Daniel,
A few points:
(1) You don't need to extract the PPI interaction term value around the given voxel. You can use the PPI contrast images in a higher level analysis with your behavioral measure as a regressor. Once you have the PPI contrast images, the analysis is the
same as your task activation analyses.
(2) If you have more than 2 conditions in your experiment (e.g. A, B, fixation or A, B, C, etc.), you should create separate PPI terms for each condition or create a series of PPI terms that span the experimental space (McLaren et al. 2012). This can be
done in FSL as well as SPM. The only difference between the programs is the deconvolution step in the SPM implementation of gPPI through the gPPI toolbox.