[VoxBo] Questions about Modify>Convolve

[Daniel Y Kimberg]

Joonkoo Park wrote:
> Q1 - Why do I see a portion of hemodynamic response decreasing at
> the very beginning of the run? (I do not have any "1" at the
> beginning of the run; the first "1" happens at 25 sec) I checked
> that the CONVOL function in IDL does not do this. I guess VoxBo is
> doing something more here, but I don't understand why VoxBo is
> assuming a decline of hemodynamic response at the beginning.

The convolution is done in frequency space, so the response to the
last event wraps around.  It shouldn't make much difference
ultimately, and I don't remember (or maybe never knew) the
justification for doing it this way.  Perhaps one of the old-timers
can chime in, but it probably has something to do with making sure the
convolution leaves you with a nice clean frequency structure, lacking
in high frequencies.  Obviously it does mis-represent the expected
signal a bit.

> Q2 - When I load the G matrix later, I can see the hemodynamics
> response graph again, but this time with jagged lines instead of
> smooth curves. Is this just a graphics issue or have my covariates
> actually changed?

VoxBo usually creates your covariates with high temporal resolution
(so that you can do things like create responses to events that don't
occur right on volume boundaries), and then resamples the final
product to the resolution of your data.  Usually it saves both a .G
file (same res as your data) and a .preG file (higher res).  If the
two get separated, and you browse the G matrix in gdw, you'll just see
the low resolution version.  Or if you open the G matrix in vecview,
you'll see the contents of your .G file.  That's what's usually
happening when you see a jagged version of your design matrix.  I
believe up until recent versions, only the G matrix got copied into
your GLM directory.

dan