dear fsl experts
this thread would like to go on a discussion over multiple comparison when correlating whichever score with RSNs activities.
in a previous post (dual regression: significant deactivation, group mean and masked analysis) I was confirmed that randomize is performed independently for each component.
so how should i correct my results for when i found voxels significantly correlating with that score in more than one RSN ??
if i found correlating voxels in two RSN, should i report (in both RSNs) only voxels > 0.975, if i want to report correlation in 3 RSN, only voxels with > 0.98333 ???
what happens If i want to report correlation among RSN and two different scores.
-1) If I create one single model with both the scores. it should be as I had one score, their specific contribute to the results is already calculated and automatically corrected for multiple comparison (applying the rule defined for a single score).
-2) I test two different models, i have to later correct for multiple comparison (requiring > 0.975 if two covariate and one RSN)..and so on. or....you ALWAYS have to use one GLM with one column of each score.
again in that post I hypothesized to perform one separate randomize for each RSN, using a different mask (the group mean of that RSN).
I was told that i cannot necessarly expect that my correlations might be included within the common areas of a RSN.
ok, but how could I discuss the meaning of a correlation if happens outside the classic voxel of that network ???
compared to e.g. VBM where voxels meaning are discussed as-is, in RSN analysis should I always interpret the voxels as part of a network, isn't it ? but if are outside that RSN ?? what happens
thanks in advance
regards
Alberto
|