3dDeconvolve to calculate within-individual-subject change in contrast activation between T1 and T2


I’d like to examine change over time (weeks) in activation by a particular two-run task, within an individual subject (i.e. get a map of that person’s contrast-elicited activation at T2 minus contrast-elicited activation at Time1).

What I’d like to do is essentially what was recommended in this ancient post:


However, I’m wondering what the syntax would be. I want to AVOID having to actually spatially align and concatenate my two time series into one long one, and change all the elapsed times of the task stimuli/events of the T2 scan by adding the total duration (s) of the first scan’s time series etc.

Is there a way to specify that 3dDeconvolve invoke one time-series EPIRT dataset as session 2, and a different time-series EPIRT dataset as session 1 and contrast them somehow, to yield a bucket dataset? What would the model specifier be to denote within-subject analysis?

I see no exemplar for this in the -help or in the 3dDeconvolve pdf manual.




I’m not so sure what issues or difficulties are involved in alignment between the two sessions. Anyway, you don’t have to analyze both sessions in one model. Instead of you can perform the analysis separately for each session, and take the effect estimates to the group analysis. You still would have to align the data between the two sessions, and the difference is whether the alignment is done before or after the regression analysis.

Hi Gang,

My query is actually a follow-up to your offline instructions to me, where perhaps by posting here, many others could benefit from advice on the same scenario. It pertains to how I could use 3dMEMA (which seems to be more sensitive than 3dMVM) to conduct a group X time analysis to get the interaction-based activation maps.

Your suggestion below:

"If I understand your case accurately, you have a 2 x 2 design with two groups of subjects, each of them went through two scanning sessions. If you use the ANOVA approach, you would go with 3dANOVA3 -type 5.

On the other hand, if you want to use 3dMEMA, you can do this:

  1. You should have the following three beta values (plus t-stat) available from each subject’s 3dDeconvolve/3dREMLfit: time1, time2, and time2-time1.
  2. Create SIX 3dMEMA scripts (if you want all six), one for each of the following case: time1 for each group separately, time2 for each group separately, and time2-time1 for each group separately.
  3. For interaction, create a 3dMEMA script for two-group comparison with time2-time1"

Was followed up by this clarification:

“Yes, concatenation is the way to go with 3dDeconvolve; however, you don’t need to physically glue to the two runs/sessions. Instead, you provide both files as two separate input files in the same 3dDeconvolve command (and you don’t have to feed both stimulus timing files though).”

Hence, my post here. Could you perhaps provide some final (?) advice on how to get the time2-time1 within-subject map for each subject singly, so that I could perform this three-stage analysis?

Many thanks



In that case, you do need to align the two sessions and feed them into 3dDeconvolve/3dREMLfit, and construct a contrast between the two sessions.