Posted by B. Douglas Ward on March 14, 2001 at 10:13:01:
In Reply to: Using Deconvolution to Estimate Impulse Response Functions posted by Hillary Schaefer on March 13, 2001 at 17:16:38:
Hillary Schaefer:
1. Since different slices are acquired at different offset times, you can
think of each slice as sampling the hemodynamic response at different points
on the curve. For interleaved slices, contiguous slices differ by about 1/2 TR
in time offset. Hence, the kth IRF coefficient from one slice refers to a
slightly different point on the curve than the kth IRF coefficient from a
neighboring slice. This should account for the "striped" appearance.
If this bothers you, you could try using 3dTshift to align all of the slices
to the same offset time, and then use this interpolated 3d+time dataset as
input to 3dDeconvolve.
I prefer not to interpolate the input data. I just accept the fact that
different slices are sampling the data at different offset times.
2. This can be done, but I'm not sure that it makes sense to add the baseline
to the IRF. I think this is like adding apples and orangutans. The baseline
coefficients (b0 and b1) are stored in the output bucket dataset. You could
use 3dcalc to add b0 to each of the IRF coefficients. Or, you could add
b0+b1*N/2, where N is the number of time points.
Perhaps what you really want is the (full model) fit to the time series data,
which would include the baseline. This is provided as a 3d+time dataset
by the -fitts option. (See the documentation in file 3dDeconvolve.ps for
more details.)
Doug Ward