Longitudinal MBA

AFNI Message Board
1

Dear AFNI Gurus,

I have rsfMRI at three time points (baseline [Base], post intervention1 [PI1], post intervention2 [PI2]), and I'm interested if the intervention resulted in changes to the default mode network.

I would like to use the MBA approach and have two questions. First is with respect to the input into MBA: I have computed a correlation matrix via 3dTcorr1D using the errts.fanaticor file and a seed (posterior cingulate cortex; PCC) and have 11 other coordinate-based ROIs. Do I simply extract the PCC-ROI correlations, yielding an R-to-Z transformed dataframe

Subj Sess ROI1 ROI2 Y
s1 Base PCC ACC 0.123
s1 Base PCC LHC 0.456
..
s50 PI2 PCC LHC 0.789

or, as I'm not exactly interested in PCC correlations but the DMN dynamics, can I derive values with another approach (e.g. 3dNetCorr)?

Second, to investigate the effect of the intervention, is it possible to model the within-subject factor (Y~1+Sess+(1+Sess|Subj)+(1+Sess|ROI1:ROI2)) or would running two difference models suffice (PI1-Base, PI2-Base; deltaY = YPI1 - YBase; deltaY ~ 1+(1|Subj)+(1|ROI1:ROI2))?

Separately, it seems that the ability to test the association with a behavioral measure is only supported by BayesianGroupAna (but it appears that RBA might support this now)?

2

If you're only interested in a specific set of ROI pairs, I recommend using RBA with a command like this:

RBA -prefix ... -chains ... -iterations ... \
    -mean 'Y ~ 1 + Sess + (1 + Sess | Subj) + (1 + Sess | ROIpair)' \
    -ROI 'ROIpair' -EOI 'Sess' -WCP ... -dataTable ...

This setup above assumes that your focus is on the session effect (Sess).

You may also try the following command with the session difference data:

RBA -prefix ... -chains ... -iterations ... \
    -mean 'deltaY ~ 1 + (1 | Subj) + (1 | ROIpair)' \
    -ROI 'ROIpair' -EOI 'Intercept' -WCP ... -dataTable ...

Gang Chen