Synopsis

Keywords: Quantitative Imaging, Brain, T2 mapping

While recent works have determined that the DESS sequence has the ability to produce 3D T₂ mapping of the brain, it remains sensitive to B₀-related variation due to respiration. Multiple encoding strategies, using cartesian-spiral trajectories with different lengths combined with a variable density Poisson undersampling mask and Compressed-Sensing reconstruction, were consequently employed and compared in order to suppress respiration artifacts as well as to establish accurate and repeatable 3D T₂ maps of the brain at 3T.

Introduction

The Dual Echo Steady State (DESS) sequence has recently been extended to 3D brain¹ and breast imaging². However, its sensitivity to motion and B₀ variation^2,3 notably hinders any potential reproducible and accurate 3D T₂ mapping.
Previous work⁴ demonstrated that respiration induced ghosting artifacts on the DESS brain images could be corrected by making use of self-gating and re-ordering of the data combined with an acceleration approach, called SG-CS-DESS thereafter. Nevertheless, repeatability of the method had not been established.
As a result, the objective of this current work is to study this correction alongside different encoding strategies in order to measure their impact on 3D T₂ estimation and repeatability.

Methods

MR system
Acquisitions were carried out on a Siemens 3T PRISMA scanner, on two healthy volunteers. Image reconstruction was achieved using Matlab in conjunction with the Berkeley Advanced Reconstruction Toolbox⁵ (BART).

Sequence
A regular 3D Cartesian DESS sequence was firstly implemented (referred to as Fully DESS). The k-space lines were then acquired following a completely random order (Random DESS). The final scheme was a golden-step Cartesian trajectory with spiral profile ordering⁶ in the phase encoded directions (SP-DESS) (Fig. 1). This reordering was varied according to the size of the spiral interleaf i.e the number of acquired lines from one end of a spiral to the other. The previously described SG-CS-DESS⁴ was then implemented and the spiral reordered patterns were applied on it as well (SG-CS-SP-DESS). These versions used a trajectory with a variable density Poisson disk Compressed-Sensing (CS) mask in the k_y-k_z plane, an undersampling factor of 8, and a fully sampled center of 24x24 voxels. The mask sampling was modified between each repetition of the acquisition in order to maximize the total number of encoded lines obtained at the end of an acquisition.

Reconstruction
For the accelerated versions, the data was grouped into 7 respiratory bins by using the SG signal^4,7. Parallel imaging Compressed-Sensing reconstruction was then performed with spatial wavelet regularization and temporal regularization along the respiratory bins. This led to the reconstruction of 7 images that were averaged for SNR maximization through a sum-of-squares.
3D T₂ maps were then obtained by fitting the image data to a previously computed dictionary simulated with extended phase graph algorithms⁸, by using a range of T₁ and B₁ values.

MR Acquisition
The Fully, Random and SP-DESS sequences were performed in 6min43sec. When CS was used in the SG-CS and SG-CS-SP-DESS, 8 repetitions were set, lasting 7min11sec. The spiral interleaf length was equal to 26640 (all lines), 13320, or 144 for the SP-DESS, and 3560 (all lines), 500 or 144 for the SG-CS-SP-DESS. Each acquisition was performed twice (test-retest) with a 1.2mm isotropic resolution on each volunteer.
A MP2RAGE sequence⁹ (3min03sec for a 1mm isotropic voxel size) and a PreSat-TFL B₁ mapping sequence¹⁰ (33sec, 1.7 x 1.7 x 6 mm³) were also acquired to obtain voxel-wise T₁ and B₁ values and increase T₂ estimation accuracy.

Analysis
The brain white matter region was segmented with SPM12¹¹. Violin plots were computed from all voxels in this region for the 9 T₂ maps resulting from the 9 different DESS acquisitions. Bland-Altman plots for each DESS test-retest were then established by measuring the T₂ difference between both tests with respect to their T₂ average.

