2973

Clinically viable g-ratio imaging with unified readout at 3T: evaluation and comparison
Francesco Grussu1,2, Marco Battiston1, Ferran Prados1,3,4, Torben Schneider5, Enrico Kaden2, Rebecca S. Samson1, Daniel C. Alexander2, and Claudia A. M. Gandini Wheeler-Kingshott1,6,7

1Queen Square MS Centre, UCL Queen Square Institute of Neurology, Faculty of Brain Sciences, University College London, London, United Kingdom, 2Centre for Medical Image Computing, Department of Computer Science, University College London, London, United Kingdom, 3Centre for Medical Image Computing, Department of Medical Physics and Biomedical Engineering, University College London, London, United Kingdom, 4Universitat Oberta de Catalunya, Barcelona, Spain, 5Philips UK, Guildford, Surrey, United Kingdom, 6Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy, 7Brain MRI 3T Research Centre, IRCCS Mondino Foundation, Pavia, Italy

Synopsis

The current way of performing g-ratio imaging is via multi-modal MRI with mixed readouts. This approach can limit the geometrical correspondence of multi-modal maps required for g- ratio calculation. Here we compare g-ratio imaging performed with innovative unified spin echo EPI readout to standard mixed-readout measurements (spoiled gradient echo and spin echo EPI). A unified readout is a feasible alternative to mixed readouts: both provide biologically plausible metrics, with comparable scan-rescan repeatability. Additionally, a unified EPI readout is compatible with multiband acceleration, and enables joint multi-contrast modelling. Our work paves the way for richer multi-contrast acquisitions that could improve g-ratio imaging.

Introduction

G-ratio MRI is a recent approach that estimates myelin g-ratio (e.g. unmyelinated/myelinated axon radius ratio) by combining myelin and axonal density indices1–3. These are typically obtained with protocols encompassing various readouts, e.g. spoiled gradient echo (SGrE) for myelin-sensitive measurement and spin echo (SE) echo planar imaging (EPI) for axonal measurement via diffusion MRI (dMRI)4.

Distinct readouts exhibit different susceptibility to motion, physiological/thermal noise and distortions, limiting the spatial correspondence of myelin and axonal indices. Here we investigate multi-contrast microstructural imaging with a unified signal readout that provides geometrically-matched myelin and axon metrics. For the first time, we compare directly this new approach to standard g-ratio imaging based on mixed readouts. We also test acceleration strategies with multiband (MB) imaging, ultimately aiming at faster acquisition protocols that favour clinical translation.

Methods

MRI protocols

We implemented on a 3T Philips Achieva scanner 3 protocols enabling macromolecular tissue volume (MTV) and neurite density (ND) mapping (figures 1-2), which provide g-ratio3 as

$$g \,\,\,=\,\,\,\ \sqrt{1 \,\,-\,\, \frac{MTV}{MTV \,+\, (1 - MTV)\, ND}}. \,\,\,\,\,\,(eq.1)$$

  • UniEPI (proposed): unified readout5 with inversion recovery6 (IR) and multi-TE SE for MTV and dMRI for ND, all based on 2D SE EPI7.
  • SGrE+EPI (standard): 3D multi-echo SPGrE with variable flip angle, including actual flip angle measurement, for MTV3; 2D SE EPI for ND via dMRI (same scan time as UniEPI).
  • UniEPImb: faster version of proposed UniEPI (MB acceleration8).

All protocols included a 3D T1-weighted scan (3DT1-w; see captions, figures 1-2).

MRI sessions

We scanned two male volunteers (age: 28 and 29) twice with each of UniEPI and SGrE+EPI. Subject 2 was also scanned once with UniEPImb.

Pre-processing

Protocol-specific pre-processing was as follows.

  • UniEPI and UniEPImb: EPI distortion, eddy current and motion correction (FSL topup and eddy9), denoising (MrTrix3 dwidenoise10) in dMRI; EPI distortion correction and rigid alignment to dMRI for IR and multi-TE (FSL flirt11 and applywarp), which creates a unified EPI space.
  • SGrE+EPI: same pre-processing for dMRI as for UniEPI/UniEPImb; rigid alignment of SGrE data to 3DT1-w scan for motion correction.

For all protocols, 3DT1-w scans were segmented using GIF12 into white/grey matter (WM/GM) and affinely co-registered to dMRI using NiftyReg reg_aladin13.

