Dear Carlo,
On Fri, Aug 17, 2012 at 11:16 AM, Carlo Huber <[log in to unmask]> wrote:
> Hej, Hej,
>
> I am working on a simple motor task and am trying to model connectivity patterns in core motor regions.
> Concerning DCM for induced responses, I have two basic questions and am happy for any consideration.
> .
> First, you have the option for several inputs when you define your model. I am considering a time window of 1000ms and
> the induced activity linked to movement starts around 150-250 ms (with trial-to-trial variability). I am thinking now if I may use
> several inputs, e.g. 100 200 300 400 500ms because 1. of the variability of the onset 2. the induced components are not nearly as sharp
> as an averaged ERP and 3. the movement needs to bei maintained steadily consciously which I interpret as ongoing input which I want to put into the model;
You could also consider changing the width of the input pulse. This is
an option available in the latest SPM8 update. You could for instance
have one input in the middle of your hold period with sd wide enough
so that there will be some input throughout that period + maybe
something for the onset, you can play with it and do model comparisons
to find the optimal input configuration. I think it's preferable to
multiple discrete inputs which will generate 'bumps' that the model
will have to try to suppress by dynamics.
> I donīt have any results with just one input, which might be due to the prior for the A-Matrix (see DCM.M.pE.A), so that the modell reaches very early convergence?? The other thing that prevented too fast convergence was changing the scaling factor in the spm_cond_units.m script (e.g. to fixed number of 1000) though I am not completely aware of the effect of this.
> .
This just scales the data. I would try changing the inputs with the
scaling as it was. All those things have been optimized to work
together by Karl mostly by trial an error so by changing some random
bit in the code it's much easier to break it than to fix it.
> My second question: When I am only analyzing one condition, the values for the A- and B-Matrix are the same. Is it therefore advisable in general to use a second, contrasting condition (e.g. rest). I think that I should not use a similiar condition (e.g. fast and slow finger movement). In general, when I am sure about the sources I am using in the modell and that these are engaged in the task, is it even better to use only one condition without the contrast to a control condition.
> .
In general it is more interesting to look at the condition effects (B
matrix) as they are more likely to reveal something interesting.
However, I don't think having rest as the second condition is a good
idea as rest doesn't have any interesting dynamics to model so for
DCM-IR the best thing to do is just to leave the coupling at zero
which is not very interesting. Fast and slow finger movements are more
interesting to compare, the only issue is how to keep the timing in
the trial comparable. In general there are many things you can do, but
only a small subset of them is likely to yield interpretable results
and that's what I'd suggest you to focus on.
Best,
Vladimir
> Dear all, thank you very much for reply. All considerations are welcome.
> Carlo
>
> (Carlo Huber, University Hospital Cologne, Neurology, Germany)
>
|
