3313

Diffusion functional MRI with isotropic b-tensor encoding

Arthur Spencer¹, Inès de Riedmatten¹, Jasmine Nguyen-Duc¹, Filip Szczepankiewicz², and Ileana Jelescu¹
¹Department of Radiology, Lausanne University Hostpital (CHUV), Lausanne, Switzerland, ²Medical Radiation Physics, Clinical Sciences Lund, Lund University, Lund, Sweden

Synopsis

Keywords: fMRI Acquisition, fMRI (task based), non-BOLD fMRI; diffusion tensor encoding; white matter

Motivation: Diffusion fMRI (dfMRI) has the potential to overcome some of the limitations of BOLD fMRI. However, acquisitions with linear diffusion encoding are sensitive to the underlying fibre orientations and may therefore give variable sensitivity, particularly in white matter.

Goal(s): To assess the utility of isotropic b-tensor encoding in a diffusion fMRI acquisition (iso-dfMRI).

Approach: We acquired iso-dfMRI data during a visual stimulation task. We compared this to dfMRI with linear diffusion encoding (dir-dfMRI) and to BOLD.

Results: Iso-dfMRI detected activity in the visual system with a larger spatial extent than dir-dfMRI and was less dependent on underlying fibre orientations.

Impact: We highlight the utility of isotropic b-tensor encoding dfMRI for detecting activity independently of underlying fibre arrangement. Thus, temporal resolution can be increased compared to acquiring multiple linear encoding directions and fMRI sensitivity in white matter boosted compared to BOLD.

Introduction

Diffusion fMRI (dfMRI) measures changes in diffusivity resulting from microstructural alterations associated with neural activity, avoiding limitations of neurovascular coupling.^1–3 dfMRI acquisitions with linear diffusion encoding require a compromise between temporal resolution and sensitivity to underlying tissue directionality. We evaluated isotropic dfMRI (iso-dfMRI) using spherical b-tensor encoding to sensitise the signal to diffusion in all directions per signal acquisition.^4,5 We compared its functional contrast to that of dfMRI with linear diffusion encoding (dir-dfMRI) and to BOLD contrast.

Methods

We acquired T1-weighted, multishell DWI, and dfMRI data in (n=5) healthy volunteers using a 3T Siemens Prisma. In each subject, we acquired iso-dfMRI data with a prototype sequence,⁶ and dir-dfMRI with a twice-refocussed spin-echo EPI sequence with bipolar linear encoding gradients, during visual stimulation (Fig.1a). To minimise BOLD signal contributions via T2-weighting, we calculated ADC from the log-ratio of volumes acquired at alternating b-values (Fig.1b), using b₁=200 and b₂=1000s/mm². The dir-dfMRI scheme was designed to minimise BOLD contributions via T2-weighting (ADC timecourse), blood signal (b≥200s/mm²), and cross-terms with background gradients (TRSE);⁷ iso-dfMRI minimised the first two contributions. For each acquisition, we also analysed the b=200s/mm² time series as an approximation of SE-BOLD with a small diffusion weighting (pseudo-BOLD). We also acquired b=0 volumes with linear and reversed phase encoding for B₀ field inhomogeneity correction. Imaging parameters are given in Figure 1c. The image volume encompassed the occipital lobe and optic radiation. MP-PCA denoising^8–10 was applied to each b-value time series separately. We then applied Gibbs unringing,¹¹ Topup,¹² motion correction and skull-stripping,¹³ and calculated ADC. CSF and white matter (WM) were segmented from the coregistered T1-weighted image. Task-induced activation was mapped using FSL FEAT,¹⁴ as follows. CSF voxels were excluded. Data were highpass filtered with a 100s cutoff, then a general linear model was used to investigate association with the task, modelled as a boxcar function. For both iso-dfMRI and dir-dfMRI, we measured voxelwise z-scores for: i) negative association with ADC; and ii) positive association with pseudo-BOLD. Cluster correction was applied to the resulting z-maps (z≥2.3, p<0.05). To measure the sensitivity of dfMRI methods to directionality, DWI data were preprocessed (MP-PCA denoising, Gibbs unringing, Topup and Eddy¹⁵) then the diffusion tensor was fitted to the b=1000s/mm² shell using FSL's FDT.¹⁶ This was used to estimate the diffusivity along the dir-dfMRI encoding gradient, calculated within significant voxels (cluster-corrected z-maps) which fell within the WM mask for dir-dfMRI and iso-dfMRI separately.