Metrics

Fitting provided several metrics:

  • UniEPI and UniEPImb: apparent proton density (aPD) and T1 (inversion recovery); T2 (multi-TE SE); neural diffusivity (D) and ND (dMRI via two-compartment spherical mean technique14). aPD, T1 and T2 were processed to obtain MTV using the pseudo-PD method15.
  • SGrE+EPI: aPD and T1 (variable flip angle SGrE); T2* (SGrE echoes); D and ND (dMRI). aPD, T1 and T2* were combined into MTV15, and then all warped to the mean b=0 (EPI) image.

MTV and ND were combined into g-ratio in WM according to eq. 1.

Analysis

We aligned via affine co-registration mean b=0 images of all sessions for each subject16, creating a subject-specific space. In this space we evaluated tissue-specific distributions of metric values and scan-rescan percentage differences.

Results

Figures 3 shows voxel-wise metrics. All protocols provide metrics with comparable between-tissue contrasts, e.g. higher MTV and ND or lower T1 and T2/T2* in WM than GM. All protocols provide biologically plausible WM g-ratio values (circa 0.7).

Figure 4 shows tissue-specific distributions. Systematic between-protocol differences exist, e.g.: higher T1 and MTV or lower g-ratio from SGrE+EPI than UniEPI and UniEPImb; higher g-ratio and lower MTV from UniEPImb than UniEPI.

Figure 5 shows scan-rescan differences. UniEPI and SGrE+EPI respectively offer better repeatability for T1 and MTV (narrower distributions). Residual non-zero mean scan-rescan differences are seen, e.g. MTV (UniEPI, subject 1; SGrE+EPI, subject 2) and g-ratio (SGrE+EP, subject 1; UniEPI, subject 2).

Discussion

We studied clinically viable signal readouts for g-ratio MRI at 3T. We compared a unified SE EPI readout approach (protocol UniEPI) to a mixed readout (protocol SGrE+EPI), and tested MB acceleration (protocol UniEPImb). The proposed unified readout appears a feasible alternative to standard mixed-readout approaches: it provides metrics with similar contrasts and comparable overall scan-rescan repeatability of g-ratio and other microstructural indices. Additionally, a unified EPI readout is compatible with MB acceleration and enables joint multimodal computational modelling. This can potentially lead to significant gains in total scan time, which could translate in significant savings for the healthcare system. Finally, our work reveals systematic differences between myelination/relaxation indices obtained with different protocols. While a ground truth in vivo is not readily available, future cross-protocol metric harmonisation17,18 is warranted to deploy our approach in multi-site studies.

Conclusion

We demonstrate the potential of alternative g-ratio imaging with unified signal readout. Our work paves the way for joint computational modelling of rich multi-contrast acquisitions, compatible with the latest acceleration techniques.

Acknowledgements

This project has received funding under the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 634541 and 666992, and from: Engineering and Physical Sciences Research Council (EPSRC EP/R006032/1, M020533/1, G007748, I027084, M020533, N018702); Spinal Research (UK), Wings for Life (Austria), Craig H. Neilsen Foundation (USA) for INSPIRED; UK Multiple Sclerosis Society (grants 892/08 and 77/2017); Department of Health's National Institute for Health Research Biomedical Research Centres (BRC R&D 03/10/RAG0449); Guarantors of Brain post‐doctoral non‐clinical fellowships.

References

1. Stikov, N., et al., In vivo histology of the myelin g-ratio with magnetic resonance imaging. NeuroImage, 2015. 118: p. 397-405.

2. Mohammadi, S., et al., Whole-brain in-vivo measurements of the axonal g-ratio in a group of 37 healthy volunteers. Frontiers in neuroscience, 2015. 9: p. 441.

3. Duval, T., et al., g-Ratio weighted imaging of the human spinal cord in vivo. NeuroImage, 2017. 145 (Pt A): p. 11-23.

4. Ellerbrock, I. and S. Mohammadi, Four in vivo g‐ratio‐weighted imaging methods: Comparability and repeatability at the group level. Human Brain Mapping, 2018. 39(1): p. 24- 41.

5. Grussu, F., et al., A unified signal readout for reproducible multimodal characterisation of brain microstructure. Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM), 2017: p. 3399.

6. Battiston, M., et al., Fast and reproducible in vivo T1 mapping of the human cervical spinal cord. Magnetic resonance in medicine, 2018. 79(4): p. 2142-2148.

