Dear Antoon,
In the `spm fmri' program a similar thing happens. In that case it is
caused by supersampling of the data. the temporal resolution of the HRF
is TR/16 instead of TR. Differences between the fitted response and the
real response are likely to be interpolation artifacts. However, I don't
know how much / where the code used by `spm fmri' and `spm pet'
overlaps.
Best regards
Alle Meije Wink
On Wed, 2005-01-19 at 14:15 +0100, Willemsen, ATM wrote:
> Hi,
>
> we have been struggling with the fitted and adjusted response
> for PET data. Both from literature and the archives I understood
> that the GLM could be written as:
>
> Y = [X1:X2][beta1:beta2] + error
> with [X1:X2] the design matrix split into effects
> of interest and confounding effects
> and [beta1:beta2] the corresponding model parameters.
>
> giving for the adjusted and fitted response:
>
> Y(adjusted) = Y - X2*beta2
> = X1*beta1 + error
> Y(fitted) = X1*beta1
>
> So for a PET data set with conditions only, I expect the fitted
> response to equal the betas for each condition. However, this is
> not what I see in the figure or when I examine the data in the
> Matlab workspace (SPM99).
>
> So obviously, I am missing something. I just can't figure out what.
>
> Any insight is appreciated,
>
> Antoon Willemsen
> Groningen University Hospital
> The Netherlands
|