Dear Georgia
I didn't fully understand the question, but hopefully I can still clear up the confusion. You are modelling a task with a single experimental factor that has two levels. There are three separate stages you need to follow:
1. Identifying where in the brain there is an effect of fearful faces
The design matrix you specified for this isn't ideal, because it is "rank deficient". The regressor for the task condition is equal to the fearful faces regressor plus the neural faces regressor. When estimating a GLM, you don't want one regressor to be a linear combination of other regressors. (I know the attention demo included with SPM does this, causing some confusion!).
If subjects were constantly looking at neutral or fearful faces, with very small gaps and no rest periods, you could have just one condition in your design matrix: fearful faces. Neutral faces then become the implicit baseline. Alternatively, if there were resting periods, then I recommend modelling both conditions: fearful and neutral. Use the contrast of fearful vs neutral to select your regions of interest.
2. Extract representative timeseries from these regions.
Use the design matrix described above to extract timeseries. You need to tell SPM which effects are interesting. So prior to extracting timeseries, create an F-contrast (it doesn't matter whether this is the first contrast) with a name like "effects of interest" (you can call it anything you want). If you have two face conditions in your design matrix, and they are the first two regressors, then define this as:
[1 0;
0 1];
Then when you extract timeseries, use your fearful vs neutral contrast to select where in the brain the VOI should be placed, and when you're asked about "adjusting " the timeseries for nuisance effects, select the effects of interest F-contrast.
3. Specify your DCM. You will need to choose an SPM.mat, so that DCM knows the timing of your conditions. It could be the same SPM.mat you used above for VOI extraction, or it could be a different one with the conditions re-arranged for convenience. As you suggested, some people collapse multiple conditions into a "task" condition to use as the driving input, which captures the average driving of the conditions. This assumes you have some rest periods in the experimental - so that effectively you've got two experimental factors - task vs rest and fearful vs neutral. Task becomes the driving input and fear because the modulator.
If you'd like to do that, you can use the design matrix you described (with three conditions: fearful, neutral, Task) during DCM specification. Then use Task as the driving input and either fearful or neutral as modulators (you don't need both - as one will have the negative effect of the other).
All the best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Georgia
Sent: 25 April 2022 11:48
To: [log in to unmask]
Subject: {SPAM?} [SPM] Time-series extraction DCM
⚠ Caution: External sender
Hello,
It is my first time doing a DCM analysis. We are performing an emotional task (fearful, neutral faces) DCM analysis.
Our SPM design for the DCM model has 3 columns: the fearful faces, neutral faces and the task (all faces-neutral and fearful). The fearful and neutral is the modulation (B matrix) and the task is the driving input (C matrix). For extracting the time-series we need an F-contrast ("effect of interest") to adjust the data. We cannot set up the contrast to be CON1 as the effect of interest, but if we didn't have the 3rd column (task) the contrast for calculating the effect of interest would be CON2. Can we extract the time-series from CON2 but do the DCM analysis on the first design matrix? But isn't this going to produce wrong time-series?
CON1 = [1 0 0
0 1 0]
CON2 = [1 0
0 1]
Best regards,
Georgia
|