3448

Feasibility of cervical spinal cord cross-sectional measurements from 3D T1w sagittal head MRI at 7T
Vanessa Wiggermann1, Henrik Lundell1, Mads Alexander Just Madsen1, Christopher Fugl Madelung1, and Hartwig Roman Siebner1,2,3
1Danish Research Centre for Magnetic Resonance, Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark, 2Dept. of Neurology, Copenhagen University Hospital Bispebjerg, Copenhagen, Denmark, 3Institute for Clinical Medicine, Faculty of Medical and Health Sciences, University of Copenhagen, Copenhagen, Denmark

Synopsis

Spinal cord atrophy is a highly relevant measure of disease progression in multiple sclerosis. Here, we assessed whether cord area measurements from head MRI scans at 7T are feasible with semi- and fully-automated tools. Although different tools can yield systematically different absolute cord areas, measurements were highly consistent within studies as well as across field strengths, if images with similar voxel sizes were used. At higher spatial resolution, the volume of partial signal voxels is reduced and smaller cord areas and diameters were measured than typically reported in literature.

Introduction

Spinal cord atrophy occurs at about 2% per year in multiple sclerosis (MS)1, making it a relevant marker of disease progression and a potential endpoint measure in MS clinical trials2-5. However, due to scan time limitations, dedicated cord assessments are often replaced by measurements from large field-of-view (FOV), sagittal head MRI scans that visualise the upper part of the cervical spinal cord. The cross-sectional area (CSA) at the C1/C2 cervical segment, which has been widely assessed, shows a moderately positive relationship with the extended disability status scale (EDSS) in MS6. Increased signal-to-noise ratio and improved spatial resolution are driving MS research toward 7T. However, head-neck coils are not yet widely available for 7T and enhanced susceptibility artifacts can lead to a significant signal drop in infratentorial brain regions, potentially hampering cord measurements. We assessed whether high spatial resolution MPRAGE scans at 7T provide reliable CSA measurements at C1/C2 and whether automatised tools, e.g. the spinal cord toolbox (SCT)7, can be employed. Finally, we compared CSA-estimates at 7T and 3T.

Methods

7T data were collected in 46 MS patients (34 relapsing-remitting (RR)MS, 12 secondary-progressive (SP)MS, median EDSS 3) and 23 healthy controls (HC) on a Philips Achieva (Best, Netherlands) with a 32Rx/2Tx-channel head coil (Nova Medical, Wilmington, MA, USA) and motion correction8. Spinal cord segmentation was performed on 3D sagittal MPRAGE images (0.65 mm isotropic) after N4 bias-field correction with ANTs9. Using SCT version 4.3, the whole cord was segmented with sct_deepseg_sc and manual labels were placed relative to the C2/C3 intervertebral disc. Then, individual cord levels and their average spatial dimensions were determined. For comparison, CSA was also estimated using a semi-automated, in-house developed tool (LT, implemented in Matlab)10 that assesses CSA over five adjacent slices by intensity thresholding, following the method by Losseff5. A second 7T dataset, which included 17 Parkinson’s patients and HC, was assessed relative to 3T data (Prisma, Siemens, Erlangen, Germany, 64Rx-channel head coil). At 7T, segmentation was performed on lower resolution 3D MPRAGE images (0.85x0.85x1 mm3) and at 3T on a multi-parameter mapping T1w scan (1 mm isotropic). In addition to CSA, LR (left-right) and AP (anterior-posterior) diameters were assessed. Segmentation results were compared by Pearson’s correlations and group differences were determined by Kruskal-Wallis test.