7. Stehling, M.K., R. Turner, and P. Mansfield, Echo-planar imaging: magnetic resonance imaging in a fraction of a second. Science, 1991. 254(5028): p. 43-50.

8. Feinberg, D.A. and K. Setsompop, Ultra-fast MRI of the human brain with simultaneous multi-slice imaging. Journal of Magnetic Resonance, 2013. 229: p. 90-100.

9. Andersson, J.L. and S.N. Sotiropoulos, An integrated approach to correction for off- resonance effects and subject movement in diffusion MR imaging. Neuroimage, 2016. 125: p. 1063-1078.

10. Veraart, J., et al., Denoising of diffusion MRI using random matrix theory. NeuroImage, 2016. 142: p. 394-406.

11. Jenkinson, M., et al., Improved optimization for the robust and accurate linear registration and motion correction of brain images. Neuroimage, 2002. 17(2): p. 825-841.

12. Cardoso, M.J., et al., Geodesic information flows: spatially-variant graphs and their application to segmentation and fusion. IEEE Transactions on Medical Imaging, 2015. 34(9): p. 1976-1988.

13. Ourselin, S., et al., Reconstructing a 3D structure from serial histological sections. Image and Vision Computing, 2001. 19(1): p. 25-31.

14. Kaden, E., et al., Multi-compartment microscopic diffusion imaging. NeuroImage, 2016. 139: p. 346-359.

15. Volz, S., et al., Quantitative proton density mapping: correcting the receiver sensitivity bias via pseudo proton densities. NeuroImage, 2012. 63(1): p. 540-552.

16. Leung, K.K., et al., Consistent multi-time-point brain atrophy estimation from the boundary shift integral. Neuroimage, 2012. 59(4): p. 3995-4005.

17. Stikov, N., et al., On the accuracy of T1 mapping: searching for common ground. Magnetic Resonance in Medicine, 2015. 73(2): p. 514-522.

18. Fortin, J.-P., et al., Harmonization of multi-site diffusion tensor imaging data. Neuroimage, 2017. 161: p. 149-17.

Figures

FIGURE 1: Details of the UniEPI (unified 2D SE EPI readout, no multiband) and UniEPImb (unified 2D SE EPI readout, multiband acceleration of 2) protocols. Both also included a 3D T1-weighted (3DT1-w) turbo-field echo scan (tip angle: 8°; TR: 6.9 ms; TE: 3.1 ms; TI: 829 ms; resolution: 1×1×1 mm3; field-of-view: 256×256×180 mm3; TFE: 230 for 131 TFE shots at 3000 ms intervals; SENSE: 2).

FIGURE 2: Details of the SGrE+EPI protocol (mixed 3D GrE and 2D SE EPI readout, without multiband acceleration for the 2D SE EPI). The protocol also included a 3D T1-weighted (3DT1-w) turbo-field echo scan (tip angle: 8°; TR: 6.9 ms; TE: 3.1 ms; TI: 829 ms; resolution: 1×1×1 mm3; field-of-view: 256×256×180 mm3; TFE: 230 for 131 TFE shots at 3000 ms intervals; SENSE: 2).

FIGURE 3: Examples of quantitative maps from subject 2, who was scanned with all protocols. Top row: metrics from protocol UniEPI (first scan); middle row: metrics from UniEPImb; bottom row: metrics from SGrE+EPI (first scan). Left to right: MTV, T1, T2 or T2*, ND, D and g-ratio. Maps are co-registered, easing visual comparison. Ventricles have been masked out.

FIGURE 4: Distribution of metric values from the first scan of all protocols in both subjects. First two rows on top: subject 1 (scanned with UniEPI and SGrE+EPI; white and grey matter); Last two rows on bottom: subject 2 (scanned with UniEPI, UniEPImb, SGrE+EPI; white and grey matter). Left to right: MTV, T1, T2, ND, D, g-ratio. Note that the acquisition protocol for ND and D (diffusion) is the same for UniEPI and SgrE+EPI .

FIGURE 5: Distribution of scan-rescan differences for protocols UniEPI and SGrE+EPI in the two subjects. First two rows on top: subject 1 (top to bottom: white and grey matter). Last two rows on bottom: subject 2 (top to bottom: white and grey matter). Left to right: scan-rescan percentage differences of MTV, T1, T2, ND, D, g-ratio. Note that the acquisition protocol for ND and D (diffusion) is the same for UniEPI and SGrE+EPI.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
2973