Results

Group-average spatial maps show robust activation in the visual cortex and proximal WM (Fig.2). In the iso-dfMRI spatial maps, grey matter made up 53% of voxels for ADC and 78% for pseudo-BOLD. For dir-dfMRI, the proportion of grey matter voxels were 47% for ADC and 72% for pseudo-BOLD. Task responses averaged across significant voxels, epochs and subjects (Fig.3) show the ADC response function reached a similar peak amplitude for iso-dfMRI and dir-dfMRI (around -1.5%), and reached peak amplitude sooner than pseudo-BOLD. Dir-dfMRI sharply returned to baseline upon task termination while iso-dfMRI mirrored the delayed BOLD. Figure 4 shows the distribution of diffusivity along the dir-dfMRI encoding gradient across active voxels in all subjects. The median for dir-dfMRI was lower than iso-dfMRI (p=0.0056, Mann-Whitney U-test).

Discussion

For both iso-dfMRI and dir-dfMRI, the ADC spatial maps had smaller spatial extent than pseudo-BOLD, possibly reflecting higher specificity to neural activity due to not being dependent on neurovascular coupling. The ADC spatial map for iso-dfMRI had larger spatial extent than for dir-dfMRI, suggesting improved sensitivity but possibly also BOLD contamination due to limited compensation of cross-terms with background gradients, as apparent in the sluggish return to baseline. In future, cross-term waveform designs may alleviate this contamination.¹⁷ ADC spatial maps for iso-dfMRI and dir-dfMRI contained a higher proportion of white matter voxels than pseudo-BOLD, suggesting increased preference for detecting activity in WM. In active WM voxels identified by dir-dfMRI, diffusivity along the diffusion encoding gradient was significantly lower than in those identified by iso-dfMRI, suggesting the isotropic encoding scheme is less sensitive to underlying fibre direction and has great potential for yielding unbiased dfMRI activation maps.

Conclusion

Iso-dfMRI enables balanced mapping of neural activity across grey and white matter, with less sensitivity to fibre orientation and higher temporal resolution compared to linear encoding dfMRI acquisitions which rely on acquiring multiple diffusion directions to achieve directional independence. Future work will focus on further eliminating BOLD contamination from the iso-dfMRI scheme, with the goal of reconstructing WM activation pathways across the brain.

Acknowledgements

This work was supported by ERC Starting Grant 'FIREPATH', SERI no. MB22.00032.

References

1. Le Bihan D, Urayama S, Aso T, Hanakawa T, Fukuyama H. Direct and fast detection of neuronal activation in the human brain with diffusion MRI. Proceedings of the National Academy of Sciences. 2006;103(21):8263-8268. doi:10.1073/pnas.0600644103

2. Nicolas R, Gros-Dagnac H, Aubry F, Celsis P. Comparison of BOLD, diffusion-weighted fMRI and ADC-fMRI for stimulation of the primary visual system with a block paradigm. Magn Reson Imaging. 2017;39:123-131. doi:10.1016/j.mri.2017.01.022

3. Aso T, Urayama S ichi, Fukuyama H, Le Bihan D. Comparison of diffusion-weighted fMRI and BOLD fMRI responses in a verbal working memory task. Neuroimage. 2013;67:25-32. doi:10.1016/j.neuroimage.2012.11.005

4. Szczepankiewicz F, Westin CF, Nilsson M. Gradient waveform design for tensor-valued encoding in diffusion MRI. J Neurosci Methods. 2021;348:109007. doi:10.1016/j.jneumeth.2020.109007

5. Nunes D, Ianus A, Shemesh N. Layer-specific connectivity revealed by diffusion-weighted functional MRI in the rat thalamocortical pathway. Neuroimage. 2019;184:646-657. doi:10.1016/j.neuroimage.2018.09.050

6. Szczepankiewicz F, Sjölund J, Ståhlberg F, Lätt J, Nilsson M. Tensor-valued diffusion encoding for diffusional variance decomposition (DIVIDE): Technical feasibility in clinical MRI systems. PLoS One. 2019;14(3):e0214238. doi:10.1371/journal.pone.0214238

7. Olszowy W, Diao Y, Jelescu IO. Beyond BOLD: Evidence for diffusion fMRI contrast in the human brain distinct from neurovascular response. bioRxiv. Published online June 30, 2021.

