Individual Subject Differences

In Group Analysis, one typically tries to normalize, or remove, individual differences. Thus the result reflects what is in common among a group of subjects.

Once you've done that, however, you might be interested in going back and looking at those differences.

One common thing to look at is how the results covary with some behavioral measure, such as reaction time, accuracy, etc.

Here are several ways to do that using AFNI.

3dfim+

Say you have a % correct score for a given condition. You've already computed SAM contrasts for that condition (they can be anything, poststim vs. prestim, condition block vs. rest block, etc.). Thus you have, for each SAM volume, one number. First, create a 3d+time dataset where the time axis is not time, but ranges across those SAM volumes, in % correct order.

For example, let's say we have scores like this:

s1+tlrc 50
s2+tlrc 35
s3+tlrc 65
s4+tlrc 45

Order the datasets according to score, and create a 3d+time dataset:

3dTcat -prefix inorder s2+tlrc s4+tlrc s1+tlrc s3+tlrc

Also create an ideal file consisting of just the scores, also in the same order:

echo 35 45 50 65 > ideal

Now you can use 3dfim+ to do a voxelwise regression:

3dfim+ -input inorder+tlrc -ideal_file ideal \
    -polort 0 -out All -bucket result

The result+tlrc bucket dataset will contain correlation coefficients (and various other things, see the 3dfim+ documentation). If you open that dataset in one AFNI controller, plus the inorder dataset as a Graph in another controller, you'll be able to spot voxels that correlate with the behavioral measure.

For example, you might find less beta-band desynchronization for the subjects that performed more accurately, and explain this using an efficiency argument.

3dRegAna

This method yields the same basic result as the above, but is slightly simpler.

3dRegAna -rows 4 -cols 1 \
    -xydata 35 s2+tlrc \
    -xydata 45 s4+tlrc \
    -xydata 50 s1+tlrc \
    -xydata 65 s3+tlrc \
    -model 1 : 0 \
    -fcoef 1 result

waver and 3dDeconvolve

This one is covered in detail on the AFNI blog here.