AFNI gods,
I have been trying to play with "raw" timeseries data recently. The goal (after preprocessing and run concatenation) is to use AFNI to remove the trends in the data (on a run-by-run basis), average these raw data within an ROI, and then import the resulting 1D file into MATLAB for further analysis. I saw Doug Ward's post from a few years ago (http://afni.nimh.nih.gov/afni/community/board/read.php?f=1&i=122&t=122#reply_122) in which he argues that the order of steps should not matter (whether detrending before averaging or vice versa). However, there seem to be (at least) two ways to accomplish the detrending (both of which are giving me troubles):

1) Using 3dDeconvolve before averaging over the ROIs with commands like:

3dDeconvolve -mask Subj_ROIs+orig -input Subj_allruns+orig \
-fdisp 123456 -concat cat1.txt -polort 2 -num_stimts 0 \
-errts Subj_allruns_detrend
3dmaskave -mask Subj_ROIs+orig -mrange 1 1 -quiet Subj_allruns_detrend+orig > Subj_ROIave_detrend.1D

When I do this, AFNI barfs on me with following message:
++ Program 3dDeconvolve: AFNI version=AFNI_2006_03_21_1314 [32-bit]
++ Authored by: B. Douglas Ward, et al.
Killed

This may be because the file 'Subj_allruns+orig' is large (~1 GB), which is why I am using a mask. However, when I use this same dataset in a standard 3dDeconvolve multiple regression analysis (i.e. with more than 0 num_stimts and a glt), I have no problems. Any ideas on what the issue could be?

2) Using 3dDetrend after averaging over the ROI's with commands like:

3dmaskave -mask Subj_ROIs+orig -mrange 1 1 -quiet Subj allruns+orig > Subj_ROIave.1D
3dDetrend -polort 2 -prefix Subj_ROIave_detrend Subj_ROIave.1D

When I run this, I get the following error message:
++ Program 3dDetrend: AFNI version=AFNI_2006_03_21_1314 [32-bit]
*** Illegal inputs to startup_lsqfit
** FATAL ERROR: Choleski factorization fails: linearly dependent vectors!

I'm not sure what to make of this error. However, I realize the 3dDetrend doesn't know about my concatenation. I assume that it can't take a -cat option (like 3dDeconvolve), so if/when it does work, it will be removing trends from the entire file (rather than on a run-by-run basis). Any thoughts?

Any and all help is greatly appreciated.

Thanks,
Adam