Dear Marina
When you select 'adjust for everything', it regresses everything in your design matrix out of your signal. This is typically used for resting state, where all regressors are uninteresting (e.g. motion). So in the case, you've regressed your task out of the timeseries, making it unsurprising the DCM didn't fit :-)
You need an effects of interest F-contrast - please see https://en.wikibooks.org/wiki/SPM/Timeseries_extraction
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Marina Shpaner
Sent: 14 June 2019 21:51
To: [log in to unmask]
Subject: [SPM] DCM analysis low explained variance
Dear experts,
I'm getting a very low explained variance (even when modeling 2 states) in individual dcm's, which is likely due to weak activations in some of the regions I'm interested in. I modeled the task differently for the DCM based on the demo example (visual motion) and also included multiple motion parameters.
My task has 9 blocks of pain presented under three different conditions, and I was interested in descending pain modulation. The acquisition is not ideal for DCM (interleaved and 3-sec TR). I used data that I have previously preprocessed in FSL as input (motion corrected, slice-timing corrected, and normalized), and I included a bunch of nuisance regressors (motion derivatives and motion spikes). Is this generally worth doing to improve modeling for DCM? In the attached example, the first 4 parameters are task-related, and the rest are nuisance.
When I extracted VOI timecourses, I adjusted for "everything." I'm not 100% sure what that means though. I'm assuming that's adjusting for the nuisance. Could you please clarify?
I really appreciate any feedback you may have, as I'm completely new to DCM and I don't have a ton of experience with SPM.:)
Thank you in advance,
-Marina
|