It is not clear what example processing script you are
referring to. In general, scaling each time series before
running 3dDeconvolve is typical (except for resting state).
But I am not sure where you are picturing 3dTcat in there.
In CD_CorrAna.html, the input is not detrended because
there still a full regression model to apply. But the
seed is detrended in order to carry out the deconvolution
and reconvolution steps. And the seed scaling matches the
input, since it explicitly starts from the scaled input.
The differing trends between the original input and the
computed seed time series should not be important, since
the trends are part of the regression.
Having said all that, I think the seed time series should
really be initialized from the errts time series, but that
should not have a big effect. At some point, I will write
a little program to help accomplish all of this.
- rick