Using 3dMEMA for One Sample T-test

Hello AFNIfam,

Apologies if this question has already been asked; I have had trouble finding any sort of documentation for this question.

We’ve generated a 3dMEMA file according to example 1 in the -help file. If I’m correct this whole brain models asks the question “where in the brain is effect > 0” by using the b and t values. If we have an ROI-based hypothesis, what is the proper way to use 3dMEMA output to answer the question “is effect > 0 for this ROI”? Some ideas we had:

  1. Mask 3dMEMA on the ROI rather than a group anatomical ROI and then use the AFNI GUI to look for significant signal in the ROI
  2. Extract betas from the 3dMEMA output using 3dROIstats and then run a post-hoc t-test.

But both of these seem inefficient and a bit…post-hoc-y. Is there a way to do this in the box with 3dMEMA? Any help with this would be greatly appreciated!

Here’s our 3dMEMA specification:

3dMEMA -prefix gain_ei
-set gain_ei
sub-001 /data1/psychology/narps/results/sub-001/stats.sub-001_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-001/stats.sub-001_REML+tlrc’[Trial#1_Tstat]’
sub-002 /data1/psychology/narps/results/sub-002/stats.sub-002_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-002/stats.sub-002_REML+tlrc’[Trial#1_Tstat]’
sub-003 /data1/psychology/narps/results/sub-003/stats.sub-003_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-003/stats.sub-003_REML+tlrc’[Trial#1_Tstat]’
sub-004 /data1/psychology/narps/results/sub-004/stats.sub-004_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-004/stats.sub-004_REML+tlrc’[Trial#1_Tstat]’
sub-005 /data1/psychology/narps/results/sub-005/stats.sub-005_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-005/stats.sub-005_REML+tlrc’[Trial#1_Tstat]’
sub-006 /data1/psychology/narps/results/sub-006/stats.sub-006_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-006/stats.sub-006_REML+tlrc’[Trial#1_Tstat]’
sub-008 /data1/psychology/narps/results/sub-008/stats.sub-008_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-008/stats.sub-008_REML+tlrc’[Trial#1_Tstat]’
sub-009 /data1/psychology/narps/results/sub-009/stats.sub-009_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-009/stats.sub-009_REML+tlrc’[Trial#1_Tstat]’
sub-010 /data1/psychology/narps/results/sub-010/stats.sub-010_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-010/stats.sub-010_REML+tlrc’[Trial#1_Tstat]’
sub-011 /data1/psychology/narps/results/sub-011/stats.sub-011_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-011/stats.sub-011_REML+tlrc’[Trial#1_Tstat]’
sub-013 /data1/psychology/narps/results/sub-013/stats.sub-013_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-013/stats.sub-013_REML+tlrc’[Trial#1_Tstat]’
sub-014 /data1/psychology/narps/results/sub-014/stats.sub-014_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-014/stats.sub-014_REML+tlrc’[Trial#1_Tstat]’
sub-015 /data1/psychology/narps/results/sub-015/stats.sub-015_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-015/stats.sub-015_REML+tlrc’[Trial#1_Tstat]’
sub-016 /data1/psychology/narps/results/sub-016/stats.sub-016_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-016/stats.sub-016_REML+tlrc’[Trial#1_Tstat]’
sub-017 /data1/psychology/narps/results/sub-017/stats.sub-017_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-017/stats.sub-017_REML+tlrc’[Trial#1_Tstat]’
sub-018 /data1/psychology/narps/results/sub-018/stats.sub-018_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-018/stats.sub-018_REML+tlrc’[Trial#1_Tstat]’
sub-019 /data1/psychology/narps/results/sub-019/stats.sub-019_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-019/stats.sub-019_REML+tlrc’[Trial#1_Tstat]’
sub-020 /data1/psychology/narps/results/sub-020/stats.sub-020_REML+tlrc’[Trial#1_Coef]’ /data1/psychology/narps/results/sub-020/stats.sub-020_REML+tlrc’[Trial#1_Tstat]’
-mask narps_group_3_7+tlrc \
-missing_data 0
-no_residual_Z
-no_model_outliers

If we have an ROI-based hypothesis

I assume that you can define the ROI as a mask. If you have already performed whole-brain voxel-wise analysis, you can apply a mask for the ROI to the result. It’s not clear to me whether you want to maintain the voxel integrity within the ROI or you want to treat the ROI as one measuring unit.

Yes, we have the ROIs as masks (in BRIK and HEAD formats). If we are interested in getting a single answer for the entire ROI (i.e. averaging across voxels), what is the correct way to do this? Are you suggesting we mask the 3dMEMA on our ROI instead of the group anatomical mask?

Best,
Andrew

If we are interested in getting a single answer for the entire ROI (i.e. averaging across voxels), what is the correct way to do this?

By “a single answer for the entire ROI)”, do you mean that you want to obtain the effect estimate and some statistical stamp about the effect? In that case, you can simply perform a Student’s t-test on the averaged effect size of the ROI from each subject with a generic statistical tool such as R, SPSS, …