Dear Haoran,
On Wed, Nov 23, 2011 at 3:02 PM, 飞鸟 <[log in to unmask]> wrote:
> Dear Vladimir,
> Thank you very much for your long term help. Last time you said you
> would update the spm and improve DCM, thus, I thought the problems I
> confronted might faciliate your work. I have confronted several problems
> about DCM for induced responses recently. But, I could not make them clear.
> The problems come as follow:
> 1. As mentioned in last letter, several steps in EM iteration might be
> time-consuming during dcm_ind computation, and might take 1 or 2 hours.The
> ensuing results appeared in matlab command window would be "...actual: NaN
> (3285.06 sec)". At the same time, other steps only took several seconds.
> This phenomenon would appeare if I changed the time-bin of DCM.options.Tdcm.
> Anyhow, time-consuming phenomenon was an event of low probability. I
> wondered whether it has special relationship with my dataset I choosed or
> the time-window I selected.
As I said, you should try the next SPM update. At the moment we are
still testing the modified code so I'd suggest that you wait till we
officially release it.
> 2. Considering the above problem might be resulted from the time-window, I
> checked the "spm_dcm_ind_data.m". As Chen said "Baseline power was removed
> by subtracting the frequency-specific power at the first time-bin "(Chen,
> C.-C., S. J. Kiebel, et al. "A dynamic causal model for evoked and induced
> responses." NeuroImage(0).). I checked the code which realizes this
> function in "spm_dcm_ind_data.m". But it seemed that we just subtracted the
> frequnecy-specific power at the first time point. I didn't know if it was
> the reason caused that problem. And I wondered whether it would be better to
> subtract the averaged frequnecy -specific power over a time window (eg. from
> -50 to 0 ms) but not at a time point '-50'.
Yes, this is one of the things the inversion is sensitive to. The
modified code averages over slightly longer window.
> 3. The results I got under one condtion showed that non-linear frequency
> coupling existed, though I just set it as linear in matrix A. Was it
> reasonable?
Because of reduction of frequency spaces to a small number of modes
linear coupling shows as things which are on the diagonal or symmetric
with respect to the diagonal. I suspect what you saw is some coupling
which was not on the diagonal but symmetric. It is not non-linear.
Non-linear would be something asymmetric with respect to the diagonal.
> 4. In Chen's paper(aforementioned above), it said we could not only deal
> with evoked reponses but also induced reponses by DCM_IND. I was confused
> with this because the framework of DCM_IND kept same but its funciton
> expanded a bit sudden relative to the first paper of DCM of induced
> responses (Chen, C. C., S. J. Kiebel, et al. (2008). "Dynamic causal
> modelling of induced responses." NeuroImage 41(4): 1293-1312.). Does
> spm8_4290 supply this function?
> 5. What if we take averaged trials into dcm_induced analysis? I know we
> can also get cross-coupling results. Intuitively, it seems to reflect the
> frequncey features of ERP, but is it meaningful?
> Thank you again for your great help to our research work. Best regards!
>
In the later paper what is meant by evoked responses is exactly ERPs
which are subjected to time-frequency analysis. So in this sense
DCM-IR can handle 'evoked responses' but it does not do the same thing
as DCM-ERP. The idea was to ask whether the features of the dynamics
requiring nonlinear coupling to model them are present in the evoked
response (i.e. things which are phase-locked to the stimulus) or they
are only in the part which is not phase-locked. For asking this kind
of question it makes sense to use averaging prior to DCM-IR but it
does not make sense in general.
Best,
Vladimir
> --
> Haoran LI (MS)
> Brain Imaging Lab,
> Research Center for Learning Science,
> Southeast University
> 2 Si Pai Lou , Nanjing, 210096, P.R.China
>
>
|