Hi Yi,

In the regression step, the event responses and
polort terms need temporal continuity. That means
running 3dDeconvolve with the complete 1200 TR time
series, but censoring out the dummy scans.

To do so, you would want to create a censor time
series (1 means keep, 0 means censor out). This
vertical text file would look like 240 sets of:

0
0
1
1
1


The question becomes when to include dummy volumes
in the time series, and even how to include them.

If you skip the tshift block, the preprocessing of
720 volumes can probably be done as is typical,
though the motion parameters will not be quite so
accurate (there may be little you can do about it).

But for the regression step, all inputs (EPI,
motion parameters, any tissue-based regressors,
typical censor files (motion, outlier)) would need
to be padded out to length 1200.

For this you could use the 1200 line 0,0,1,1,1
censor file and 1d_tool.py (see example 12 c from
the -help output) on the 1D files. But I know of
no convenient way to do it on the EPI data. That
might mean writing a little script to generate a
3dTcat command the fills in the extra 480 TRs
(with 0-volumes, or whatever you want).

Does that seem reasonable?

- rick