I would add that because ICC is sign independent and only tells one the "magnitude" of any degree centrality changes, it is necessary to conduct post-hoc seed-to-voxel (or ROI-to-ROI incorporating your ICC seed) analyses to determine the directionality of functional connectivity changes that account for what may be driving any ICC blobs that survived in your initial analyses.  

A good example of this is a paper we recently published in HBM (see here: https://www.ncbi.nlm.nih.gov/pubmed/2916...).  We found group-wise ICC (degree centrality) changes pre/post-surgery in the precuneus/posterior PCC.  Using this location as the seed for subsequent post-hoc analyses (using the same model as the initial ICC analysis) it was determined that the significant change in degree centrality magnitude was driven by a combination of functional connectivity increases and decreases to the precuneus/posterior PCC.

One consideration that I have recently pondered is the prospect that the current ICC settings and formula is likely to be most attuned to find significant degree centrality magnitude changes (or differences) in widely distributed network hubs.  It would be nice to add the ability to constrain ICC analyses as a function of connectivity difference (i.e., analyze for short/local network ICC differences, as well as the current global ICC analysis).  I believe this functionality is something Alfonso may be considering (see here: https://www.nitrc.org/forum/forum.php?thread_id=9994&forum_id=1144).

Hope this helps,
Jeff