Group analysis and clustering using surfaces

AFNI Message Board
1

Dear AFNI experts:

Is it possible to run AFNI group analysis related functions (3dttest++, 3dANOVA, 3dLMEr) using surfaces instead of volumetric data?

If so, I am interested to do a multiple regression with the following variables:

  • Dependent variable = the surface.
  • Independent variables/regressors = 7 physiological measures (3 main effects and 4 interaction effects).
  • 4 covariates.

Could 3dLMEr may be a good approach to handle that regression?

Finally, I am interested to performed permutations and apply FWEr correction. I knew that ETAC could do something like that in t-test analysis. But, Is it possible to do something similar as ETAC does in 3dLMEr’s output (also considering that I am using surfaces as dependent variable)?

If you could help me with my queries I would greatly appreciate it

Best regards,
Karel

2

Karel,

Yes, you can perform population-level analysis on surface in AFNI. Could you describe the types of those explanatory variables (including covariates)? Are they between-subject or within-subject?

Is it possible to do something similar as ETAC does in 3dLMEr’s output (also considering that I am using surfaces as dependent variable)?

3dLMEr is based on linear mixed-effects modeling at the voxel level, and does not offer a way to adjust for multiplicity.

3

Gang,

Could you describe the types of those explanatory variables (including covariates)? Are they between-subject or within-subject?

We will perform two analyses. The first one include 7 within variables of interest (3 main effects and the 4 possible interactions) , 2 covariates of no interest (within) and 2 factors of no interest (between). The second one will include 2 variables of interest (within), one between and the 4 possible interactions (1 within and 3 between), two covariates of no interest (within) and 2 factors of no interest (between).

3dLMEr is based on linear mixed-effects modeling at the voxel level, and does not offer a way to adjust for multiplicity.

In that case, which approach could offer a way to adjust for multiplicity?

Thank you so much,
Karel

4

Karel,

If you have within-subject quantitative variables, use 3dLME or 3dLMEr. Otherwise, go with 3dMVM.

which approach could offer a way to adjust for multiplicity?

I may set a voxel-wise p-value of 0.05. If I see some clusters with, for example,10 or more voxels, that would be some useful information for me.