Results

The mean CSA, RL and AP measurements for cohort 1 are summarised in Table 1 and examples are shown in Figure 1. 13/69 initial cord segmentations needed manual annotation. On average, all SCT-measured dimensions were smaller than LT-measured dimensions. Both reported smaller CSA than typically found in literature (e.g. HC mean SCT-CSA=61.9 mm2 and LT-CSA=71.5 mm2, literature 80 mm2 1,11). Nevertheless, measurements were highly correlated, and absolute mean CSA differences between groups were consistent and similarly significant (Figure 2). Likewise, SCT-obtained LR and AP diameters were smaller than with LT (Figure 3). AP group differences were consistent for both approaches. For LR, only SCT-measured group differences were significant. CSA measurements in the second cohort were also smaller with SCT compared to LT, but overall closer to literature values. CSA was not significantly different between field strengths (p > 0.1, Wilcoxon rank sum test) (SCT 3T=63.2 ± 6.2 mm2 – 7T=66.2 ± 8.8 mm2; LT 3T=76.3 ± 6.3 mm2 – 7T=77.3 ± 9.1 mm2) (Figure 4).

Discussion

CSA estimates can vary due to differences in methodology and voxel resolutions. LT is intensity-dependent and known to segment up to 4.5-10% larger CSA1. In both of our datasets, LT-CSA was about 15% larger than SCT-measured CSA. We assessed C1/C2 as signal loss at lower C-spine levels becomes more frequent and cord motion and CSF flow increase10. The significant difference in CSA between RRMS and SPMS agrees with current literature12,13. Further, group differences in LT-estimated AP diameter, but not with LR, have been previously reported14. However, SCT-LR did differ between cohorts. This discrepancy may be due to the presence of nerve roots in the LR direction, which were captured with LT, but less so with the more conservative SCT segmentation (Fig. 1). Further work to validate and understand potential LR diameter group differences is needed. In trauma patients, AP and LR changes independently predict motor and sensory deficits10. With higher spatial image resolutions at 7T, the volume of partial signal voxels at the cord edge, and thereby CSA, are expected to be reduced. However, when we compared field strengths, there were no significant differences in CSA for either segmentation approach. Notably, mean CSA in cohort 2 was larger than for cohort 1 at 7T, despite older age. This was likely due to the different image voxel sizes. In cohort 2, the voxel size was still 72.3% of the 3T voxel volume, but for cohort 1 the MPRAGE voxel size was reduced to 27.5%.

Conclusion

We found that large-FOV MPRAGE images can be used to produce reliable C1/C2 CSA estimates at 7T, especially if signal loss in the neck region can be minimised. At high spatial resolution, CSA may be smaller than typically reported as the amount of partial volume voxels is reduced. More work is needed to systematically assess the effect of voxel size, together with signal-to-noise ratio and field strength, to accurately determine CSA and diameters.

Acknowledgements

This work was supported by the Danish Multiple Sclerosis Society [Grant numbers: A33409, A35202, A38506] and the Independent Research Fund Denmark [Grant number: 9039-00330A]. Data acquisition of the second cohort was funded by the Danish Council for Independent Research [Grant number: 7016-00226B], the Novo Nordisk Foundation [Grant number: NNF16OC0023090] and the Danish Parkinson’s Disease Association [Grant number: A289]. HL has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program [grant agreement No 804746]. The 7T scanner was donated by the John and Birthe Meyer Foundation and The Danish Agency for Science, Technology and Innovation [Grant no. 0601-01370B].

References

1. Lin X, et al. Measurement of spinal cord atrophy in multiple sclerosis. J Neuroimaging 2004;14:20S-26S.

2. Kearney H, et al. Spinal cord MRI in multiple sclerosis – diagnostic, prognostic and clinical value. Nat Rev Neurol 2015;11:327-338.

3. Cawley N, et al. Spinal cord atrophy as a primary outcome measure in phase II trials of progressive multiple sclerosis. MSJ 2018;24(7):932-941.

4. Prados F, Barkhof F. Spinal cord atrophy rates. Ready for prime time in multiple sclerosis clinical trials? Neurology 2018;91(4).

5. Losseff NA, et al. Spinal cord atrophy and disability in multiple sclerosis. Brain 1996;119:701-708.

6. Xiaodong S, et al. Correlation between EDSS scores and cervical spinal cord atrophy at 3T MRI in multiple sclerosis: A systematic review and meta-analysis. Mult Scler Relat Disord 2020;37:101426.

