Hi Nicolas,
> - My merged_ph and th 4d images consist of 3d nodif_brain mask multiplied by a number ranging from 0 to 2pi and merged together. So all voxels from a 3d volume have the same value, but the distribution of values for a voxel across time ranges from 0-2pi.
My understanding is that you generate a random set of (theta,phi) pairs and assign this same set to all voxels in the brain. Is there any reason for doing that rather than assigning a different random set to each voxel? If the number of direction samples (i.e. time points) you generate is not very large, picking at each voxel from the same direction set may reduce the "randomness" of the tracking results, which is what you are after for this application.
> - I am not sure what the merged_f values should be, particularly if you are not constraining the tractography to FA values, so I presumed this would not make any difference (I've used all voxels with FA of 0.5 or 1 without much changes).
You are right, if you do not use anisotropy to constrain tractography the f_values should not influence your results.
> And yet with this approach the diffusion process seems quite limited to the seed region, with almost no tracts progressing further in the brain as you would expect. Probably there is something very wrong that I am doing, and was hoping you'd be able to advise me on this.
Using uninformative orientation pdfs, will "slow down" tracking, as there is no directional coherence. Therefore, if you increase the step length and/or the number of steps in probtrackx might help. Furthermore, you can reduce the curvature threshold and remove loopchecking as many streamlines will be excluded because of these criteria.
Hope that helps,
Stam
|