Motion-correction enabled ultra-high resolution in-vivo imaging of the human brain at 7T
Daniel Gallichan1

1CIBM, EPFL, Lausanne, Switzerland

Synopsis

We extended previous work using 3D-FatNavs to enable motion-correction of ultra-high resolution structural acquisitions, including T1-, T2- and T2*-weighted images. Images are of exceptional quality and detail for in-vivo acquisitions.

Introduction

We recently successfully demonstrated the use of a highly-accelerated 3D-fat-excitation GRE at 2 mm isotropic resolution (3D-FatNav) to allow tracking and retrospective correction of T1-weighted MP2RAGE and T2-weighted 3D-TSE with sufficient precision to follow the microscopic movements of the head which are inevitable even for compliant subjects during extended scan durations [1]. Previous work involved incorporating the 3DFatNav into sequences where sufficient dead-time already exists for their insertion without increasing scan time or perturbing the steady-state of the host sequence. In the current work we extend their use to the GRE sequence, where the steady-state of the host sequence must also be considered. We also apply the technique to each of these sequences with ultra-high resolution protocols (<400 μm isotropic) covering the whole brain with very long scan times (up to 42 mins per scan, some scans repeated for averaging) which would be unachievable without motion correction at high precision.

Methods

All experiments were performed on a Siemens 7T head-only system fitted with a single-channel transmit, 32-channel receive RF coil (Nova Medical Inc.). We have recently demonstrated that the spatial resolution and acceleration factor of the 3D-FatNav can be selected to reach the desired compromise between the precision of the motion estimates and the duration of the 3D-FatNav [2]. For this work we decided to use a conservatively long FatNav duration for very high quality motion estimates – using the same parameters as described in ref. [1]: 2 mm isotropic resolution, 4×4 GRAPPA acceleration, ¾ partial Fourier undersampling in both PE directions – resulting in 1152 ms per 3D-FatNav. For the MP2RAGE and TSE protocols this still fits in existing dead-time without extending the scan-time, but for GRE will result in a longer scan. Previous work using motion-navigators with GRE was careful to create segments for the navigator with a matched-TR to the host sequence in order to preserve the steady-state condition [3]. We hypothesized that insertion of a longer navigator would also be possible after the acquisition of each k-space partition – as the transition into the steady state would then be the same during each partition (except for the very first). Figure 1 shows the relative timing of the 3 sequences. Estimated rigid-body motion-parameters were obtained using ‘spm_realign’ (http://www.fil.ion.ucl.ac.uk/spm/) and the full 3D k-space of the host sequence was corrected for this motion, using a NUFFT (http://web.eecs.umich.edu/~fessler/code/) to account for the resulting non-Cartesian k-space sampling. For MP2RAGE and TSE each reconstructed motion-corrected image was co-registered using rigid-body FLIRT (http://fsl.fmrib.ox.ac.uk/) before averaging. As the MP2RAGE data were acquired over 2 scanning sessions each image was also unwarped for gradient nonlinearities (https://github.com/ksubramz/gradunwarp) prior to co-registration. TSE and GRE images were bias-field corrected using ‘smoothn’ (http://www.biomecardio.com/matlab/smoothn.html). For the reconstructed TSE data the hippocampus was segmented manually using itk-SNAP (http://www.itksnap.org/).

Results and Discussion

Figure 2 shows the MP2RAGE ‘uniform’ image with 350 μm isotropic voxels, acquired with 4×31-minute scans over 2 sessions. Figure 3 shows the TSE image at 380 μm resolution (2×29-minute scans) and Figure 4 shows the GRE image at 380 μm resolution (1×42-minute scan). There is no evidence of image artifact arising from the disturbance to the steady-state caused by the addition of the 3D-FatNav. All the images show excellent contrast and detail. Figure 5 shows a 3D rendering of the hippocampus manually segmented from the TSE image. The exceptionally high isotropic resolution of the scan and the strong contrast allow both the digitations of the surface, as well as the intricately convoluted ‘tail’ to be clearly visualized.

Conclusion

Motion-correction enables high-quality images to be acquired at 7T at ultra-high spatial resolution for a range of image contrasts. We expect these kinds of protocols to be useful for probing the in-vivo brain in exceptional detail.

Acknowledgements

This work was in part supported by the Centre d’Imagerie BioMédicale (CIBM) of the EPFL, UNIL, UNIGE, HUG, CHUV and the Leenards and Jeantet Foundations, as well as SNSF project number 205321_153564.

References

1. Gallichan et al, MRM (Early View online)

2. Gallichan and Marques, MRM (under review)

3. Tisdall et al, proc ISMRM 2014 p882

Figures

Diagram showing relative timing of the 3D-FatNav inserted into the MP2RAGE, 3D-TSE and GRE sequences. For the GRE there is an increase in the total scan time, whereas for MP2RAGE and TSE it fits in existing dead-time.

Motion-corrected 350 μm isotropic MP2RAGE scan with matrix size 552x442x416 acquired in 4x31-minute scans (2 sessions). TE/TI1/TI2/TR = 3.84/800/2700/6000 ms, BW = 240 Hz/Px, α12 = 5°/7°, 3/4 partial Fourier undersampling in both phase-encoding directions.

Motion-corrected 380μm isotropic vendor-supplied 3D-TSE scan with variable flip-angle readout train acquired in 2x29-minute scans. Matrix size = 512×320×448, TE/TR = 381/2700 ms, Turbo factor = 166, echo train duration = 997 ms, BW= 296 Hz/Px, 3/4 partial Fourier undersampling in both phase-encoding directions.

Motion-corrected 380μm isotropic GRE scan. Matrix size = 512×444×320, TE/TR = 15.6/27 ms, flip angle = 11°, BW = 60 Hz/Px.

Hippocampus manually segmented from 380μm TSE data, rendered using Blender (www.blender.org).



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1816