3dMVM - graphical representation

Dear AFNI experts,

I ran the following group analysis after applying a TENT model to estimate brain activity:

3dMVM
-prefix groupAna_factor1
-wsVars ‘ABTime’
-wsMVT
-bsVars ‘factor1’
-qVars ‘factor1’
-qVarCenters ‘0’
-dataTable
Subj A B Time factor1 InputFile
S1 A1 B1 t1 0.1968 …/S1/SoA_S1_GAM.results/stats.SoA_S1_GAM_REML+tlrc’[SA#1_Coef]’
S2 A1 B1 t1 0.8258 …/S2/SoA_S2_GAM.results/stats.SoA_S2_GAM_REML+tlrc’[SA#1_Coef]’

To note: A and B have two levels (A1, A2; B1, B2), Time has four levels (t1, t2, t3, t4).

After this analysis, I have two cluster-corrected maps of interest: the interaction of Axfactor1 and Bxfactor1.
Now I need to represent in a scatter plot(s) the interactions within the significant clusters (for example: factor1 is related to brain activity in the significant cluster, but only for the first level of A -A1-, and not for the second level of A -A2-).

I was wondering - following the multivariate model - which is the correct way to extract data across time in order to represent such interaction. Do you think that timepoints (t1-t4) should be sipmly averaged? Any suggestion is appreciated.

Thanks in advance,
Simone

I ran the following group analysis after applying a TENT model to estimate brain activity

You captured the HDR with only four time points?

I have two cluster-corrected maps of interest: the interaction of Axfactor1 and Bxfactor1.

You may throw away a lot of informative results if you stick to such a dichotomous approach. Furthermore, sharp thresholding with the current correction approach can be overly conservative.

I need to represent in a scatter plot(s) the interactions within the significant clusters (for example:
factor1 is related to brain activity in the significant cluster, but only for the first level of A -A1-, and
not for the second level of A -A2-).

Extract at each cluster the effect estimate (beta) for each level of A (A1, A2) at each time point (t1, t2, …) (averaging across the B levels) from each subjects, and plot them against factor1.

Dear Gang,

Thanks!

You captured the HDR with only four time points?

The model was a TENTzero with 10 non-zero estimates, but I focused on specific timepoints given the group analysis results. Do you think this is uncorrect / introducing bias?

(ME) I have two cluster-corrected maps of interest: the interaction of Axfactor1 and Bxfactor1.
(Gang Chen) You may throw away a lot of informative results if you stick to such a dichotomous approach. Furthermore, sharp thresholding with the current correction approach can be overly conservative.

I don’t think I completely understood this point. If you mean that I also have to check the other results, of course I did (n.s. after cluster correction). Otherwise, I’m not sure how to change my approach to avoid being overly conservative…

Extract at each cluster the effect estimate (beta) for each level of A (A1, A2) at each time point (t1, t2, …) (averaging across the B levels) from each subjects, and plot them against factor1.

Clear! Thanks!

Simone