Hello AFNI minds,
I am interested in transforming 3d+time EPIRT datasets into a percent-signal-change 3d+time dataset, where each voxel at each point in the time series is transformed into percent-different from the mean signal from that voxel across the time series. I do this not only to normalize datasets for between-subject comparisons, but also to apply VOI masks and dump out time course data ready-to-go as percent signalchange.
My event-related paradigm has no "baselines" so I use the session-average signal as a surrogate baseline, even though this artficially blunts signal change values since the baseline includes the occasional event of interest.
I can generate this alternative dataset (in native orig space) as part the following script which calculates a functional dataset:
****
foreach subject ( cr la sw ks jg jb mw mrm mn yt vw tc )
cd ../${subject}*
# Calculate percent signal change from session-wide mean voxel signal
if ( -e normalized+orig.BRIK ) then
rm -rf normalized+orig.*
endif
if ( -e TimeSerAvg+orig.BRIK ) then
rm -rf TimeSerAvg+*
endif
3dTstat -prefix TimeSerAvg EPIRTseries+orig
3drefit -abuc TimeSerAvg+orig
3dcalc -datum float -a EPIRTseries+orig -b TimeSerAvg+orig -expr "((a-b)/b)*100" -prefix normalized
# apply model to the normalized datasets
3dDeconvolve -input normalized+orig -concat runs.1D -mask 3dCL_mask+orig -nfirst 0 -num_stimts 12 \
-polort 2 \
-stim_file 1 'antc.1D' -stim_label 1 'ant' \
-stim_file 2 'rvnantlc.1D' -stim_label 2 'rewvsneu' \
etc....
*********
The "Normalized" variant .BRIK for some reason (additional decimal places?) is exactly twice the original EPIRTseries.BRIK and results in fairly similar individual statistical maps. With the normalized dataset, I can better compare event-related Beta coefficients between different patient groups. Also, just as individual subject *statisitical maps* made from the raw time series can be tlrc-warped and merged in to a group map, so can the individual maps derived from the normalized data as above.
My problem is, that Talairach warping of this normalized dataset ITSELF (using a hi-res MP-RAGE reference scan)
adwarp -apar MPRAGE+tlrc -dpar normalized+orig
for some reason results in unmanageable file sizes and "core dumps." Thus, I cannot warp out the normalized dataset so as to pass the data through a VOI mask placed in common stereotactic coordinates with 3dmaskave etc.
What can I do to warp out the %signalchange normalized dataset? I am using 3.8mm isotropic voxels with a 1mm gap. Is there some re-sampling parameter or command I could include that would not appreciably dilute the data but would make things more manageable? -dyxz 2.0?
Jim B
