FATCAT: Missing files 3dTrackID

Hi there,

I was taking a look at my 3dtrackID output when I noticed that there is a file missing. I have 5 ROI’s in total, however I believe there should be a file titled ROI_001_002 and that does not exist.

I’m not sure what the reasoning for that is exactly. I’ve attached the 3dtrackid output files if you’d like to take a look at them.
I’ve attached the ROI’s. ROI1 is the yellow ROI and ROI2 is the purple ROI2.

Any idea what could be happening here?



Hi, Sondos-

I think that isn’t an error or anything, but a sign that there was no direct connection found between ROI1 and ROI2; there were apparently connections found between pretty much all of the other possible pairs in that case (a large amount of interconnectedness which I haven’t really seen before, but can be OK in principle).

If you look at the *.grid files, my guess is then that in the “NT” (=“number of tracts”) matrix will have a 0 in the ROI1 column and ROI2 row (and equivalently in the ROI1 row and ROI2 column).


Yes, exactly. They showed up as 0’s (zeros) in the grid file

Yep, that’s normal. See, for example, from pg 32 onward here describing more about the *.grid files:
But, all the targets in a network are not necessarily connected directly to all other targets (that is the case biologically, and also for tracking, possible for related or different reasons); in general (and esp. for larger networks), that is not the case, and the “zeros” in the NT matrix (and also appearing in others) show where tracts were not found.


Hi again,

Now that I checked the grid files again, there are actually ROI’s missing and they don’t show up.
For example if its 5 ROI’s, ROI 1 might not show up at all, and this can be computationally complicated.
However, in the case of 3dNetCorr all ROI’s are present and have values in the matrix.
So does that mean that there is a funcitonal correlation and no structural correlation?

Also does that mean that there connections there don’t exist(NaN)? or does it mean that there is connection values of 0?

Also, does that possibly mean that my ROI’s don’t inflate to the white matter, therefore we are not getting any values?
Because for my data what I ended up doing is not using the inflated ROI’s (GMI) because my ROI’s are already quite large( as you have seen in the photo in my question initially), was that a mistake on my part to not use the inflated ROIs?
Or could I possibly get missing tracts in all cases even if I used the inflated ROIs?

Sorry for bombarding you with a million questions.

Thanks for your time

Hi, Sondos-

When you say that “ROIs are missing”, does that refer to the target ROIs or to the WM-connection ROIs? In general, there will be target ROIs without direct connections between them – that is true physiologically as well as when running tractography.

Please see:

  • Taylor et al 2015: Brain Connectivity Vol. 6, No. 2, “Open Environment for Multimodal Interactive Connectivity Visualization and Analysis” for more about this, specifically the 2nd column of pg 116, in the paragraph starting with “The general implementation …”.
  • THe AFNI Bootcamp presentation: FATCAT_02_dti_tracking_intro.pdf, pg. 37 and the few following it mentioning sparsity
  • The first 11 slides of FATCAT_03_dti_tracking_funcs.pdf, which ends by mentioning specific tools for dealing with this when doing post-tractography group analysis
  • and perhaps most importantly, the FAT-MVM Demo (downloadable with “@Install_FATMVM_DEMO” which has a detailed readme file and scripts for looking at this matter when preparing to run 3dMVM.

—> Not finding connections tractographically doesn’t mean they necessarily don’t exist. In brief, it is very difficult to interpret them. By default, the MVM-based analysis in AFNI finds just the WM-connections that were found across all subjects and uses those.

Re. inflating target ROIs to white matter-- you can/should look at that in those data sets. Hopefully your target ROIs are close/mostly bordering WM already, before inflation; if they aren’t, how were they created? Inflation should be a small effect, in general.

<< Also, does that possibly mean that my ROI’s don’t inflate to the white matter, therefore we are not getting any values? >>
It is hard for me to tell the exact location of those regions in the initial post, because there is no accompanying WM (FA>0.2) image…

You can make a WM skeleton map (FA>0.2) with something like this :

3dcalc -a FA_VOLUME -expr '(a-0.2)' -prefix wm_skel_FA02.nii.gz

… then underlay that beneath your current ROIs and check how/if they border it.