Thanks, Rick. I'll look into detrending as you suggest.

I have also gone a different route by mean padding the time series to appear continuous. You suggested this here.

As a proof of concept, I want to calculate a sound>no sound contrast using the padded time series. Previously in this thread you showed me how to generate this contrast by specifying my own design matrix using -ortvec and my own -glt contrast. So I wanted to compare the two results with the understanding that the should be identical.

1. I created indices to insert the mean time series in the proper location

1deval -expr 'bool(mod(t+1,4))' -num 240 > pad_times.1D

2. I created a mean volume and appended it to the end of the scan.

3dTstat                                   \
     -prefix dummyMean          \
     -datum 'short'                    \
      pb04.scale+tlrc                \
3dTcat                                    \
     -prefix 1pad.ts                   \
     pb04.scale+tlrc                  \
     dummMean+tlrc
3dTcat                                     \
     -prefix pb05.dummyPad    \
     -1pad.ts+tlrc'[1dcat pad_times.1D']

3. Then I pad the motion (and derivatives) and create a censor file to censor the mean-padded volumes. Initially I created a censor file for volumes that exceed 0.5mm FD, but lets just say that this person did not move so that I have a vector of all 0s.

1deval -expr 'bool(mod(t+1,4))' -num 240 > issues_pad.1D
1d_tool.py                                        \
     -infile MoPar.1D                          \
     -censor_fill_parent isss_pad.1D \
     -write MoPar_pad.1D
1d_tool.py                                        \
     -infile censor.1D                          \
     -censor_fill_parent isss_pad.1D \
     -write censor_pad.1D

4. Then I regress...

3dDeconvolve 												\
	-input $runs -jobs 12										\
	-polort A 	                                                                                                \
	-ortvecMoPar_pad.1D              demean 							\
	-ortvec MoPar_derv_pad.1D    derv 							\
	-censor censor_pad.1D 							                        \
	-num_stimts 2 											\
	-stim_times 1 trials.1D 'BLOCK(0.5,1)' 	            -stim_label 1 trial 	\
	-stim_times 2 catch_trials.1D 'BLOCK(0.5,1)'     -stim_label 2 cat    	\
	-gltsym 'SYM: trial' -glt_label 1 sound 							\
	-overwrite -tout -bucket trial.stats 							        \
	-x1D xmat.1D -x1D_uncensored uncensor.xmat.1D

I have attached 2 images. The one that looks like the auditory cortex was created without padding the time series, but I specified the design matrix using -ortvec and you suggested earlier in this thread. The image where the whole brain is deactivated are the results of the same sound>no sound contrast, same voxelwise threshold (p<0.01) and same cluster correction (arbitrary k=50, just to clear it up) but I used the mean-padding procedure outlined above.

Do you have any recommendations? Thanks for your help!

Dustin