Biomedical Imaging and Instrumentation
Sameer Faruquee
Undergraduate Student
Northwestern Univeristy
Evanston, Illinois, United States
Kimberly Hemmerling
Ph. D. Candidate
Northwestern University, United States
Molly Bright
Assistant Professor
Northwestern University, United States
Spinal cord MRI scans were collected for 28 healthy subjects using a 3T Siemens Prisma MRI scanner. Two 10-minute fMRI (TE/TR=30/2000ms, 1x1x3 mm3) scans covering approximately the C5-C8 region were acquired during which subjects squeezed hand-grips, targeting 25% maximum voluntary contraction. Sixteen smoothing kernels were assessed, defined by their in-plane (i.e., transverse plane) and through-plane (i.e., along the superior-inferior axis) size. Kernel dimensions ranged from zero to three times the voxel size in each dimension, and all combinations of through-plane and in-plane smoothing defined a matrix of tested kernels (Figure 1A). Smoothing was applied within a motion-corrected mask of the cord excluding cerebrospinal fluid (CSF) to avoid noise[2][5]. Subject-level analysis was performed with right-grip and left-grip task regressors and nuisance regressors (motion, CSF, CO2, respiratory and cardiac cycles). Output parameter estimate maps were averaged across runs and group-level analysis was performed using a 1-sample t-test using threshold-free-cluster-enhancement and thresholded (p< 0.05, corrected for multiple comparisons). Significant voxels were counted in the entire cord, in each spinal segment, hemicord, horn, and in the gray matter for each kernel. The percentage of significant voxels contained in gray matter (%GM) was calculated as the proportion of significant gray matter voxels to total significant voxels. Laterality index (LI) was calculated for significant voxels in the ipsilateral and contralateral hemicords (LI=I-C/(I+C)), ranging from +1 (entirely ipsilateral) to -1 (entirely contralateral). Similarly, Ventral-Dorsal Index (VDI) ranged from +1 (entirely ventral) and -1 (entirely dorsal).
Group-level t-statistics maps for left and right grip activation for each kernel were thresholded to assess significance (Figure 1B). When kernel dimensions increased, clusters of significant voxels were accentuated. Heatmaps summarizing significant voxel counts for left-grip were generated.
Total significant voxels always increased with increased in-plane or through-plane smoothing (Figure 1C). Extending through-plane smoothing by 3mm and in-plane smoothing by 1mm led to similar increases in significant voxel count, indicating that activation sensitvity may be related to kernel size as a scaling of the voxel size in a given dimension.
%GM is highest with high through-plane smoothing and low in-plane smoothing (Figure 1D), which is consistent with current literature. LI was always greater than 0, indicating that the ipsilateral cord contained most activation (Figure 1E). LI was inversely related to through-plane and directly related to in-plane smoothing, however, this is not strictly true, as LI was highest for 3x3x3 mm3 and not 3x3x0 mm3. VDI was always greater than 0, indicating more ventral activation (Figure 1F). VDI was directly related to through-plane smoothing while being indirectly related to in-plane smoothing. The cord’s curvature may affect the effect of smoothing on VDI, but future work is needed here.
Most activation occured in the C6 to C7 segments of the cord (Figure 1G). Additionally, through-plane smoothing added significant voxels within each segment, whereas in-plane smoothing seems to preferentially increase significant voxels in segments that already contained more active voxels. Therefore, it appears that in-plane smoothing improves specificity of activation to spinal cord segments, and the opposite effect occurs for through-plane smoothing.
In conclusion, we recommend selecting kernel dimensions based on the research question. If the aim is to study activation in gray matter or between ventral/dorsal regions, higher through-plane smoothing may be advantageous. On the other hand, if the aim is to study activity between right/left hemicords or spinal cord segments, higher in-plane smoothing may be advantageous. In conclusion, we recommend considering the potential benefits of smoothing in-plane and through-plane with respect to the study objectives when selecting dimensions for spatial smoothing in spinal cord fMRI.