Dear Peter and DCM experts,
I have a result from some DCM fMRI analyses that I find a bit puzzling, and I was wondering if you could help explain what might be going on (or what I might have done wrong).
In brief, the puzzling finding is that Bayesian model reduction on a 9-node model that had 2 of those nodes receiving driving inputs produced a result showing a positive effect (i.e. a positive DCM.C.Ep value) on one of those nodes but a negative effect (i.e. a negative DCM.C.Ep value) on the other of those nodes. Yet the univariate GLM analysis showed enhanced activation for the relevant contrast (all stimuli > implicit baseline) in both those regions from which these 2 nodes/VOIs were extracted.
The 2 nodes receiving driving inputs were located in left and right lateral occipito-temporal cortex (LOTC). They were selected for the DCMs as suitable relatively early visual-processing regions that would be directly driven by the stimulus inputs (short movie clips of human movements) and because of their known involvement in action observation and understanding. I specified these nodes as having bidirectional inter-hemispheric connections, as well as bidirectional within-hemisphere connections to other nodes in the network. I specified the stimuli as both driving (C matrix) inputs into these nodes and as modulatory (B matrix) inputs into all extrinsic (and self) connections in the base model (see e.g. Allen et al., 2016, Neuroimage). Our main focus of interest in these analyses is the effects of the subject’s task on the extrinsic and self connections, and so all specified extrinsic and self connections were specified as being modulated by the task as well as by the stimuli.
In case it’s relevant, I used SPM12, version 6470. The DCMs were deterministic, bilinear, and two-state, with mean-centered inputs.
I look forward to hearing from you.
Best wishes,
Anthony
|