Hi guys,

Sorry for the tardy reply on this. Just getting back to my desk after some time away.

To be concrete, I present the following.

We're working with a DTI-based index that estimates the amount of extra-cellular free water (FW) at each voxel in each participant. Our primary interest is in comparing depressed and healthy control samples with respect to this index (which is thought to be an indicator of neural inflammation). This index, however, can be influenced by a number of factors that have nothing to do with neural inflammation so we want to mathemagically address them through regression (in this case, 3dttest++ with covariate center = SAME). Note that we proceed with this kind of testing with the knowledge that regression isn't actually magic and that we can't "overcome" confounds between our depressed and control samples this (or any) way. Rather, if there are "weak" confounds between our MDD and CTL groups, we'd like to partial them out; further, we'd like to strengthen our tests by removing some noise associated with these covariates.

There are four noise covariates:
1. a single estimate per participant of rigid body rotation during acquisition of different DTI directions
2. a single estimate per participant of rigid body translation during acquisition of different DTI directions
3. age
4. a voxel-wise VBM-based estimate of structure volume

What I would like to have, for each subject, is a 3D map of "corrected" FW, based on having taken out effects (if any) of motion, age, and VBM.

Then I can mess around with these data that, presumably, are more reflective of actual FW in the brain (i.e., have linear effects of any associated noise covariates taken out). If the key data in doing this for, say, subj_01 are:
--subj_01_FW+tlrc (individual subject free water map)
--subj_01_VBM+tlrc (individual subject VBM map)
--subj_01's rotation, translation, and age
--and- from the 3dtest++ output, the voxelwise coefficients for VBM, age, rotation, and translation
then perhaps you could address two things:

1) there are a few kinds of coefficients for each noise covariate in the 3dttest++ output. For VBM, for example, we have MDD-CTL_VBM, MDD_VBM, and CTL_VBM (where MDD and CTL are our two groups). I assume that since we set center=SAME, that the coefficient associated with MDD-CTL is the appropriate one.

2) provide a sample 3dcalc command to generate the kind of map I'm looking for? I have a good sense for how to do this using a simple regression equation (like the one's you've provided) but I'd like to find out about Question 1, above, before I start messing around too much.

Thank you!

Paul



Edited 1 time(s). Last edit at 04/16/2015 02:02PM by paul.hamilton.