Yes absolutely fine to try varying the structure first then varying the modulatory inputs. Or you could try all at the same time. You might want to look at the family option in the model comparison (BMS) batch to group your models according to something they have in common, such as particular intrinsic connections.
I would make one model with the GUI and then do the rest with scripts. The matrices are stored in DCM.a, DCM.b and DCM.c. Your script would need to modify one or more of these matrices then save each modified model. Let us know if you need help writing that script.
From: Laura Enax [mailto:[log in to unmask]]
Sent: 10 June 2015 17:18
To: Zeidman, Peter
Subject: AW: RE: [SPM] DCM - define the GLM
Thanks a lot, this was a very helpful explanation. I think I was confused by the spm example. This makes sense now. The driving input could possibly enter at every region, if there is no anatomical restriction right?
I'm not sure about the network involved yet - is it all five regions or possibly less, and which region is connected with which region. Would it be valid to test for intrinsic coupling first and then test for the picture with feature modulator on the connections ? Or do you do this all at the same time with your dcm?
Last - do you know an efficient way to construct the dcm matrices? I know how to do it with the GUI and by hand one by one in a script, but are there more efficient ways?
Thanks a lot!
Zeidman, Peter schrieb am Mi., 10. Jun 2015 16:54 MESZ:
>The way you set up your GLM and DCM depends on the question you want to ask. What would you like to find out from the DCM?
>Let me give you an idea of what’s typical. I’ll assume you expect the feature to increase or decrease coupling between certain regions, when a picture is being displayed.
>1. For extracting ROIs, you would create a GLM with two columns: picture and picture+feature (plus any nuisance regressors). You should not model the fixation baseline in your GLM. Extract your ROIs from this GLM. Ensure you have an effects of interest F-contrast defined, and use this when you extract ROIs.
>2. Create a new GLM for use in DCM. This will have two columns called ‘stimulus’ and ‘feature’. The stimulus column contains the onsets of all picture trials, with or without the extra feature. The feature column only models trials with the extra feature. Now, construct a DCM with stimulus as driving input. Include ‘feature’ as a modulatory input, to test which connections are strengthened or weakened in the presence of the feature.
>From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Laura Enax
>Sent: 10 June 2015 14:21
>To: [log in to unmask]
>Subject: [SPM] DCM - define the GLM
>Dear SPM experts,
>I am currently struggling with the very first step in DCM - the GLM specification for my DCM analysis. As I read in several tutorials and also in the SPM manual, the GLM needs to be set up a little bit differently, compared to standard analyses.
>I have three conditions: subjects see a low-level baseline (fixation cross), a picture, or a picture with an additional feature. I am interested in how the third condition (i.e., picture with an additional feature) influences the connectivity between several regions.
>If I understand correctly, one has to have a factor that varies the driving input (i.e., this would be picture vs. fixation ?) and one factor that varies the modulating input (i.e., this would be picture with feature vs. picture?).
>Now I am not sure how to set this up in the GLM, which I need for further analyses. Is the driving input the fixation, or a combination between fixation and picture, or even between all of them? Is the modulating input only the picture with the feature, or both, the picture with the feature and the normal picture? If I need both pictures as modulators, would I have to test all combinations (i.e., test if picture modulates connectivity beween x and y, AND test if both modulate connectivity between x and y AND test if picture with feature modulates connectivity between x and y ? Or would I always specify that both (together) should modulate the connectivity beween region x and y?
>Sorry if I am unclear, feel free to ask questions.
>Thanks a lot