Hi all!
Wondering if it’s necessary to perform whole-brain tractography (using tckgen) prior to isolating tracts of interest (tckedit) in order to compare tract counts (or afdconnectivity) between subjects or time points. I have a dataset with several subjects who have undergone a neurosurgical lesioning procedure, and had baseline and then several followup scans. I’m interested in quantitatively measuring any changes in a few tracts of interest. My plan is to perform whole brain tracography (using tckgen), and then tckedit down to the tracts of interest, but wanted to confirm the whole-brain tractogram is necessary.
Hey @jdtournier, read the latest draft of that manuscript?
I have a manuscript describing precisely this problem and the various solutions… kind of an expansion of this documentation page on the afdconnectivity command. It’s become a self-deprecating meme at this point, because this question - along with how to properly compare quantities across subjects - comes up all the time on here, and yet it’s still not published… hence the persistent pings to @alan-connelly and @jdtournier
What I consider to be the most appropriate solution from a physics perspective is:
For within-subject quantification:
Whole-brain tractography
SIFT2
Extract pathway of interest
Sum SIFT2 weights of streamlines within the pathway of interest
For between-subject quantification:
Do requisite pre-processing for AFD, i.e. bias field correction, intensity normalisation, common response function
Multiply result of 1.4 by the SIFT proportionality coefficient; demonstrated here, explanation and proof of why this is appropriate to come.
@rsmith - just a followup. I’m planning to perform a within-subject comparison, and then move to between subject (with a linear mixed model). Can I still use the first method you outlined? And also, is there a command to perform step 4 (sum SIFT2 weights of streamlines with the pathway of interest).
Yes, those are not intended to be mutually exclusive: point 1 is the requisite steps to achieve within-subject quantification, point 2 are the additional steps necessary to achieve between-subject quantification.
Is there a command to perform step 4 (sum SIFT2 weights of streamlines with the pathway of interest).
Once you have extracted the pathway of interest, you should have a text file that contains the SIFT2-calculated weights of only those streamlines within your pathway of interest. So all you need to do is obtain the sum of a list of values. If you want a way to do such at the command-line, something like this would be one way to do it (though you’d want to verify it manually):
awk '{ sum += $1 } END { print sum}' pathway_weights.csv
Historically I’ve used: tr '\n' '+' | bc
to produce a long summation equation and feed it to a calculator, but the addition of comment lines to output text files in 3.0.0 means that won’t work without some additional trickery, so awk is simpler.
Hi @rsmith, thanks for the help. All of those steps worked well. I’m having some difficulties and I’m hoping to get your expertise.
The population I’m studying are patients who have undergone an anterior capsulotomy - a lesional procedure in the anterior limb of the internal capsule, essentially lesioning the tract connecting the thalamus to the orbitofrontal cortex. The lesions are small (~5-8mm diameter). I’m hoping to detect any changes in the connectivity between the thalamus and the orbitofrontal cortex between baseline and 6 month scans. When I ran the above steps, I got several subjects where the connectivity in fact increased (sometimes by 50-100%) from baseline to 6 months, although that does not make biological sense. I’m worried that it may have something to do with the normalisation step - could I be in fact ‘normalizing’ or smoothing out the differences I’m hoping to detect?
Wondering what you’d advise in this situation.
Thanks a lot!
Ben
While I can’t provide any insight into @BennyD’s particular cohort, I can at least point you in the direction of this preprint I pushed out subsequent to this thread, which includes within it the physically appropriate mechanism by which to normalise connection density estimates following SIFT2 across subjects. No guarantee that it will change results such as this, but it at least resolves the normalisation issue (albeit a half decade later than intended).
