We have a set of 3dmvm-generated results limited to cortical grey matter (per a mask) and are interested in seeing if the full set of significant clusters occur above chance in particular intrinsic functional connectivity (IFC) networks--e.g., Does a set of clusters where we observe significantly altered functional connectivity with a default-mode network seed region as a function of polygenic risk for major depression overlap above chance with somatomotor network?
It's easy enough to get estimates of overlap between our results and different IFC networks. It's more challenging to do Monte Carlo simulations assuming a uniform spatial distribution to generate a null hypothesis model to test the likelihood of the overlap of a set of N clusters with sizes i, j, k... with a given set of IFC networks. It would be easy if all we needed to do was generate a set of N appropriately sized spheres, but to meet assumptions, we need to generate sufficiently complex blob-oids that fall within our grey-matter mask.
This all sounds so close yet so far away from 3dClustSim that we wonder if you might have something within dancing distance of our needs that we might not be aware of? The feature in question is one of those that strikes me as conceptually simple and implementationally difficult.
Indeed, running 3dClustSim would make most sense in a close-to-sphere, blobby object like a brain mask; doing so in a restricted, skeletony GM mask or further shrunk network mask is tricky.
I guess my initial thought we be to use 3dClustSim, still, on the whole brain mask or at least on an inflated GM mask, and see if/where any clusters fall in your data, and then evaluate which networks those are in. While admittedly not exactly what you asked for, the benefits of this are:
that you only run 3dClustSim once, rather than having a multiple comparisons issue across multiple networks
cluster simulation (or random field theory, too, methinks) will have more consistent assumptions in a sphere-like mask region rather than in complicated skeletony things.
the noise in the collected data know no boundaries or networks in the data---noise smoothness patterns are just noise smoothness patterns across the brain, they don't know about underlying networks
if the data have been smoothed during processing (which I assume is the case for voxelwise analysis), information has already been smeared across strict network delineations, anyways
So, if I'm picking up your trail here, you recommend using 3dClustSim, probably with the option: -ssave:TYPE ssprefix and then I can check out the incidence of clusters within these volumes with various network masks.
I'm betting this is the best option because of the NB for -ssave option: "This option will slow the program down a lot, and was intended to help just one specific user."
--The Paulest of Them All (but only here in Norway where there are no other Pauls)
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.