 |
 |
Hi Andries The easiest and most ram efficient way of doing what you want to do is to use --omatrix1. If I understand correctly, your seed is the whole cortex? In this case, what you could do is seed from the whole cortical surface (white matter/grey matter interface), define the pial surface as a stopping mask, and add the --omatrix1 flag. This will produce a matrix of size #seeds X #seeds, that stores the number of samples between each pairs of seed vertices on the cortical surface. You can then use this matrix to extract your ROI by ROI connections (e.g. in matlab). Cheers Saad On 6 Feb 2013, at 23:21, Andries R. Van Der Leij wrote: > Dear Sad and FSL experts, > > I've been struggling getting probtrackx connectome reconstruction to work. I am trying to figure out a pragmatic approach which will work on my cluster (24GB ram limit on a node). > > I started off in surface space, then went back to volumetric masks. Working with all the targetmasks still gives a too high ram load. > I figured I would just track from each seed in structural space only defining a single -stop termination mask. this binary mask consists of all the seed rois with the current mask set to 0, which leaves a 'hole' in the surface. Tracking is only constricted by the DTI mask and the termination mask. I figured I would just mask the resulting fdt_path files with all the ROIS post-hoc to read out the values of the overlapping voxels. This approach has the huge benefit that ram load <3GB, which would even allow for more jobs in parallel on a node (as opposed to a single job with the targetmasks since it would consume all the RAM). Before I submit a big batch of this to the grid, I would kindly like to have some feedback though. I have one remark, one general question and one specific question: > > Remark: In probtrackx2, this approach doesn't seem to work, as the tracks do not enter the termination mask at all which leaves little to mask afterwards. In probtrackx it works better. This has been discovered by another poster some days ago. > > Q1: does this approach make sense at all? > > Q2: If it does, I still think I need to correct the values for the size of the seed and the waytotal in a way. Are there rules of thumb for this correction? > > Thank you, > > > Andries van der Leij > > > > > > > -- Saad Jbabdi University of Oxford, FMRIB Centre JR Hospital, Headington, OX3 9DU, UK (+44)1865-222466 (fax 717) www.fmrib.ox.ac.uk/~saad