I think the UCSF group led by Bill Seeley has been looking at this, trying to covary for local GM volumes when analysing the resting-state data from their patients with semantic dementia (see Guo et al., Brain 2013; doi: 10.1093/brain/awt222 ).

A Canadian group (see Farb et al., Cortex 2013; doi: 10.1016/j.cortex.2012.09.008 ) has also looked at resting state in patients with SD and bvFTD and they actually used BPM to correct their fMRI differences using the patients VBM-derived gray matter maps.

Overall, I guess the correction need to be applied at the regional level (or voxel level if you are performing voxel wise analysis) correcting for global GM would not be relevant if you are considering regional connectivity metrics.

Hope that could help.

Thank you for your answer. I completely agree with you, I recently applied Regional Homogeneity index in Frontotemporal Dementia patients related to Granulin mutation using Biological Parametric Mapping to take into account gray matter atrophy in the statistical model. With this approach, I obtained a pattern of local connectivity alteration quite similar to VBM analysis in this group (frontal, temporal and parietal asymmetric atrophy). However, it is quite interesting that without correction, only posterior regions (parietal and precuneus) were impaired.

Interesting.

Actually, it's pretty hard to predict how atrophy is going to induces changes in the resting-state signal; the idea that atrophy just reduces the connectivity values might be over simplistic. The problem is that atrophy certainly increases the noise in the measurements : fMRI voxels are usually big (3x3x3mm) and in patients with bvFTD or SD have major atrophy and their cortical GM gets so thin that fMRI voxels get really subject to important partial volume effects. I'm not sure we can really perfectly deal with that but looking at results with and without BPM seems like a very good first step!

Good luck with your study