To seed and target from the specific ROIs, you would need to use the --network mode with --omatrix1 and give a list of ROIs in a text file as seeds.
What you are calling below, generates a dense connectome, i.e. treats every non-zero voxel in the atlas as a seed/target. Therefore if you have N voxels in this atlas within the ROIs, you are asking for a N x N matrix, which can easily cause memory problems depending on how big N is.
> On 29 Aug 2017, at 15:55, Melanie Matyi <[log in to unmask]> wrote:
> Hi all,
> We are trying to use probtrackx2 to obtain a connectivity matrix of pairwise connections between 360 rois (using a single volume containing all 360 rois as the seed and target). From the FSL documentation, it seems like matrix3 was the option that would provide this. However, running the command below fails (Centos7, 40 GB RAM, FSL 5.0.10), even after reducing the number of samples to 100. The RAM appears to max out during the run, so this may be the cause of the failure. It does finish when only using 10 samples, but this takes about 11 hours and is a drastic reduction in samples (from the default 5000). Are we using the correct command? Any ideas why this might be failing and why so much RAM is being used?
> probtrackx2 -x roi_atlas.nii.gz -l --onewaycondition --omatrix3 --target3=roi_atlas.nii.gz --waypoints=wm_mask.nii.gz -c 0.2 -S 2000 --steplength=0.5 -P 5000 --fibthresh=0.01 --distthresh=0.0 --sampvox=0.0 --forcedir --opd -s merged -m nodif_brain_mask --dir=test