Dear Haoran,
2011/6/21 飞鸟 <[log in to unmask]>:
> Dear Vladimir,
> I'm sorry to disturb you again, but as for the problem presented in the
> title, I still have two questions.
> 1. Since cingulate is below cerebral cortex but not very deep, can I
> always regard the cingulate as the non-hidden source?
You can if there are any components in the ERF coming from there. I
think there should be but in any case specifying the correct
anatomical location for this source will not be wrong (as well as for
other 'hidden' sources). The only question is whether there is any
evidence for that location being meaningful. Whether you want to
explore this question or not depends on your research aims.
> 2. Must the input be specified only on sensory cortex(e.g. primary visual
> cortex, primary auditory cortex)?
> Take an example:
> input----->V1---->Precuneus<---->PFC<----->Cingulate
> ↑ ↑ ↑
> ∣
> ↓ ∣
> ----→ hidden
> source ←----
> In this condition, can I specify the input on the precuneus directly? I
> think I can if V1 is not necessary for my network.
If you thing that the input to precuneus comes from the thalamus you
can specify it directly. But if you thing that the input comes from a
primary sensory area you should model that. First, direct input and
input via a cortical source will differ in their timing thus leading
to different prediction for precuneus activity. Second, usually the
activity from primary sensory areas involved in the task is stronger
than anything else and persists for hundreds of milliseconds. If you
don't model this activity it will be incorrectly attributed to other
sources so your whole analysis will be wrong.
> Further step, does the hidden source here play a role of all subcortical
> regions(e.g. amygdala, hippocampus and thalamus) ? And thus, I couldn't
> distinguish which regions make great effects to this network?
>
No, usually you would use different hidden sources for those different
areas. The question you should think about is what will make each of
those sources distinct in your model. For instance in the example I
gave you before V1 connected to 'hippocampus' and 'amygdala' you could
exchange the labels between the two hidden sources and it wouldn't
make any difference for the model. So you couldn't really say that
your 'hippocampus' has something to do with the real hippocampus. If
each of the hidden sources is connected differently to the other
structures and you can relate that to the real anatomy then it can
make sense. Again, try to have as few hidden sources as possible.
Perhaps you should fit a series of models with increasing complexity
and use model comparison to see whether a more complicated model is
justified.
Best,
Vladimir
> Thank you very much!
>
>
>
> --
>
> Haoran LI (MS)
> Brain Imaging Lab,
> Research Center for Learning Science,
> Southeast University
> 2 Si Pai Lou , Nanjing, 210096, P.R.China
>
>
|