Dear Maria,
If I understand your question correctly, then you need to first perform
10 t-contrasts (one for each subject) that measure how much the thalamus is
activated in condition 1 relative to condition 2. In the context of a fixed
effects analysis, you would perform a t-contrast by placing, for example, a
1 in the contrast manager every time condition 1 occurs, a -1 in the
contrast manager every time condition 2 occurs, and zeroes elsewhere. Note
that in a fixed effects analysis, all runs from all subjects are lumped
together. Therefore, the 1s and -1s are averaged across all runs and all
subjects.
It sounds as if you want to correlate the extent of thalamic activity in
condition 1 relative to condition 2 with a behavioral measure of emotion on
a subject-by-subject basis. To do this, you need to perform 2 steps.
First, compute N t-contrasts, one for each subject. As in the fixed effects
analysis, each contrast should contrast activity in condition 1 with that in
condition 2. However, each contrast should be performed using the data from
just one of the N subjects in your design. Let's assume that you have 10
subjects in your design, that each one was tested in 10 runs, and that
you've entered the 100 runs in the order that they were collected into your
fixed effects model. In that case, the first 10 runs belong to subject 1,
the second 10 runs belong to subject 2, etc. To compute a t-contrast for
subject 1, enter 1s and -1s for conditions 1 and 2 in the first 10 runs and
zeroes for all other subjects. A similar procedure can be used compute
t-contrasts for the other subjects. Unless you have specific hypotheses
about temporal or dispersion differences between your trial types, you'll
likely want to enter the 1s and -1s for the Betas that represent the
amplitude of the BOLD response for conditions 1 and 2. Each time you
perform a t-contrast, both an SPM_T*.img file and a con*.img file will be
produced. Therefore, at the end of this procedure, you should have N
SPM_T*img files and N con*.img files, assuming that you have N subjects.
After all the t-contrasts are made, you can select Basic Models on the
SPM GUI. You'll then see an option for 'Simple Regression (Correlation).'
Select this choice and you will be prompted to select some images to enter
into the correlation. If you want to correlate the t-values with the
behavioral measure, select the N spm_T***.img files. If you want to
correlate the difference in Beta weights for conditions 1 and 2 with the
behavioral measure, select the N con***.img files. Correlating the t-values
with the behavioral measure will assess the degree to which a reliable
statistical result is associated with your behavioral outcome. Correlating
the difference in Beta weights for conditions 1 and 2 with the behavioral
measure will reveal the degree to which a difference in the BOLD response
between conditions 1 and 2 is associated with your behavioral outcome. When
you are finished selecting all N images, hit done and you will be asked
several questions.
Here are some possible answers, which should get you started.
1. If you've already performed global scaling during the estimation of your
fixed effects model, I believe you should say no when it asks whether you
want grand mean scaling.
2. I generally say no to threshold masking at this stage.
3. As for the implicit and explicit masks, these are the same as for other
contrasts so if you don't want any masks just say no.
4. I think you can omit the global calculation.
5. Finally, you'll be asked to enter the covariate (i.e., behavioral
measure). If you entered the images in order of subjects 1 - N, you should
enter 1 behavioral measure per subject in order of subject 1,2, .....up to
subject N. In your case, this will be the single score from the emotion
scale for subject 1, the single score from subject 2, etc.
6. Next, you may choose whether to estimate the model now or later.
Once you have estimated this model, I believe that an SPM.mat file and
associated images will be deposited into the current working directory. If
you open this file from within the results section, you can make new t
contrasts. I think that entering a 1 into the contrast manager will show
voxels where greater activity in condition 1 than in condition 2 is
positively correlated with your behavioral measure. Entering a -1 into the
contrast manager will create a t contrast that reveals negative
correlations.
Hope that's helpful,
:> Daniel
Daniel Weissman, PhD
Center for Cognitive Neuroscience
Duke University
Durham, NC 27705
phone: (919)-681-1029
fax: (919)-681-0815
e-mail: [log in to unmask]
----- Original Message -----
> Dear spmers,
>
> I have run a fixed effects fMRI analysis and now that I have certain areas
of
> activation I have been asked to covary, for example the thalamus
activation,
> with the emotion scale results that we have for each subject. I'm not at
all
> sure how to go about this.
>
> Any guidance is greatly appreciated.
>
> Maria
>
>
|