Hi Peter - thanks so much for your quick reply.
- I extracted ROIs based on the 2nd level T-map (one sample T test). Not
quite sure what you mean by effects of interest F contrast. If you are
referring to using an F contrast to look at correlations with canonical
HRF and/or time, dispersion derivatives, then I have done that and the map
looks basically the same as the T-map.
- I went with 2 state DCM because I tested out BMS using stochastic 1
state vs stochastic 2 state in a couple of subjects, and the 2 state
models had dramatically (~4000) greater log-evidences.
- We did not collect intracranial recordings - just fMRI.
- Driving input: No, I did not include a driving input. I am wondering if
this is the key? Last night, instead of including the effect of stim in
the B matrix, I tried including it as a driving input (C matrix) instead.
The log evidences for these models came out much greater. Would it make
more sense, therefore, to include the effect of stimulation as a driving
input as opposed to a modulatory input, and to vary the anatomical priors?
Intuitively, this seems to make more sense to me, since what I am trying
to test is: how does the stimulation-evoked activity travel through the
network nodes that we have identified. Wouldn¹t the A matrix values
represent the possible ³paths² that the stimulation-evoked activity is
taking to travel between these nodes?
Thoughts?
Thanks
-Will
On 4/28/15, 3:41 AM, "Zeidman, Peter" <[log in to unmask]> wrote:
>Dear Will,
>This is indeed surprising, which probably means there are some
>inappropriate priors somewhere. Some initial questions please:
>
>- How did you extract your ROIs? Did you have an effects of interest
>f-contrast?
>- Did you have a driving input in your DCM?
>- Have you tried switching off two-state DCM? It shouldn't cause a
>problem, but I'm not aware of it having been used with stochastic DCM.
>- Were you also collecting intracranial recordings from the subject's
>brain, or was this non-invasive?
>
>Best
>Peter
>
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
>Behalf Of Will Gibson
>Sent: 28 April 2015 03:04
>To: [log in to unmask]
>Subject: [SPM] Confusion about DCM result
>
>Dear DCM experts: I just ran bayesian model selection on a dataset (16
>subjects, 1 run per subject), and got a rather confusing result. Please
>let me know if you can help:
>
>The experiment was simple: an electrical stimulus was applied to each
>subject's brain in a block design (subject at rest). Activation (2nd
>level - one sample t-test, pFWE < 0.05) was observed in 2 brain areas
>with established connectivity. I then set up 16 competing models. The A
>matrix included within- and between-region connections for the 2 regions.
>The A matrix was held constant across all models. The B matrix
>representing the effect of the stimulus was varied across model space,
>with the stimulus affecting all possible combinations of the 4 A Matrix
>connections (2^4 = 16 models per subject).
>
>The winning model (by both RFX and FFX BMS) was the model in which the
>stimulus affected None of the connections. Since the group activations
>must be a result of the stimulus (the subjects were otherwise at rest),
>how can it be possible that the stimulus did not affect connectivity
>between (or within) the activated regions? This result makes absolutely
>no sense to me.
>
>I used stochastic, two-state DCM within SPM12.
>
>Please let me know if you have any thoughts/suggestions.
>
>Many thanks.
>
>-Will
|