Hi-
Re.
But, I’ve some question about doing mini-probabilistic tractography using 3dTrackID. Is there anyway to make tracts only in a certain direction?
There isn’t an option for affecting directionality of propagation. However, you can use ‘-thru_mask …’ to require tracts to go through a certain region to be kept, or alternatively use ROIs with negative values in the ‘-netrois …’ input to prohibit tracts from going through somewhere.
Re.
I want to draw tracts between OFC and caudate. but among the output tracts, I see some of those tracts start from caudate, make small loops around caudate, and end in caudate. I’m not sure this makes sense to you. It’s hard to explain without pictures 
Yes, there is difficulty posting images, but feel free to email me.
Re.
Anyway, this is not what I wanted since I want tract between OFC and caudate ONLY. So, I changed max. angle for turning when going to a new voxel during propagation to 55, but it is not helping. could you please give me any suggestion to solve this?
The issue of false positives are best dealt with using fully probabilistic tracking, via ‘-mode PROB’. Mini-prob is typically much better than just deterministic for visualization and reducing false negatives (because it includes tensor uncertainty information and some perturbed Monte Carlo iterations), but full probabilistic is better for consistent tracking.
As much as I like viewing (computated) tracts, their important function is to point the way of likely connections within hte WM skeleton-- each one isn’t really believeable/useful on its own. Full probabilistic uses a large number of perturbed iterations (again based on the data’s own tensor uncertainty) to build up maps of sets of voxels that keep getting hit by tracts-- and are therefore the most likely ones to be involved in a particular connection. This provides a better way of ‘trimming the fat’ of a few (well, hopefully the few) errant tracts on obviously wrong paths due to noise/randomness-- after a large number of iterations, you should be able to threshold those voxels away, which is how the fully probabilistic tracking works. (The thresholding within the miniprob tracking is, at the moment, just based on total bundle count, which isn’t as good-- but miniprob is still generally much superior to pure deterministic).
So, executive summary-- full probabilistic tracking! You might not have the pretty tract images, but you should get a more consistent set of WM voxels in a given connection. And… actually there should be a tool available soon for visualizing probabilistically tracked WM regions (voxel sets) in RGB determined by local tensor orientation-- that’s part of the OHBM poster I’m supposed to be finishing…
And, I’ve one more question. Some subjects are not able to build tracts very well. (only very short tracts are in the output). Is there any solution to solve this? or should I just take those people off from my set?
That sounds like a Randy Newman song… But, to the point at hand-- it is worth checking a whole brain tracking of subjects to make sure that the data doesn’t look too noisy/unreasonable overall, or that there is not a dreaded ‘flip problem’ with the gradients; flipping is described here:
https://afni.nimh.nih.gov/pub/dist/doc/htmldoc/FATCAT/DealingWithGrads.html#flipping-gradients-if-necessary
and there is a tool in AFNI called ‘@GradFlipTest’ to help guess if there is a flipping problem.
If that’s not the case, it could just be the data is very noisy/distorted. How did you process it? My favorite way tends to be TORTOISE. Visually checking the DWIs to make sure there aren’t volumes significantly affected by subject motion and/or dropout is useful, as well. I’m happy to take a look at any of the data sets, as well, if that’s useful. In the end, though, it is possible that some data might be too corrupted to use-- a judgment call then has to be made about loss of power vs skewness of results based on ‘rough’ data.
Re.
Following this the command I used.
I note that miniprobabilistic tracking works best by having only a few perturbed iterations-- that’s the ‘mini’ part of its name. This is described more here:
afni.nimh.nih.gov/pub/dist/papers/ASF_2015_draft_BCinpress.pdf
Basically, by adding a few perturbations, you help reduce false negatives by taking into account more possible pathways; however, there isn’t a superstrong way to filter out false positives, so you don’t want to have an infinite number of extra iterations-- something like 5-7 seemed to be a good balance (see Figs 2-3 in the paper; and actually, the Appendix there describes more about gradient flipping).
Therefore, if you want to stick with miniprob tracking, I’ve amended your command a bit to have a smaller ‘-mini_num …’ that seems more representative for the methodology:
3dTrackID \
-mode MINIP \
-dti_in DT -uncert tract_UNC+orig \
-alg_Thresh_FA 0.2 \
-alg_Thresh_ANG 55 \
-alg_Thresh_Len 20 \
-alg_Nseed_X 2 \
-alg_Nseed_Y 2 \
-alg_Nseed_Z 2 \
-logic AND \
-netrois ROI+orig. \
-prefix tract \
-mini_num 7
But, for most quantitative work, using full prob is still the way to go. And, you’re probably already doing enough iterations for that. Therefore, you could just do something like the following:
3dTrackID \
-mode PROB \
-dti_in DT -uncert tract_UNC+orig \
-alg_Thresh_FA 0.2 \
-alg_Thresh_ANG 55 \
-alg_Thresh_Len 20 \
-alg_Nseed_Vox 5 \
-alg_Nmonte 1000 \
-netrois ROI+orig. \
-prefix probtract \
-mini_num 7
–pt