It can definitely be done in FEAT as beta-series is a strategy based on the GLM. However, I don't think there is an official script of set of instructions indicating the steps, so you'd have to do some homework. In principle you'd create one EV per trial convolved with the HRF, then concatenate the pe*.nii.gz files to form 4D files per subject, which can then be subjected to a 2nd level (still within-subject) GLM to test for an interaction between the seed betaseries and the condition(s), which now becomes akin to a PPI. If instead of multiple conditions within subject there is a single condition to be analysed across subjects, then maybe the dual_regression could be a shortcut, using the mask with the seed in the place of "melodic_IC" and the concatenated pe*.nii.gz files in place of the filtered_func_data for each subject.

I haven't run this myself so this general advice. Maybe others in the list can give better details.

I would however discourage subject-level inferences, i.e., running a 2nd-level within-subject correlating betaseries and obtaining a p-value for a particular individual. The reason is that the modelling of temporal autocorrelation in most FMRI software assumes an actual timeseries, whereas here the beta-series may have serial correlation patterns that don't follow usual assumptions. And of course, independence cannot be considered. For between-subject inferences things should be ok.

If you could, post here your results/scripts, and good luck!

All the best,

Anderson