Since I was the one who decided on the threshold in the referenced paper, I might warn you that it was based on what worked well for those datasets to exclude low probability voxels that were not of interest while preserving higher probability voxels that were. I just completed arcuate tractography on another dataset, and got a rather different threshold value that seemed to work best 2.0e-6 % of total samples sent out from the ROIs. One difference between the two datasets is that my seed space in the below referenced paper was in diffusion space, whereas in the new study, it is in structural space. Thus, if the DTI resolution were 2x2x2mm and the structural resolution were 1x1x1mm, each voxel of DTI space would correspond to 8 voxels of structural space, and this may be the cause of the difference. Because probtrackx produces a continuous probability distribution, there is no “rule of thumb” that allows you to threshold results from any tract in any dataset at the same absolute number. I think two things are important: 1) That you use a consistent percentage of the total number of samples sent out for each tract across subjects (so that tracts created by larger ROIs, and thus have more total samples sent out, have higher thresholds) and 2) That in setting the percentage that you will use, you try a variety of values and see what seems to produce clean results showing the pathway of interest without many extraneous pathways not clearly connected to the ROI but at the same time does not remove large parts of the pathway of interest (because it is too high). Tim et al may have further thoughts. You can reduce the amount of extraneous pathways by carefully choosing your method of tractography (i.e. the ROIs you use) and using the colors and lines displays of FSLView to get an understanding of what the diffusion is doing in your pathways of interest and/or restricting your pathways to white matter by segmenting a structural image and turning greymatter and CSF into a “stop” mask. If you use fairly large ROIs and get pathways of few samples, that may suggest that any pathway between the regions is either very weak or does not exist.
Peace,
Matt.
This is a great paper to start with. Be sure to read the supplementary material for more info on the methods.
Rilling et al. Nature Neuroscience (2008) vol. 11 (4) pp. 426-428
On Jun 10, 2008, at 9:28 PM, Ruth Carper wrote:
I'd like to use ProbTracks to identify some white matter tracks. Do you
have any suggestions for selecting a threshold value to define a tract?
i.e. I have seed and target masks which produce an intervening tract, but
many of the voxels have very low values (indicating few streamlines). How
do I select a good cutoff value for what I can be confident is the real
tract? Any rule of thumb?
