Propose a new method to automatically assign visually distinct colors to presegmented bundles of streamlines. This is useful to researchers and neuroscientists, to ensure an easy distinction between close bundles, and ensure color consistency throughout a study.Various tools and methods^{1, 2} now allow to automatically extract white matter (WM) bundles from whole brain tractograms. To reduce manual interaction, most tools automatically assign a color to each bundle. No matter the strategy chosen for this assignment, there never is a consideration of spatial relationship between bundles, contrarily to, for example, brain surface coloring³. This frequently results in neighboring bundles sharing almost exactly the same color, making them difficult to distinguish visually (see Figure 1). This can be quite limiting when trying to visualize those bundles.

To overcome this problem, we propose the following method: 1) Compute the bounding box of bundles, to have information on their spatial reaches; 2) Create a graph representing spatial relationships, having the bundles as vertices and connecting bundles having intersecting bounding boxes (see Figure 2 for an example); 3) Generate a colormap where each sequential color is the farthest from all previously chosen colors in a perceptually uniform color space (CieLAB⁴); 4) Beginning with the most connected vertex, assign a color from the colormap to each bundle; 5) Find N, the number of bundles that have a color similar to one of their neighbors; 6) Generate N new colors following the same principles as step 3; 7) For each of the N bundles, iteratively choose, in the new colormap, the color that maximizes the distance to its neighbors' colors.

To validate that our method produces a color assignment which facilitates distinguishing neighboring bundles, we used a Scilpy⁵ production pipeline to create a 500k streamlines, whole-brain tractogram. Dipy⁶ CSD modeling was ran on a 64 directions, b=1000 s/mm² DWI. Deterministic tractography was done through Mrtrix⁷, seeding from the WM mask. 29 of the main WM bundles were extracted using the TractQuerier¹ with a custom query.

Bundles were then divided in various configurations: 1) all bundles, 2) Corpus Callosum (CC) bundles, 3) all bundles except those from the CC. Those groups of bundles were then colored using 2 techniques: 1) a basic technique⁸ based on sampling the HSV color space using a golden ratio based spacing, sequentially assigning a color to each bundle, 2) our proposed technique. Each bundle group / coloring technique combination was then visually analyzed to check how often 2 neighboring bundles were indistinguishable.

Table 1 shows the occurrence of visually indistinguishable bundles for the various bundles groups and coloring technique combinations. Using the basic technique on the full 29 bundles resulted in 13 pairs of bundles that were not separable. Our technique generated no such pair. This is clearly a gain, since it makes it easier for users to see the full content of their extracted bundles, with no manual intervention to assign new colors to bundles. Figure 3 shows bundles extracted in the right hemisphere and their coloring, whereas Figure 4 shows a subset of those bundles demonstrating the problem of distinguishing bundles. Using the basic technique, each of the 2 pairs of bundles has only one color for the 2 distinct bundles of the pair. Our technique assigned easily distinguishable colors to those same 4 bundles.Using our proposed technique enhances the distinction between neighboring bundles, to more efficiently work with those datasets. An additional advantage of the technique is that no manual intervention is needed. Colors are automatically optimized based on the spatial relations of bundles.We have shown that using our automated color assignment technique results in bundle color assignments that enhance the ease of visual identification of bundles, by assigning sharply different colors to neighboring bundles. This will help users developing automated processing pipelines to obtain consistent and more usable bundle datasets. It will also help neuroscientists in quickly identifying missing bundles when some colors are not present.