Constructing connectome using FSL ProbtrackX

I would like to construct a connectivity matrix of the FreeSurfer Desikan-Killiany atlas using FSL’s Probtrackx2. I would like matrices showing ROI-to-ROI streamline count and mean path length. Using the example of this paper, I have gone ahead and set up a script that loops through each cortical ROI (--seed) and computes streamlines to every other ROI (--targetmasks). I was wondering if I could get some feedback on the current set up and see if there are specific options I should enable.

I am using all default options (i.e. streamline count of 5000, steplength of 0.5, curvature of 0.2, etc.), with -l and --pd flags enabled.
For outputs, I have specified --ompl, --os2t, --s2tasrext, and '–network`.

For waypoint masks, I have specified the left and right cerebral white matter.
For exclusion masks, I am combining the DK-atlas CSF and ventricle parcellations into a single mask.

Does this seem like an efficient setup? Is there a recommended way to make running this more efficient? Thank you for any clarification you can provide!

That seems fine. Without knowing your computational set up it is hard to know how to make it more efficient. For example, if you are on an HPC, you can have a script that produces multiple jobs, each doing different ROIs, to parallelize the process. You can also use the GPU accelerated probtrackx2/bedpostx versions if you have GPUs.

However, if you are okay using a different atlas, I will also recommend simply running your raw diffusion images through qsiprep and qsirecon. The end result is a set of connectivity matrices that include the metrics you want.

QSIprep and QSIrecon are pipelines we are currently still evaluating.
We already have BedPostX output for most individuals, and we’ve published previously using that output, so we need to use that in the meantime.

I unfortunately do not have access to our cluster just yet. I have access to a single server, which has equivalent hardware to a single cluster node. All I can think of at this point is to parallelize using xargs --max-procs=max-procs