Posted by B. Douglas Ward on April 05, 2002 at 14:11:20:
In Reply to: Re: condition comparion using 3ddoconvolve or anova2 posted by Gateman on April 04, 2002 at 18:51:39:
Hello Gateman:
The regression coefficients and GLT linear combination coefficients are
stored in separate sub-bricks in the output bucket dataset. The format
of the output bucket dataset depends on which output options you use.
For a description of the various output options, and examples illustrating
their use, see the documentation in file 3dDeconvolve.ps. Also, you can use
program 3dinfo to display the contents of the bucket dataset.
If you wish to "normalize" some output parameter as percent change relative
to the baseline, you can use program 3dcalc as follows:
Suppose that the output bucket dataset is myResults+orig. Suppose that
sub-brick #0 contains the baseline constant, and sub-brick #1 contains the
linear drift coefficient. Suppose that the parameter you wish to "normalize"
is contained in sub-brick #14. Then you could use the following command line
to calculate % change relative to the average baseline:
3dcalc \
-a "myResults+orig[14]" \
-b "myResults+orig[0]" \
-c "myResults+orig[1]" \
-expr "100.0*a/(b+c*(NT/2))" \
-fscale -prefix PCResults
where NT/2 must be replaced by a constant (NT = number of time points in the
fMRI time series).
If you want to plot the IRF as % change from baseline, then you would want to
use the entire IRF time series itself (not just the sum of coefs.). That is,
use the 3dDeconvolve "-iresp" command to write the IRF as a 3d+time dataset
(say, myIRF+orig), and then use 3dcalc to divide by the baseline:
3dcalc \
-a myIRF+orig \
-b "myResults+orig[0]" \
-c "myResults+orig[1]" \
-expr "100.0*a/(b+c*(NT/2))" \
-fscale -prefix normedIRF
The output, normedIRF+orig, will be a 3d+time dataset containing the IRF
"normalized" as % change from baseline.
For more details, please see the documentation in file 3dDeconvolve.ps.
Doug Ward