Hi Floris, Hi James,

I don't really have any ideas beyond what James
suggests: basically going to a 1 second TR and
censoring everything that does not apply. That
applies to the actual data analysis as well.

Using a 1-second TR would be appropriate to put
the TRs back on a regular grid.


But this is a messy approach. It means there
would be no slice time correction and no outlier
detection. And the regression would need to be
adjusted in a similar way to the -nodata test.

Though since the "TR=1 and censoring" method
would make combining with motion parameters messy,
the actual regression could be set up by first
computing a polort baseline parameter matrix, and
then simply handing it to 3dDeconvolve (and using
-polort -1). That way, 3dDeconvolve would not
have a dependency on temporal continuity.

Note that the motion parameters do not require
temporal continuity, not even for censoring.

- rick