Improved in vivo diffusion tensor imaging of human cervical spinal cord

Abstract

We describe a cardiac gated high in-plane resolution axial human cervical spinal cord diffusion tensor imaging (DTI) protocol. Multiple steps were taken to optimize both image acquisition and image processing. The former includes slice-by-slice cardiac triggering and individually tiltable slices. The latter includes (i) iterative 2D retrospective motion correction, (ii) image intensity outlier detection to minimize the influence of physiological noise, (iii) a non-linear DTI estimation procedure incorporating non-negative eigenvalue priors, and (iv) tract-specific region-of-interest (ROI) identification based on an objective geometry reference. Using these strategies in combination, radial diffusivity (λ(⊥)) was reproducibly measured in white matter (WM) tracts (adjusted mean [95% confidence interval]=0.25 [0.22, 0.29] μm(2)/ms), lower than previously reported λ(⊥) values in the in vivo human spinal cord DTI literature. Radial diffusivity and fractional anisotropy (FA) measured in WM varied from rostral to caudal as did mean translational motion, likely reflecting respiratory motion effect. Given the considerable sensitivity of DTI measurements to motion artifact, we believe outlier detection is indispensable in spinal cord diffusion imaging. We also recommend using a mixed-effects model to account for systematic measurement bias depending on cord segment.

Fig. 1

Reduced FOV diffusion sequence scheme (A) with gradient moment nulling (both 0th and 1st orders) in the slice-selective direction. Grey areas indicate matched gradient moments. Fat suppression was achieved with a frequency selective fat saturation module before the excitation pulse and gradient reversal between the two refocusing/inversion pulses. (B) Slice position with separate slice groups and tilted slices within the group. White arrows are landmarks for slice positioning. Green boxes (i.e., shimming volume) restrict the spatial region of B₀ homogeneity optimization to a relatively homogeneous tissue (i.e., spinal cord). (C) Slice position and shimming volume in the coronal view. This arrangement accommodated the natural curvature of the human spine and enabled compensating for positioning differences in the same subject on different visits. Furthermore, slices were individually tilted within the slice group to be parallel to the cross-section of the cord. Figure adapted from (Klawiter et al., 2012; Xu et al., 2010) with modification.

Fig. 2

Iterative registration scheme tailored for rFOV spinal cord diffusion imaging. DW images were mutually coregistered within alignment groups. (A) Sagittal b₀ image at original position. (B) Registered and averaged sagittal b₀ image. (C) Axial b₀ image (shown in entire in-plane FOV) at original position, overlaid with the registration mask (transparent red-orange). (D) Axial DW image (shown in entire in-plane FOV) at original position. (E) Sagittal DW image at original position. (F) Intermediate group registered and averaged sagittal DW image. (G) Final group registered and averaged sagittal DW image. (H) Final group registered and averaged axial DW image (cropped), overlaid with the outlier rejection mask (yellow). See Methods for details. The example shows one extreme case of severe motion disrupting the slice contiguity.

Fig. 3

Example of outlier rejection procedure for robust DTI fitting. See Methods section for details about the µ_res plotted here and the outlier rejection routine. Box-plots were used to visualize the distribution of µ_res. Solid points represent data for one diffusion encoding direction in six different slices. Only one of these (slice 8, solid point) was considered to be an outlier and removed from DTI fitting. The corresponding DW image and squared residual maps are shown for each slice below the Box-plots. Signal drop-out and, consequently, high residuals are evident for the outlier. The procedure was repeated until all outliers were removed. During each iteration, no more than one outlier (with the largest µ_res) was identified for each slice; which encoding was removed was not necessarily the same across slices.

Fig. 4

Anatomical reference geometry (A) based ROI definition on a FA color maps of a representative slice from C4 (B), for lateral CST (A, pink), PC (A, yellow), and GM (C, blue). The same method was used for generating the ROIs at all cervical spinal cord levels in this study. Figure adapted from (Klawiter et al., 2012; Xu et al., 2010) with modification.

Fig. 5

Self-correlation plots of λ_‖ (A), λ_⊥ (B), MD (C), and FA (D) with (x-axis) and without (y-axis) outlier rejection show significant overestimations of diffusivity measurements and underestimations of FA when outlier rejection was not applied. The black lines indicate line-of-identity. Ordinary linear regression lines were plotted in blue with 95% confidence intervals (shaded). Color represents left or right PC (LPC or RPC), and left or right lateral CST (LCST or RCST). The outlier rejection procedure did not bias any particular WM ROI. If the outlier rejection had not had any effect on the results, points would have been randomly distributed around the line-of-identity, due to the non-deterministic nature of diffusion tensor estimation with the non-negative eigenvalue Bayesian prior (Appendix I).

Fig. 6

Representative DTI maps from the center slice in each cord segment (C1–C6) from a normal subject. The distinctive “butterfly” differentiation of GM and WM can be appreciated from FA, λ_‖, and λ_⊥ maps.

Fig. 7

Distributions of DTI parameters in left-right averaged lateral corticospinal tracts (CST, A), left-right averaged posterior columns (PC, B), left-right averaged graymatter (GM, C), and whole slice ROI (Whole, D) through cord segments (C1–C6) for normal subjects (n = 18). Each tick mark on x-axis represents one slice. Error bars represent standard deviation among the population. The first slice in C1 (shadowed) was removed from other analyses to avoid the decussation of the lateral CST at the medullary–cervical cord junction.