In vivo mapping of human spinal cord microstructure at 300mT/m

Abstract

The ability to characterize white matter microstructure non-invasively has important applications for the diagnosis and follow-up of several neurological diseases. There exists a family of diffusion MRI techniques, such as AxCaliber, that provide indices of axon microstructure, such as axon diameter and density. However, to obtain accurate measurements of axons with small diameters (<5μm), these techniques require strong gradients, i.e. an order of magnitude higher than the 40-80mT/m currently available in clinical systems. In this study we acquired AxCaliber diffusion data at a variety of different q-values and diffusion times in the spinal cord of five healthy subjects using a 300mT/m whole body gradient system. Acquisition and processing were optimized using state-of-the-art methods (e.g., 64-channel coil, template-based analysis). Results consistently show an average axon diameter of 4.5+/-1.1μm in the spinal cord white matter. Diameters ranged from 3.0μm (gracilis) to 5.9μm (spinocerebellar tracts). Values were similar across laterality (left-right), but statistically different across spinal cord pathways (p<10(-5)). The observed trends are similar to those observed in animal histology. This study shows, for the first time, in vivo mapping of axon diameter in the spinal cord at 300mT/m, thus creating opportunities for applications in spinal cord diseases.

Keywords: AxCaliber; Axon diameter; Diffusion MRI; Human; Quantification; Spinal cord.

Figure 1

Placement of slices (yellow), saturation bands (red) and shimming volume (green). Four slices were placed in the middle of the vertebral body at levels C1, C2, C3 and C4, by adjusting the slice gap for each subject. Slices were orthogonal to the SC. Optimal shim coefficients (up to 2nd order) were calculated within a small box encompassing the spinal cord. To prevent aliasing associated with reduced FOV, two saturation bands were prescribed anteriorly and posteriorly.

Figure 2

Illustration of the diffusion encoding gradients used in the AxCaliber protocol (a) and in the protocol for probing orientation dependence (b). The latter protocol aims at exploring fibers that are not oriented along Z (e.g., collateral fibers entering the dorsal aspect of the cord).

Figure 3

Examples of DW images with selected b-values in the lowest range (4302) used for motion correction. These images offer sufficient SNR for robust estimation of motion parameters without CSF contamination.

Figure 4

Estimated motion in anteroposterior direction (raw moco) and fitted spline functions (smooth moco) at each cervical level in one subject. All data with different Δ were concatenated. Here, “moco” stands for motion correction.

Figure 5

a: Mean high b-values images. b: image of the template used for registration. c: registered image after applying the deformation field. d: Five major axonal pathways with different morphological features were selected from the white matter atlas in order to extract model-based diffusion MRI metrics.

Figure 6

Example of images acquired at different q-vectors. Data are not interpolated. Contrast is kept the same for better comparison. Notice the low SNR at very-high q-value (orange), which was compensated by averaging over the four directions.

Figure 7

Data averaged along q-values in one subject (excluding images acquired at b < 430s/mm²), before and after applying the correction for eddy-currents and subject motion.

Figure 8

Top: Rician corrected q-space data in one voxel of the spinal cord white matter for one subject before LPCA correction (normalized by b=0). Bottom: same data averaged over the four directions. The purple dashed box shows the data collected for probing the orientation dependence (see Figure 2.b.).

Figure 9

Standard deviation of noise along q (blue curve) in one voxel and one subject before LPCA correction. Values are shown as percentage of the b=0 signal. This estimated noise includes thermal and physiological noise. Notice that the standard deviation is fairly constant along q.

Figure 10

a: Cuneatus (blue), gracilis (yellow) and rubrospinal (red) tracts highlighted on the mean DWI in one subject. b. Histological images of axons stained for myelin (luxol fast blue cross) over corresponding pathways of a human spinal cord (“Histology at the University of Michigan,” n.d.), reproduced with permission. c. Model fitting on signal decay acquired in one subject on a single voxel in the corresponding regions.

Figure 11

Maps of fitted parameters using single diameter model. Data histograms with range and mean value are shown at the bottom. The black arrows points to the posterior funiculus,

Figure 12

Top left: Mean DWI with overlay of ROIs for computing parameters within specific white matter tracts. Top right: Bar graph showing estimated axon diameter within tracts, laterality and subject. The estimated axon diameters range between 3.5 and 5.5μm, suggesting fairly precise estimate of axon diameters on an individual basis. Bottom table: Estimated parameters averaged across subjects. Mean axon diameter was 3.51 (+/−0.54), 4.15 (+/−0.46) and 3.71 (+/− 0.36) μm in the gracilis, cuneatus and spinothalamic tracts, respectively. The restricted water fraction (1-fh), which correlates with axon density, was 55% and 44% (+/− 2%) in the cuneatus and spinothalamic tracts, respectively. Results of the three-way ANOVA show a significant effect of pathway and subjects but no effect for laterality.

Figure 13

Difference in axon diameter estimated using two sub-sets of data with orthogonal diffusion gradient direction (X,Y;−X, −Y) and (−X,Y; X, −Y) in one subject. Symmetrical differences (red versus blue) are observed in the lateral region (especially at C1 and C2), which could be attributed to the presence of collateral fibers.

Figure 14

a. q-space sampling for orientation dependence study. b. Signal at different gradient orientations, which was detrended using cosines into a function representing the signal variation as a function of gradient orientation (“orientation dependence” plot). c. Directions of collateral fibers averaged across subjects at level C2. This map was obtained by extracting the angular value corresponding to the highest diffusion (i.e. lower signal). d: Corresponding map of orientation dependence obtained using the peak-to-peak amplitude from the orientation dependence plot.

Figure 15

Comparison of AxCaliber results with two histological resources. Left: Optical images (50×50μm²) of human thoracic spinal cord (“Histology at the University of Michigan,” n.d.), reproduced with permission. Middle: Cytoarchitecture of human spinal cord white matter at vertebral levels C1 and C5 (Nieuwenhuys et al., 2007), reproduced with permission. Axon size is gray-level coded (the darker the bigger). Note that this representation of axon diameter is qualitative. Notice that some tracts have monodisperse axonal sizes (e.g. spinocerebellar and gracilis), while others present some super-axons surrounded by tiny axons (e.g. Pyramidal tracts). For direct comparison, AxCaliber results (averaged over five subjects) are overlaid on the right portion of the cytoarchitecture map at the corresponding levels (note: given that we did not acquire lower than C4, the C4 level is shown next to the C5 level from the cytoarchitecture map). Regions corresponding to the optical imaging panel are circled on the AxCaliber maps: gracilis (yellow), cuneatus (blue), rubrospinal (green) and spinocerebellar (red).