Dear Saurabh,
In DCM-IR induced responses are fitted with a linear dynamical system
of interacting modes. When there is only one input this kind of system
cannot generate interesting dynamics forever, it either diverges or
decays to zero. This is also the case for brain induced responses to a
single input. They usually last for a few seconds at most. You could
fit something longer by generating a series of inputs, like an input
for every key event in the movie or modelling the changes in luminance
with a series of gaussians. But I suspect fitting very long TF
responses will be challenging. DCM-phase is suitable for a situation
when you have phase dynamics converging to a steady state. I suspect
during a movie clip you'd have many changes in phase dynamics so
perhaps only shorts segments at a time would work. You could use
DCM-CSD if it's reasonable to assume that some segment of your data is
stationary so just looking at the fixed spectrum is a reasonable
description of it.
Best,
Vladimir
On Thu, Mar 8, 2018 at 1:17 AM, Saurabh Sonkusare
<[log in to unmask]> wrote:
> Hi Vladimir,
>
> Thanks for clarifying this.
>
> Can I also clarify as to how long the duration of the data should be for optimal results? I assume since its the ERSP data feature that DCM is modelling, it shouldnt be more than 1 second of data.
>
> Also if I have longer sets of dataset for e.g from movies as stimulus (more than 30 seconds) would DCM-phase coupling model such long datasets? Or DCM-CSD would be better for such data?
>
> Thnaks in advance.
>
> Regards,
> Saurabh
>
|