If you are confident that the connection probabilities have converged to a stable value, then it is fine to lower the number of samples per seed point. You can try with a few values and a few seed regions to see how much you can lower it.
Note also that you can run probtrackx in parallel using different values for --rseed.
On 14 Jun 2012, at 22:54, András Jakab wrote:
> Dear experts,
> I am running whole-cortex tractographies with 150 seed regions and 150 atlas-based targets for each seed region (not using the new --omatrix options). I intend to create connectivity (symmetrical, 150*150) matrices where each element is filled up by summing relevant values of the "seeds_to_target" images and then correct it by the seed region's waytotals.
> My question is rather theoretical. You suggest that -P 5000 is optimal for probtrackx. However, I "sum" connectivity information over a mask of 1-2 ~hundred voxels to have a single variable of connection "strength" for the (i,j) of the matrix. In this case, can the usage of lower -P values be justified? Although I use a large HPC of 2200 cpus, this remains still very time consuming as I plan to do this for ~200 subjects.
> p.s. the latest patch for probtrackx2 really reduced the memory for this task, thanks!
> Andras Jakab, M.D. Ph.D.
> Post Doctoral Researcher
> Dept. of BioMedical Laboratory and Imaging Science
> Faculty of Medicine, University of Debrecen
> 98. Nagyerdei Krt.
> 4032 Debrecen
> Phone: +3652255170
> Cellular: +36305095083
University of Oxford, FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
(+44)1865-222466 (fax 717)