Correlating Brain and Behavior in AFNI

If previous posts hadn’t given it away, AFNI is my tool of choice for processing neuroimaging data.  A lot of people get confused on the sheer number of options there are and that most things can be accomplished in a variety of ways.  Today I’m going to outline three ways of doing Correlations between fMRI data and some kind of external data (e.g., a reading measure).

Data: 10 subjects of MRI data and an associated data table

subject value
1     1.59583042
2    0.30977361
3    0.17275604
4    1.77287327
5    1.91858457
6    -1.26554353
7    1.73914139
8    -0.03359282
9    0.62022641
10    -2.0418750


3dTcorr1D: Perhaps the easiest way to run a correlation between brain and behavior is to run 3dTcorr1D.  First you will need to compile all of your relevant statistical coefficient files into a single NIFTI or HEAD/BRIK pair.  You can use 3dbucket to do this.  In this case, since I have 10 subjects I use 3dbucket to extract the condition I want to correlate from each subject and put it into an all_visual+tlrc dataset.  After this, then it’s just a matter of calling 3dTcorr on the dataset and supplying the covariate file listed above.  The [1] tell AFNI to get the column of values (because 0 would be the subject number) and the {1..$} tells AFNI to get all the rows after the header.  These are in order to match the input dataset created by 3dbucket.

3dbucket -prefix all_visual+tlrc Sub01+tlrc.HEAD[visual#0_Coef] Sub02+tlrc.HEAD[visual#0_Coef]...
3dTcorr1D -prefix corr_visual all_visual+tlrc. cov.txt[1]{1..$}


3dRegAna: 3dRegAna can be used to perform simple and multiple regression (or even step-wise regression) on fMRI data.  Here we do a simple regression (one DV, one IV).  Then we use 3dcalc to take the square root of the R-squared term and then get the sign for the correlation based on the b-value from the regression.

3dRegAna -rows 10 -cols 1 \
-xydata 1.59 "all/stats.Sub1.nii.gz" \
-xydata 0.30 "all/stats.Sub2.nii.gz" \
-xydata 0.17 "all/stats.Sub3.nii.gz" \
-xydata 1.77 "all/stats.Sub4.nii.gz" \
-xydata 1.91 "all/stats.Sub5.nii.gz" \
-xydata -1.26 "all/stats.Sub6.nii.gz" \
-xydata 1.74 "all/stats.Sub7.nii.gz" \
-xydata -0.03 "all/stats.Sub8.nii.gz" \
-xydata 0.62 "all/stats.Sub9.nii.gz" \
-xydata 0.91 "all/stats.Sub10.nii.gz" \
-model 1 : 0 -bucket 0 ./corr_visual
#calculates correlation coef
3dcalc -prefix corr22 \
-a story.corr_visual+tlrc.[5] \
-b story.corr_visual+tlrc.[2] \
-expr 'sqrt(a)*ispositive(b) - sqrt(a)*isnegative(b)'

3dttest++/3dMEMA: You can also use 3dttest++ or 3dMEMA to run a correlation, with the caveat that you need to standardize both your fMRI and behavioral data before running the analysis. -command 3dttest++ \
-dsets sub?.nii.gz sub??.nii.gz \
-subs_betas 0 \
-options -covariates cov.txt

Which will generate the following command:

3dttest++ -prefix ttest++_result \
 -setA setA \
 1 "sub1.nii.gz[0]" \
 2 "sub2.nii.gz[0]" \
 3 "sub3.nii.gz[0]" \
 4 "sub4.nii.gz[0]" \
 5 "sub5.nii.gz[0]" \
 6 "sub6.nii.gz[0]" \
 7 "sub7.nii.gz[0]" \
 8 "sub8.nii.gz[0]" \
 9 "sub9.nii.gz[0]" \
 10 "sub10.nii.gz[0]" \
 -covariats cov.txt
Comments are closed.