Results

The Fully DESS sequence produced ghosting artifacts that corrupted the anatomical images, while the artifacts were spread out over the images as noise in the Random and SP-DESS acquisitions (Fig. 2). Nonetheless, the T₂ maps remained inadequate for all 5 different fully versions, reflected by the wide distribution of their T₂ values in the white matter (Fig. 3) and the large voxel differences between T₂ maps (Fig. 4).
On the other hand, the CS versions generated artifact-less images with T₂ standard deviations approximately halved compared to the fully sequences (Fig. 2). The violin plots then depicted a reduced interquartile range and a much denser distribution of voxels around the median, especially for the SG-CS-SP-DESS with maximum interleaf length (Fig. 3). Lastly, robust repeatability was established with a considerably larger amount of voxels centered around a 0% difference (Fig. 4).

Discussion

By developing different encoding strategies with a goal of reducing respiratory induced artifacts, this work demonstrated a gradual improvement of DESS 3D T₂ estimation. At first, the random and spiral trajectories disrupted the coherence of the respiratory artifact which spread it out in the form of noise instead. However, this did not necessarily increase T₂ accuracy, and the drastic gradient changes in the random and short spiral interleaf trajectories could possibly produce Eddy currents which could not be accounted for. Consequently, combining a long interleaf spiral Cartesian trajectory with CS acceleration and a respiratory binning method previously validated led to artifact-less reconstructed images and repeatable T₂ maps.

Conclusion

We have optimized and validated a DESS sequence employing a spiral reordered cartesian encoding trajectory along with acceleration and respiratory artifact correction. The data obtained from this 11min protocol provides repeatable 3D T₂ maps of the brain that can be useful for longitudinal monitoring of tissue microstructure.

Acknowledgements

This study was achieved within the context of the Laboratory of Excellence TRAIL ANR-10-LABX-57. This work was also supported by the French National Research Agency (ANR-19-CE19-0014).

References

1. Gras, V. et al. (2017) ‘Diffusion-weighted DESS protocol optimization for simultaneous mapping of the mean diffusivity, proton density and relaxation times at 3 Tesla’, Magnetic Resonance in Medicine, 78(1), pp. 130–141.

2. Moran, C.J. et al. (2015) ‘Respiration Induced B₀ Variation in Double Echo Steady State Imaging (DESS) in the Breast’, 23rd annual Meeting and Exhibition of the ISMRM, pp. 498.

3. Barbieri, M. et al. A method for measuring B₀ field inhomogeneity using quantitative double-echo in steady-state. Magn Reson Med. 2022;1-17. DOI: 10.1002/mrm.29465

4. Kadalie, E.C. et al. (2022) ‘Respiration artifact-free 3D DESS T₂ mapping of the human brain’, 30th annual Meeting and Exhibition of the ISMRM, pp. 760.

5. BART Toolbox for Computational Magnetic Resonance Imaging, DOI: 10.5281/zenodo.592960

6. Prieto, C. et al. (2015) ‘Highly efficient respiratory motion compensated free-breathing coronary mra using golden-step Cartesian acquisition’, Journal of Magnetic Resonance Imaging, 41(3), pp. 738–746.

7. Rosenzweig, S. et al. (2020) ‘Cardiac and Respiratory Self-Gating in Radial MRI using an Adapted Singular Spectrum Analysis (SSA-FARY)’, IEEE Transactions on Medical Imaging, 39(10), pp. 3029–3041.

8. Weigel, M. (2015) ‘Extended phase graphs: Dephasing, RF pulses, and echoes - pure and simple’, Journal of Magnetic Resonance Imaging, 41(2), pp. 266–295.

9. Trotier, A.J. et al. (2019) ‘Compressed-Sensing MP2RAGE sequence: Application to the detection of brain metastases in mice at 7T’, Magnetic Resonance in Medicine, 81(1), pp. 551–559.

10. Chung, S. et al. (2010) ‘Rapid B₁+ mapping using a preconditioning RF pulse with TurboFLASH readout’, Magnetic Resonance in Medicine, 64(2), pp. 439–446.

11. https://www.fil.ion.ucl.ac.uk/spm