Dear Gabriel,
Depends on the purpose. If this cluster is from an independent data set / study, then you can safely conduct inferential statistics on that cluster. If the cluster was derived from smoothed data I would suggest to smooth your data accordingly, if the cluster stems from unsmoothed data I'd extract the average value from your unsmoothed data. As you average across several voxels at least anyway there should be no issue with unsmoothed data (otherwise you might lose information = cluster is based on unsmoothed data with higher anatomical precision, but you apply this to smoothed data with less precision).
In case this cluster reflects a sig. effect from a some whole-brain analysis for which you want to present a composite score for illustrative purpose I would rely on those files that you used for the initial analysis (likely smoothed ones).
Best
Helmut
|