8. Cordero-Grande L, Christiaens D, Hutter J, Price AN, Hajnal J V. Complex diffusion-weighted image estimation via matrix recovery under general noise models. Neuroimage. 2019;200:391-404. doi:10.1016/j.neuroimage.2019.06.039

9. Veraart J, Novikov DS, Christiaens D, Ades-aron B, Sijbers J, Fieremans E. Denoising of diffusion MRI using random matrix theory. Neuroimage. 2016;142:394-406. doi:10.1016/j.neuroimage.2016.08.016

10. Veraart J, Fieremans E, Novikov DS. Diffusion MRI noise mapping using random matrix theory. Magn Reson Med. 2016;76(5):1582-1593. doi:10.1002/mrm.26059

11. Kellner E, Dhital B, Kiselev VG, Reisert M. Gibbs-ringing artifact removal based on local subvoxel-shifts. Magn Reson Med. 2016;76(5):1574-1581. doi:10.1002/mrm.26054

12. Andersson JLR, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage. 2003;20(2):870-888. doi:10.1016/S1053-8119(03)00336-7

13. Avants BB, Tustison N, Johnson H. Advanced Normalization Tools (ANTS). Insight j. 2009;365:1-35.

14. Woolrich MW, Ripley BD, Brady M, Smith SM. Temporal Autocorrelation in Univariate Linear Modeling of FMRI Data. Neuroimage. 2001;14(6):1370-1386. doi:10.1006/nimg.2001.0931

15. Andersson JLR, Sotiropoulos SN. An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. Neuroimage. 2016;125:1063-1078. doi:10.1016/J.NEUROIMAGE.2015.10.019

16. Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM. FSL. Neuroimage. 2012;62(2):782-790. doi:10.1016/j.neuroimage.2011.09.015

17. Szczepankiewicz F, Sjölund J. Cross-term-compensated gradient waveform design for tensor-valued diffusion MRI. J Magn Reson. 2021;328:106991. doi:10.1016/j.jmr.2021.106991

Figures

Figure 1: a) dfMRI runs were acquired during visual stimulation consisting of 16 blocks of baseline-active-baseline at 6-12-12s, with four interspersed 30s baseline blocks (10 minutes total). b) To minimise BOLD signal contributions via T2-weighting, ADC was calculated from the log-ratio of volumes acquired at alternating b-values. c) Imaging parameters for both iso-dfMRI and dir-dfMRI. Both were acquired with TR=1s, giving the resulting ADC timeseries a temporal resolution of 2s.

Figure 2: Group-level spatial maps showing significant association of task status with negative ADC (blue) and positive pseudo-BOLD (red-yellow), for dir-dfMRI (top) and iso-dfMRI (bottom). Subject-specific cluster-corrected (z≥2.3, p<0.05) spatial maps were binarised and coregistered to MNI standard space. The colourbar indicates the average z-score. Both dir-dfMRI and iso-dfMRI had smaller spatial extent with higher specificity to visual areas.

Figure 3: Normalised response functions for negative ADC (blue) and positive pseudo-BOLD (red), averaged across all active (cluster-corrected) voxels, epochs and subjects, for dir-dfMRI (left) and iso-dfMRI (right). The grey shaded area shows the task period. The shaded area around each line shows the standard error. ADC reached a similar peak amplitude for dir-dfMRI and iso-dfMRI, and reached peak amplitude sooner than pseudo-BOLD. However, dir-dfMRI returned to baseline sooner than iso-dfMRI.

Figure 4: Histograms of the diffusivity in the direction of the dir-dfMRI encoding gradient, within active (cluster-corrected) white matter voxels for all subjects, shown for both iso-dfMRI (blue) and dir-dfMRI (orange). For dir-dfMRI, the distribution is skewed towards lower values, indicating sensitivity to underlying fibre orientation.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)

3313

DOI: https://doi.org/10.58530/2024/3313