Hi,
If you really want to ignore volumes within an analysis then you should avoid deletion and use a separate confound regressor for _each_ volume that you want deleted. So if you wanted to delete volumes 2 and 3 then you'd need two regressors:
[0 1 0 0 0 0 0 0 0 0]
and
[0 0 1 0 0 0 0 0 0 0]
I don't quite know why you want to delete half of your volumes, but in principle you can extend this, though obviously you create a lot of regressors that way. However, this is the right way to accomplish removing volumes without causing problems in the pre-processing or autocorrelation estimation. Also, make sure that for these EVs you keep the temporal filtering turned on, as otherwise they don't capture all of the effect associated with the original unfiltered timepoint (which spreads on filtering).
All the best,
Mark
On 18 Sep 2013, at 14:43, Tynan Stevens <[log in to unmask]> wrote:
> Thanks for the reply.
>
> That is basically what I was worried about; altering the autocorrelations. I am not sure the confound regressor route is what I want either though, as I am familiar with using that when I expect some signal to be explained by my confound. In this case, for a hypothetical 10-volume fMRI run, I want something like this:
> confound1=[0 1 1 0 1 0 1 0 0 1 ]
> So that where there are 0's, the data for those timepoints are effectively ignored in fitting the model parameters / calculating residuals, etc. I would then be able to run a second analysis with:
> confound2=[1 0 0 1 0 1 0 1 1 0 ]
> To independently analyze the other half of the data. This would allow you to use the un-altered 4D fMRI file and design files that preserve auto-correlations.
>
> It is not clear to me that the confound regressors in feat would accomplish this?
>
> -Tynan
>
> HI - I'm not sure it makes sense to do "split-half" analysis on FMRI timeseries modelling in this way - hard to avoid messing up the temporal autocorrelation in the data/model. It is possible to tell FEAT to "ignore" specific volumes just by adding in appropriate confound regressors - but you will end up with a vast model if you want to add a separate regressor for every time point ignored.
>
> Cheers.
>
>
> On 16 Sep 2013, at 17:34, Tynan Stevens <[log in to unmask]> wrote:
>
> Hello,
>
> I am trying to figure out what the best way to handle volume deletion for a FEAT analysis is.
>
> My goal is to be able to do a split-halves analysis, by randomly assigning volumes in my original dataset to 1 of two subsets, and then analyzing these. I know I could do this by removing volumes from my time-series manually, and likewise from my response model, but I feel like this isn't very sophisticated, as it would alter the signal dynamics.
>
> In AFNIs 3dDeconvolve I know there is an option to provide a list of 1s/0s to include/exclude volumes from the model fit. Is something like this possible in FSL?
>
> -Tynan
|