Surface-based tracking for short association fibre tractography

Abstract

It is estimated that in the human brain, short association fibres (SAF) represent more than half of the total white matter volume and their involvement has been implicated in a range of neurological and psychiatric conditions. This population of fibres, however, remains relatively understudied in the neuroimaging literature. Some of the challenges pertinent to the mapping of SAF include their variable anatomical course and proximity to the cortical mantle, leading to partial volume effects and potentially affecting streamline trajectory estimation. This work considers the impact of seeding and filtering strategies and choice of scanner, acquisition, data resampling to propose a whole-brain, surface-based short (≤30-40 mm) SAF tractography approach. The framework is shown to produce longer streamlines with a predilection for connecting gyri as well as high cortical coverage. We further demonstrate that certain areas of subcortical white matter become disproportionally underrepresented in diffusion-weighted MRI data with lower angular and spatial resolution and weaker diffusion weighting; however, collecting data with stronger gradients than are usually available clinically has minimal impact, making our framework translatable to data collected on commonly available hardware. Finally, the tractograms are examined using voxel- and surface-based measures of consistency, demonstrating moderate reliability, low repeatability and high between-subject variability, urging caution when streamline count-based analyses of SAF are performed.

Keywords: Short association fibersl; Superficial white matter; Surface; Tractography; U-fibers.

Graphical abstract

Fig. 1

Pipeline summary. After seeding from white surface mesh (WSM) coordinates (top row, 1-3), tractograms are filtered (top row, 4) to ensure each streamline starts and ends in the neocortex (grey-grey filter) of the same hemisphere (hemisphere-hemisphere filter) and escapes into white matter along the way (grey-white-grey filter). The grey-grey filter (bottom row, 1-3) functions by finding the closest midcortical coordinate (MCC, average of the matching WSM and pial coordinates) for each streamline end (with K-means clustering of MCCs for speed - bottom row, 1). A streamline end is considered in the grey matter if it lays within the local cortical half-thickness of its MCC (bottom row, 3). Then, two intersections with WSM (one either end) are sought (bottom row, 4) at which point the intracortical portion is truncated. The optional surface-based analysis is conducted after the filtering (top row, 5).

Fig. 2

Comparison of surface- and voxel-based strategies. A: Distribution of seeding coordinates (yellow) related to a section of the WSM (green) in the paramedian premotor cortex (region-defining box in the middle). With surface seeding, most (but not all) vertices resulted in streamlines, visually achieving a more spatially uniform distribution. B: Final tractograms for the same subject showing similar appearances except for minimal streamline extension in a few areas into the deep grey matter with the surface method. Bottom, upper row: Histograms of streamline length (averaged across cohort) show a near-linear decline with the surface method and roughly a power law decline with the voxel method. Illustration in the centre shows a streamline coursing tangential to the grey-white interface (semi-transparent). Most sections are subcortical (blue) but the few escapes into the grey matter (red) would have met the conventional termination criteria resulting in a considerably shorter streamline. Bottom, lower row: Plots of streamline angle at the cortex along the gyral blade (averaged across cohort) show less acute angle with the surface method. Also see legend for Figure 6. Connectivity matrices in the centre show the number of streamlines connecting the five zones along the gyral blade (averaged across cohort) produced with the two methods. Colour represents streamline counts.

Fig. 3

Comparison of fODFs (l_max = 8, scale = 3) generated using Connectom (top row) and Prisma (bottom row), “state-of-the-art” (left column) and ”standard” (right column) acquisitions for the same subject on a 1 mm isotropic voxel grid, affinely co-registered to the same T1-weighted volume. Inset in the bottom right shows the region compared. Green outline: grey-white interface, red outline: 1 mm superficial, blue outline: 1 mm deep. State-of-the-art acquisitions (particularly from the Connectom scanner) demonstrate increased anisotropy along the gyral wall with glyphs pointing radially (yellow arrowheads) or turning more sharply (purple arrowhead). Colour of the glyphs represents orientation: SI, superior-inferior; AP, anterior-posterior; RL, right-left.

Fig. 5

Appearances of SAF tractograms depending on the choice of scanner, acquisition, resampling. A region from the right frontal lobe of the same subject is shown in coronal plane (inset in the left upper corner). Tractogram slices are 1 mm thick, all voxels are isotropic. Differences in streamline density are most noted along a sulcal wall (yellow arrow) and a gyral crown (red arrow). Local variation in the predominant direction of streamlines is also seen (green arrow, red arrow).

Fig. 6

Effect of scanner, acquisition, resampling on the streamline-cortex angle, depending on location along the gyral blade. Each plot shows average angle distributions across all subjects. Colour bars represent streamline counts. Mean angle (not adjusted for streamline count) is shown in each title as $\overline{\alpha }$; same is provided on the x-axis for the five equal-area sectors of the gyral blade. Grey vertical lines illustrate boundaries between the sectors. Green vertical lines represent 5–95% range used for average angle calculation. Most differences appear in the sulci, with Prisma scanner, “standard” sequence and larger voxels leading to increased mean angles.

Fig. 7

Effect of scanner, acquisition, and voxel size on SAF connectivity between different positions along gyral blades. Cortical surface was divided into five zones with equal areas from sulcal fundi to gyral crowns. Each image shows the absolute number of streamlines connecting different zones across all subjects. Connectivity matrices suggest that Connectom, ”state-of-the-art” sequence, small voxel size increased the overall number of connections but particularly so between gyral crowns and the surrounding areas.

Fig. 4

Effect of scanner, acquisition, resampling on the orientation of FODs (calculated for 1st peak only) in relation to the cortex, depending on the location along the gyral blade. Each plot shows average angle distributions across all subjects, sampled at three depths (green: grey-white interface, blue: 1 mm deep, red: 1 mm superficial). Mean angle is shown under each title (deep white/grey-white/cortex, respectively); same is provided on the x-axis for the five equal-area sectors of the gyral blade. Grey vertical lines illustrate boundaries between the five sections along the gyral blade. Green vertical lines represent 5–95% range used for average angle calculation. Connectom scanner, “state-of-the-art” sequences, smaller voxel data all demonstrate steeper peak rotation from near-parallel to near-perpendicular orientation (from deep white to cortex, respectively) particularly in the gyral wall.

Fig. 8

Final appearance of SAF tractogram following the filtering process. Left: SAF streamlines overlaid on the T1-weighted volume in dMRI space (coronal view, 1 mm thick slice). Right, first column: SAF bundles will often deviate from the orthogonal planes in their course. Thicker slices (5 mm) are provided for a better appreciation of their extent. Right, second column: TDI maps of the same regions are provided. Regions on the left are represented in the right-hand columns under matching letters.

Fig. 9

Repeatability of SAF using TDI map comparison in common space. All maps were superimposed on the average T1-weighted volume. CV_W and CV_B were thresholded at 100%. ICC was thresholded at <0.5 and >0.75. CV_W, coefficient of variation within subjects. CV_B, coefficient of variation between subjects. ICC, intraclass correlation coefficient.

Fig. 10

Surface-based analysis demonstrated on the lateral cortex of the right hemisphere. N_S, termination density (number of streamlines/vertex). L_S, mean streamline length/vertex. FA_S, mean streamline fractional anisotropy/vertex. CV_W, coefficient of variation within subjects. CV_B, coefficient of variation between subjects. ICC, intraclass correlation coefficient. Values at each vertex were recorded in subject space, then transformed into average subject space before running analyses. CV_W and CV_B were thresholded at 100%.