4954

Structural covariance of diffusion metrics in mild COVID19: Spatially coherent effect on fractional anisotropy but not free water.

Nick Teller¹, Jordan A. Chad^1,2, Eugenie Roudaia¹, Ali Hashemi³, Haddas Grosbein¹, Asaf Gilboa^1,2, Maged Goubran^2,4, Ivy Cheng^2,4, Sandra E. Black^2,4, Robert Fowler^2,4, Chris Heyn^2,4, Fuqiang Gao⁴, Mario Masellis^2,4, Jennifer Rabin^2,4, Xiang Ji⁴, Aravinthan Jegatheesan^2,4, Benjamin Lam^2,4, Allison B. Sekuler^1,2,3, Bradley J. MacIntosh^2,4, Simon J. Graham^2,4, and J. Jean Chen^1,2
¹Rotman Research Institute, North York, ON, Canada, ²University of Toronto, Toronto, ON, Canada, ³Psychology, Neuroscience & Behaviour, McMaster University, Hamilton, ON, Canada, ⁴Sunnybrook Research Institute, Toronto, ON, Canada

Synopsis

The impact of COVID19 on the brain’s microstructural integrity remains unclear. In this study, we examine self-isolated COVID19 patients and controls using diffusion-tensor and free-water imaging, based on single- and multi-shell acquisitions, respectively. We identify several differences in spatial covariance among patients in fractional anisotropy (in cingulate-frontal and temporal-parietal regions), but not free water fraction. Our results indicate COVID19’s implications in long-term, measurable brain deficits.

Introduction

Infection of the central nervous system has been a significant cause of death and morbidity in COVID19¹. Previous studies have established reduced grey matter (GM) mean diffusivity (MD) in hospitalized patients². However, the less severe self-isolated patients, who make up a majority of COVID19 cases, remain understudied. Specifically, many of these patients report prolonged COVID symptoms that impair sensory and cognitive function ³, for which white-matter microstructural dysfunction is almost certainly involved. Hence, we use multi-shell free-water imaging (fwDTI) to elucidate disease mechanisms in WM of self-isolated patients. Our previous work has shown widespread differences in diffusion metrics of white-matter microstructure between patients and controls ^4,5. We also noticed distinct data variabilities between patient and control groups. To better visualize the spatial distribution of these variability differences, we compute spatial covariances between tissue regions of interest (ROIs) across the control and patient groups. Our approach is also inspired by the understanding that neurotropic viral infections are known to travel along white-matter structures⁶, and past work has shown spatial covariance ⁷ as a sensitive marker of systemic changes ⁸.

Methods

Participants
Here we report findings from 40 self-isolated COVID-19+ patients (mean age: 43.7, 68% female), and 14 controls (mean age: 43, 64% female) who had flu-like symptoms but are otherwise healthy.
Diffusion MRI Data
Diffusion MRI data acquisition occurred on a Siemens Prisma 3T system with 5 b=0, 34 b=700s/mm2 and 34 b=1400s/mm2 volumes at TR = 4.3 s, TE= 62ms, matrix size=96x96x60 with (2mm)3 resolution, 2-fold through-plane simultaneous multi-slice acceleration with 2 field-of-view shifts. A 1mm isotropic T1 anatomical was also acquired for each participant.
Data Processing
Diffusion metrics were obtained by using Dipy’s multi-shell free-water code to fit the b = 0, 700 and 1400 shells to a two-compartment fwDTI model⁹. These resulted in fractional anisotropy (FA) and free-water fraction (FW) maps for all participants. Furthermore, for the region of interest (ROI) analyses FreeSurfer reconstruction was performed, producing subject-specific WM tissue parcellations.
Statistical Analysis
Regional medians of diffusion metrics were obtained from all WM parcellations. These were then used to compute inter-regional covariance patterns across subjects, as defined by Eq. (1),
$$ cov (A,B) = \frac{1}{N-1} \sum_{i=1}^N (A_i- \mu_A)*(B_i- \mu_B) $$
where A and B represent vectors of diffusion metrics from patients and controls, derived from the same ROI, Ai and Bi represent elements of these respective matrices up to length N, and μ represents the group mean in each case. Correlation is the covariance normalized by the product of the variances across groups A and B. We focused on FA and FW as two complementary diffusion metrics.

Results

The group-mean FA is comparable between patients and controls (Fig. 1a). FA standard deviation is higher in patients than controls in the corpus callosum region, and higher in controls in a cluster that spans the lateral/medial-orbitofrontal, lingual and lateral occipital WM (Fig. 1b) Accordingly, higher spatial covariance is found in the corpus callosum region and lower spatial covariance is found in this frontal-lingual-occipital cluster in patients (Fig. 2a-c). The spatial-correlation difference between the groups shows a similar pattern. Conversely, for free-water, the group mean is higher in patients in the corpus callosum region (Fig. 3), but the standard deviation was similar in both groups. Accordingly, the covariance and correlation matrices revealed no coherent differences in spatial patterns (Fig. 4).

