Dear Francis,
> I am new to SPM and have recently scanned subjects while they learn a
> visuo-motor task. I now want to correlate a range of behavioural
> values (like reaction time,error, movement distance etc.) generated
> from the learning session with the fMRI data. The idea here is to
> relate the trend of the beh. variables across the session to the time
> course of the signal in various brain regions in order to try to say
> something about when areas are more important or less important as the
> taskis learned. I am interested in both positive and negative
> correlations.
>
> I have now run an analysis in SPM 96 with the following parameters:
>
>
> # of subjects: 8
> type of response : user specified
> # of conditions : 1
> # covariate of no interest: 0
> # of contrast: 2
> contrasts: 1 =(1 1 1 1 1 1 1 1) for the pos corr
> contrasts 2 = (-1 -1 -1 -1 -1 -1 -1 -1) for the neg corr
>
> I created a vector consisting of the beh variable of interest(eg
> error)and used this vector in the user specified field (i.e. one error
> value for each scan)
>
> The scanning design was made up of:
>
> 1. a one minute period of baseline at the start (dark screen) followed by
> 2. 80 trials of the task (160 volumes) equal to a total of 400 sec. followed by
> 3. a one minute period of baseline at the end
>
> Questions
>
> Is this approach to the correlation analysis correct?
Yes indeed.
> Is there another option in SPM 96 I should use?
There are a number of ways forward. First you could model both the
behavioural effects and the response to task preformance per se and then
the interaction between these. You could model all the behavioural
variables togther after some suitable orthogonalization (and
interactions among them).
> Does SPM carry this type of analysis out as a simple regression of the
> functional data on the behavioural data?
Yes it does (more exactly a multiple linear regression).
> Which of the statistics SPM(Z) or SPM(F) should be used in this analysis ?
If there is just one explanatory variable (or contrast of variables)
then use the SPM{t}. For collective inferences about a number of
causes use the SPM{F}. For just one covariate the SPM{F} = SPM{t}
squared (but the latter preserves the direction of change)
> Lets assume that the procedure outlined above is correct and I am
> looking at an activation map (SPM(Z)) using contrast no 1 (pos corr).
> The behavioural variable is an error score from each of the trials.
> Lets say I have a cluster of voxels in BA 4 (motor cortex). Can I
> interpret this as suggesting that as the error score increases the
> level of activity (or intensity) also increases?
Yes. To say that activation has increased you would have to model a
task x error interaction.
> Likewise if I get an activation cluster with contrast no 2 (negative
> correlation) for movement distance in the motor cortex, can I interpret
> this as suggesting that as movement distance decreases there is an
> increase in the level of activation (or intensity) ?
Yes, but as above there is a difference between activity and activation.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|