Dear Mina,
As you point out, it doesn't make much sense to use a parametric
modulation for conditions that consist of a single trial.
What are you actually trying to test for and where did the parametric
modulation come into play? Wouldn't an option be to estimate your
'beta-series' in the usual way and use these in a subsequent GLM where
the parametric modulators would correspond to extra covariates?
Best regards,
Guillaume.
On 25/11/2018 14:32, Mina Shire wrote:
> Dear SPM Experts,
>
> I would really appreciate if any of you could make a comment on my
> design matrix modeling that would help me to modulate the regressors by
> using some parameters.
>
> I would like to construct a beta-series connectivity analysis modeling
> each of the trials with a separate regressor in the design matrix as it
> was explained in Rissman et al (2004). However I would like to use
> parametric regressors for each of those trials, so here comes the
> problem. First of all due to this unique structure of the design matrix
> required for the beta-series analysis, using parametric modulators
> creates a concern with the orthogonality In addition because of the
> hierarchical modeling in SPM design matrix, the order of the parametric
> modulators matters resulting in a decrease in the effect/variation from
> the first to the last regressor. So I guess using parametric modulator
> in such design doesn't work at all. You can see the example of its
> application here.
>
> https://www.dropbox.com/s/4s0ncpjpug2ucg5/parametricBetaSeries.png?dl=0
>
>
> To overcome those problems, I used two different approaches. As first, I
> used parametric modulation in the beta series model proposed by Mumford
> et al (2012). So in which case, rather than modeling each trial as a
> separate regressor in your design matrix, you model a single design
> matrix for each of your trials separately and the rest of the events are
> modeled as the nuisance regressors. Hence you end up with N-number of
> design matrix for your N-number of trials. But if I apply parametric
> modulator in this beta-series modeling proposed by Mumford et al, again
> I guess I cannot overcome the orthogonality problem don't I? I would
> really appreciate if you could share your comments on this please. Here
> is the application if it :
>
> https://www.dropbox.com/s/g05jd82bb5ggtvo/parametricBetaSeries_SingleTrial.png?dl=0
>
>
>
> As a second solution, rather than using parametric modulator, I thought
> I could manipulate SPM beta estimation. So in which case I constructed
> my design matrix using standard stick functions (without parametric
> modulators) and then I multiplied each hrf-convolved regressor in
> SPM.xX.X (simply the design matrix) with the parameter of interest,
> considering this would modulate the hrf with the parameter. This allowed
> me to overcome the orthogonality problem, however, I am not sure whether
> modifying hrf in SPM would work in a way that I want to modulate each
> regressor and I also wonder whether if this manipulation would break
> down the data structure of SPM.mat or beta-value estimation pipeline of
> SPM at some point. Here is the example design matrix:
>
> https://www.dropbox.com/s/ngm14iefzp0sqdo/BetaSerieswithHRFModulation.png?dl=0
>
>
> So I am so sorry if the question is very simple and the answer is out
> there and I can't see for some reason, but I would really appreciate if
> anyone could share their recommendations or comments on how to establish
> a beta-series connectivity estimation in a way that I could modulate the
> hrf with some parameters of interest.
>
> Thank you very much for all the helps in advance,
>
> Kind regards
>
> Mina,
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|