[VoxBo] 1.8.2 - convolution with an HRF

[Daniel Y Kimberg]

Aaron Heller wrote:
> I have been running GLMs with VoxBo 1.8.2 and have noticed an
> anomaly with some of our analyses. The problem is apparent when
> examining the fits of the GLM to the data. Strangely, the shape of
> the fitted parameter is different from the shape of HRF that was
> used to generate the covariate. It seems that the fitted response
> should appear to be a scaled version of the HRF, and that the shape
> should not be fundamentally different. To illustrate the issue, I've
> attached files showing the time course data (timedata.ref), the
> fitted parameter estimates (paramater.ref), and the empirically
> defined (subject specific) IRF that was used to generate the
> parameter (IRF.ref). I have attached those files.

Aaron, hi.  I probably need a bit more information to figure out where
this process is going wrong.  I think it has to be either a problem
with the G matrix (covariate has the wrong shape), with creating the
fit (it isn't using the right covariate), or with the averaging
(somehow not done correctly).

Did you do the time series averaging in the IDL-based voxel surfer on
in vbview?  If you could try whichever one you didn't use the first
time, it might be useful to know what it does.

In the meantime, if you look at the G matrix in gdw, does the
covariate in question look right there?  If not, I'll probably want
more detail on how you constructed it.  If so, we'll want to figure
out why it's not being rendered correctly.  If you look at the
covariate while surfing the GLM in vbview, does it have the same shape
there as in gdw, before averaging?  If not, does it look like one of
the other covariates?

dan