7. De Leener B, et al. SCT: Spinal cord toolbox, an open-source software for processing spinal cord MRI data. NeuroImage 2017;145(A):24-43.

8. Andersen M, et al. Improvement in diagnostic quality of structural and angiographic MRI of the brain using motion correction with interleaved, volumetric navigators. PLoS ONE 2019;14(5):e0217145.

9. Tustison NJ, et al. N4itk: improved n3 bias correction. IEEE Trans Med Imaging 2010;29(3):1310-1320.

10. Lundell H, et al. Independent spinal cord atrophy measures correlate to motor and sensory deficits in individuals with spinal cord injury. Spinal Cord 2011;49:70-75.

11. Kearney H, et al. Improved MRI quantification of spinal cord atrophy in multiple sclerosis. JMRI 2014;39:617-623.

12. Bernitsas E, et al. Spinal cord atrophy in multiple sclerosis and relationship with disability across clinical phenotypes. Mult Scler Relat Disord 2015;4(1):47-51.

13. Liu C, et al. Three dimensional MRI estimates of brain and spinal cord atrophy in multiple sclerosis. J Neurol Neurosurg Psychiatry 1999;66:323-330.

14. Lundell H, et al. Spinal cord atrophy in anterior-posterior direction reflects impairment in multiple sclerosis. Acta Neurol Scand 2017;136(4):330-337.

Figures

Figure 1: Example spinal cord segmentations at the C1/C2 level, SCT (left), LT (right). MPRAGE images without overlaid segmentation are shown for reference in the center. For display purposes, the SCT segmentations were upsampled to the LT standard and slices were visually matched. Arrows point toward partial segmentation of spinal nerve roots. For subject 1, (top row, female, RRMS, 33 years), mean SCT-CSA was 65.04 mm2, compared to LT-CSA=74.03 mm2. For subject 2, (bottom row, female, HC, 42 years), mean SCT-CSA was 53.49 mm2, compared to LT-CSA=61.26 mm2.

Table 1: Group mean and standard deviations of CSA, AP and LR diameter measurements in the three subgroups of cohort 1. The fourth column lists the differences in estimates between techniques (SCT vs. LT). The last two columns list the differences between the subgroups by technique. Here, the upper cell displays the difference between HC and RRMS, and the lower cell HC to SPMS.

Figure 2: Comparison of average CSA measurements at the C1/C2 level using SCT and the in-house technique (LT). Although SCT generally obtained lower CSA measurements, the relative reproducibility of CSA was very high, both on an individual level (see left regression of individual data) and on a group level (compare middle and right). Statistical analysis involved a non-parametric Kruskal-Wallis test with post-hoc multiple comparison correction by Tukey-Kramer. P-values are uncorrected for testing both SCT and LT.

Figure 3: Spinal cord anterior-posterior (AP) and left-right (LR) diameters estimated at 7T and compared between the two tools. Corresponding to the area measurements, SCT estimated diameters were consistently lower than with LT. Relative reproducibility of AP measurements was higher than for LR. LR measurements between groups were significantly different with SCT, but not with LT. Statistical analysis involved a non-parametric Kruskal-Wallis test with post-hoc multiple comparison correction by Tukey-Kramer. P-values are uncorrected for testing both SCT and LT.

Figure 4: CSA estimates at different field strengths in the second cohort. The top row compares CSA estimates of the two tools at 3T (left) and 7T (right). In line with the other cohort, SCT consistently estimated lower CSA than LT. The bottom row compares the LT (left) and CST (right) CSA-estimates at the two field strengths. CSA-estimates between field strengths were less variable with SCT. CSA estimates did not differ between field strengths. Note also that there was no difference between PD and HC, in line with current knowledge of PD pathology. (rmse - root mean square error)

Proc. Intl. Soc. Mag. Reson. Med. 29 (2021)
3448