Dear Allison,
your problem sound a bit complicated and you might have to do things
manually. Here is a quick suggestion to start off with that avoids having
to modify SPM code.
First, you only specify (but not estimate) a design with block regressors
for instruction and feedback.
Second, you retrieve the regressors for these events from the first design
matrix (from SPM.xX.X, the names of the regressors are stored in
SPM.xX.name) and you save them separately.
Third, you specify and FIR for your task and then include the save
regressors from the first design as "user regressors". If you are using the
GUI, this would be under the item "Regressors" in the configuration tree.
If you are batching using the job manager this would correspond to
jobs{1].stats{1}.fmri_spec.sess(1).regress.name and
jobs{1].stats{1}.fmri_spec.sess(1).regress.val (here you provide the
additional regressor).
Then, you finally estimate the design.
I have never done it like this, but in theory it should work.
I *think* all regressors should be high-pass filtered and also whitened
with AR(1) (if you have specified them), so you should be safe on that
side, but you might want to check that.
There may be some estimation issues when combining FIR and convolved HRF
regressors, but someone else would have to comment on this.
Good Luck,
Jan
Allison Nugent wrote:
> Hello,
> I am currently assisting a colleague in analyzing a rather problematic
> study using SPM5. Although there are problems with the design of which we
> are only too aware, at this point the data has been collected and we must
> simply make the best of it.
>
> The task block consists of a 3 second instruction, a 21.6 second task,
> and a 5 second feedback. Task type alternates, with four different flavors
> of the task, and 10 blocks per run. There is no break between the blocks,
> although there is fixation before and after the 10 blocks.
>
> I began by modeling this as a block design study, with 12 regressors -
> one each for instruction, task, and feedback for each of the four flavors.
> The results were not good. I began to inspect the time series, where I
> discovered that response to the task was not sustained over the entire 21.6
> second interval, which was not well fit by the block model.
>
> So, I've turned to a FIR model - but here's where I've run into snags.
> Ideally, I think I would like to model the instruction and feedback as short
> blocks, and only model the task itself with the FIR regressors. However,
> this does not seem possible within the SPM5 interface - is there a way to do
> this outside the interface? Alternatively, is it possible to use a
> different number of FIR regressors for different stimuli? It would be
> unreasonable, I think, to model the 2s instruction block with the same
> number of FIR regressors as the 21.6 second task. Our final option would be
> to combine the instruction and feedback and task into one giant block.
>
> In any case, I would appreciate any help with enhancing the flexibility
> of the FIR models, as well as any input on how to model a problematic task.
>
> Thanks!
>
> Allison
>
--
Jan Gläscher, Ph.D. Div. Humanities & Social Sciences
+1 (626) 395-3898 (office) Caltech, Broad Center, M/C 114-96
+1 (626) 395-2000 (fax) 1200 E. California Blvd
[log in to unmask] Pasadena, CA 91125
|