Discussion and Conclusion

This work shows preliminary data on the spatial-covariance patterns of two complementary diffusion metrics, FA and FW. FW shares interpretation with the well-known mean-diffusivity. We demonstrate structural coherence in FA, but less in FW. Our finding of elevated FA standard deviation and mean FW in the corpus callosum of the patient group suggest the greatest impact of COVID on this brain region, followed by a potential impact in the region that span the frontal, occipital and lingual regions. These latter regions are involved in a variety of functions, including visual processing and executive function. These regions were equally identifiable in the spatial-covariance matrices constructed using FA, but not when using FW. This is unsurprising, as spatial covariance is driven in large part by within-group variability, which was almost identical between the two groups in the case of FW. Our results suggest that in the case of COVID, FA and FW may represent different disease processes.

Acknowledgements

We would like to thank the CIHR and Sunnybrook Research Institute for their support.

References

1. Iadecola, C., Anrather, J. & Kamel, H. Effects of COVID-19 on the Nervous System. Cell (2020). doi:10.1016/j.cell.2020.08.028.

2. Lu, Y. et al. Cerebral Micro-Structural Changes in COVID-19 Patients - An MRI-based 3-month Follow-up Study. EClinicalMedicine 25, 100484 (2020).

3. Douaud, G., Lee, S., Alfaro-Almagro, F., Arthofer, C., Wang, C., Lange, F., Smith, S. M. Brain imaging before and after COVID-19 in UK Biobank. medRxiv (2021).

4. Chen J. J., Chad J. A., Ji X., MacIntosh B. J., Gilboa A., Roudaia E., Sekuler A.B., Lam B., Heyn C., Black S. E., Graham S. J.. COVID19 effects on brain tissue microstructure: Longitudinal study of self-isolated cases using diffusion MRI. OHBM (2021).

5. Chen J. J., Chad J. A., Ji X., MacIntosh B. J., Gilboa A., Roudaia E., Sekuler A.B., Lam B., Heyn C., Black S. E., Graham S. J.. COVID19 effects on brain tissue microstructure: Longitudinal study of self-isolated cases using diffusion MRI. ISMRM (2021).

6. Sun, N., & Perlman, S. Spread of a neurotropic coronavirus to spinal cord white matter via neurons and astrocytes. Journal of virology 69(2), 633-641 (1995).

7. Carmon, J., Heege, J., Necus, J.H., Owen, T.W., Pipa, G., Kaiser, M., Taylor, P.N., Wang,Y. Reliability and comparability of human brain structural covariance networks. NeuroImage 220, 117104 (2020).

8. Alexander, G. E., Lin, L., Yoshimaru, E. S., Bharadwaj, P. K., Bergfield, K. L., Hoang, L.T., Trouard, T. P. Age-Related Regional Network Covariance of Magnetic Resonance Imaging Gray Matter in the Rat. Frontiers in Aging Neuroscience 12 (2020). doi:10.3389/fnagi.2020.00267

9. Hoy, A. R., Koay, C. G., Kecskemeti, S. R. & Alexander, A. L. Optimization of a free water elimination two-compartment model for diffusion tensor imaging. NeuroImage vol.103 323–333 (2014).

Figures

Figure 1: Comparing WM FA mean and standard deviation by region in patients and controls. The group-mean free-water fraction values are similar across patients and controls, but patients exhibit higher group-wise standard deviation in the corpus callosum whereas controls exhibit higher standard deviation in a cluster of regions spanning the frontal, lingual and occipital regions (black boxes).

Figure 2: FA spatial covariance exhibits clear spatial clusters around the corpus callosum, frontal and temporal regions, seen as bright regions along the diagonal. Patients exhibit higher covariance in corpus callosum, whereas controls show higher covariance in the cluster spanning the frontal, lingual and occipital regions (red boxes). A similar pattern is shown in the spatial-correlation difference (f), with the exception that controls also have lower correlation in the bottom-right cluster (including frontal pole, temporal lobe and insula).

Figure 3: Comparing WM F mean and standard deviation by region in patients and controls. The group-mean free-water fraction is higher in patients than in controls (black box), but patients and controls exhibit similar group-wise standard deviations.

Figure 4: Free-water fraction only exhibits a covariance cluster around the corpus callosum (unlike FA, which is more spatially structured). Patients and controls exhibit similar covariance and correlation matrices. Free-water covariance is lower in controls (c), but there is no clear structure in the group difference in correlation (f).

Proc. Intl. Soc. Mag. Reson. Med. 30 (2022)

4954

DOI: https://doi.org/10.58530/2022/4954