Aaron H wrote:

> Hi AFNIers,
>
> This is what I want to do: I have run 3dDeconvolve on an event
> related paradigm and I want to view the fit of the model to the
> event-related data.
>
> Now, I know that the AFNI viewer has the capability to show the
> fit over the entire time series - i.e. by loading the raw data
> as the underlay - and loading the output of -fitt. But what I
> want to do is to view the averaged, event-related data and it's
> corresponding fit. For example, every time a negative stimulus
> is presented I want to be able to view the averaged, raw BOLD
> response (baseline corrected or not) to that stimulus along
> with the corresponding fit/error (also averaged for that
> condition, of course). As far as I can tell, Gang's page on
> viewing model fit only looks at the entire session - not event
> related.
Well, if it is not a SLOW event related design, there is no raw BOLD response to get this way. When event responses overlap, we use some model to disentangle them. The most widely used model is a linear one. So what you might want to see is the estimated impulse response function to each stimulus category. This IRF and its standard deviation would be output with the -iresp and -sresp options of 3dDeconvolve. Obviously, you should be using an appropriate set of basis functions for that type of analysis. I would recommend you use the 'TENT' for this type of analysis. If you are not familiar with these concepts, you can find documentation here:
[afni.nimh.nih.gov]
under AFNI Regression, AFNI Deconvolution and Advanced 3dDeconvolve


>
> What would be super is if I could also dynamically click around
> the brain to look at the averaged BOLD response/fit/error
> across separate regions. It would also be nice to alter the
> length of the window (number of TRs) about which I am viewing
> the averaged time-series.
Interactive Clusterizing would do the trick.
[afni.nimh.nih.gov]

>
> I know VoxBo has had a very nice capacity to do this.
>
> Does AFNI do this natively? If not, has anyone written code to
> do this? One option, of course, would be to parse the output of
> fitts and errts based upon the timing of the paradigm, but I
> would have to do some interpolation because the timing was in
> seconds, while TR = 2.0sec.

>
> I haven't seen this discussed in the threads, so if it has been
> discussed earlier I'd be happy to take that suggestion
>
> Thanks.
>
> -aaron