Hi Dane,

Sorry I didn't respond earlier!

> 1a. Which subbrick or output file from 3ddeconvolve should I using as the input data?
> ( ie. The full MSE model, Fstat, All_Betas, fitts, stats?)

Not one specific sub-brick, but the same whole time series you used for individual subject analysis with 3dDeconvolve.

> 1b. Why does this input need to already be deconvolved if we are putting pretty much all
> the same motion and stim onset regressors into 3ddeconvolve in step 6? Seems like we
> are removing effects of no interest twice...

No, not really. When you create the interaction regressor, you want to construct an explanatory time series that purely reflects the interaction effect without any confounding effects, just as the same concept when you create the presumed (or idealized) hemodynamic response regressors.

> 2. Why do you need to detrend ROI data if you are masking it from 3dDecon output. Im pretty
> sure most preprocessing pipelines leading up to activation analysis with 3dDeconvolve already
> incorporate detrending.

The reason is that, in AFNI, the input time series still contain slow shift (long-range trends) and the associated effect is typically modeled through polynomials with option -polort. Therefore, after you extract the seed region times series, you need to remove any confounding effects including slow drift through detrending.

> 3a. Can you use -cbucket components of an original activation analysis for any of this, to skip
> complex use of waver and 3dtfitter (ie. Gang's steps 3 and 4.... or..... "the part that really flummoxed
> me")? For instance the "ideal_cond.1D" files multiplied by TR contrast -1/0/1 onsets be sufficient?
> Or is it still necessary to create the Impulse response GAM function with waver and other steps etc...

Yes, those steps are necessary in general. It's not clear to me what specific steps you are suggesting with option -cbucket.

Gang



Edited 1 time(s). Last edit at 09/08/2014 10:22AM by Gang.