Hi,
This is in response to:
[
afni.nimh.nih.gov]
>Along those lines, I see that you are using motion parameters
>as regressors of no interest. This is actually a bit of a
>problem if you plan to compute %change _after_ 3dDeconvolve.
>A good subject can have some motion regressors that are pretty
>much flat. Such a regressor can "eat up" some of your baseline
>for the run.
I interpret the "good subject" in this explanation to be one who moves very little, is this correct?
Could you explain further how the motion parameters as regressors of non-interest "eat up" some of the baseline.
>The result is that the constant term of the baseline may not
>be accurate for each run (maybe if one removed the mean from
>the motion parameters).
>In any case, that is why we recommend converting to a %change
>_before_ 3dDeconvolve. See this post for a discusstion of
>the difference between computing a percent of the mean vs a
>percent of the baseline.
We have been calculating %signal change using the baseline constant (Run#NPol#0_Coef for N=1,2,3,4) averaged for 4 concatenated runs from the sub-bricks of the bucket dataset output of 3dDeconvolve.
motion parameter issues aside, Is this a correct way of calculating the baseline to then calculate %change?
Also, is just using the baseline constant term(Run#NPol#0) a reasonable baseline if the deconvolution used a cubic baseline ( -polort 3)?
I've been trying to hunt through pdfs and message board posts, but was still a little unclear on some of these issues.
Thanks in Advance,
Gabe