compared to the Reference sequence, the Accelerated results show only subtle loss of detail and do not appear to introduce a bias in the estimated T₂ values.

In Figure 5, T₂w and FLAIR anatomical images are supplemented with our in-vivo T₂ map to demonstrate an example protocol that can be achieved with an ultra-low field system.

Discussion

Despite the inherently lower signal-to-noise ratio, some aspects of ultra-low field MRI physics are uniquely favourable to T₂ mapping. Negligible specific absorption rate limitations allow for long echo trains of closely spaced 180° refocusing pulses. Shorter T₁ relaxation times at ultra-low field do not require as long an effective TR (TR-TE_max) to recover the signal, which boosts SNR efficiency by minimizing dead time in T₂-weighted spin echo sequences.

Noise robustness of the CALIPR framework reconstruction is especially apparent at later echo times, where nuFFT and CS reconstructions show artifacts resulting from low signal, due to T₂ decay.

Our T₂ estimation in the phantom vials with the lowest MnCl₂ concentrations have large standard deviations suggestive of imprecise T₂ estimation. This variability is unsurprising, as T₂ >> TE_max for these vials and the temporal sampling is insufficient to characterize these long T₂ components. However, our current echo train is sufficient to characterize most brain tissue, which has T₂ on the order of 100 ms at 0.064 T¹.

Conclusion

Here we developed an approach for performing rapid, in-vivo T₂ mapping on ultra-low field MRI systems. Despite rapid acquisition (6-9 minutes), this technique achieved sufficiently high SNR to characterize a wide range of T₂ times in phantom and distinguish brain tissue types in-vivo.

Acknowledgements

We thank Hyperfine Inc for their involvement and support. FP, RPAGT, MEP, MH, and SR are employed by Hyperfine Inc. EL, SW and SHK received funding from the Bill & Melinda Gates Foundation. This research is additionally funded by Natural Sciences and Engineering Research Council of Canada (AVD - Alexander Graham Bell Canada Graduate Scholarship, SHK- Grant RGPIN-2018-03904) and Michael Smith Healthcare BC (SHK).

References

1 O’Reilly, T. & Webb, A. G. In vivo T1 and T2 relaxation time maps of brain tissue, skeletal muscle, and lipid measured in healthy volunteers at 50 mT. Magnetic Resonance in Medicine 87, 884-895 (2022).
2 Jordanova, K. V. et al. Quantitative MRI measurements for T1 and T2 at 0.064T. In Proceedings of the 31st Annual Meeting of ISMRM (2022).
3 Dvorak, A. V. et al. The CALIPR framework comprehensively improves acquisition, reconstruction & analysis of multi-component relaxation imaging. In Proceedings of the 31st Annual Meeting of ISMRM (2022).
4 Uecker, M. The Berkeley Advanced Reconstruction Toolbox (BART). DOI: 10.5281/zenodo.592960. doi:10.5281/zenodo.592960 (2019).
5 Uecker, M. et al. ESPIRiT--an eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA. Magn Reson Med 71, 990-1001, doi:10.1002/mrm.24751 (2014).
6 Beck, A. & Teboulle, M. A Fast Iterative Shrinkage-Thresholding Algorithm for Linear Inverse Problems. Siam Journal on Imaging Sciences 2, 183-202, doi:10.1137/080716542 (2009).
7 Bloembergen, N. & Morgan, L. Proton relaxation times in paramagnetic solutions. Effects of electron spin relaxation. The Journal of Chemical Physics 34, 842-850 (1961).