I am conducting a resting-state functional connectivity seed-to-voxel analysis on the 25 participants from the NYC test-retest dataset using the CONN toolbox with the Posterior Cingulate Cortex of the Default Mode Network as a seed region.
I am using a liberal uncorrected voxel-level "height" threshold of p < .001 with a p-FDR corrected cluster-level "extent" threshold of p < .05, which is recommended when the expected effects are distributed/broad but weak as opposed to focal and strong (when a p-FDR- or FWR-corrected voxel-level "height "threshold is recommended instead) according to the developer of the CONN toolbox (https://www.nitrc.org/forum/forum.php?th...).
I have searched for literature on which those recommendations are based but could only find tangential reasons and nothing concrete (of course, having a more liberal voxel-level threshold would allow one to detect weaker effects because statistical power increases for the same sample size). If anybody has a reference for those specific recommendations, I would be grateful if they can share it because the developer of the CONN toolbox has expressed them in several posts so they must be some sort of standard.
My first question relates to the appropriateness of this thresholding convention for my particular case.
While seed-to-voxel connectivity during resting state can be considered distributed when the analysis is carried out on the whole brain and no a-priori predictions are made because large-scale networks (e.g., default mode network)are identifiable during rest, I am uncertain about connectivity during resting state being necessarily "weak" as the recommendations assume.
The reason for this doubt is the knowledge that the BOLD signal during a task-related activity is very small compared to the noise, with up to 80% of the BOLD modulation being discarded as noise (1st reference below). On the other hand, precisely the low-frequency spontaneous fluctuations in the BOLD signal that are discarded as noise in task fMRI are taken as signals of interest in resting-state fMRI.
From this, I would expect stronger effects during rest vs. task, giving roughly a 80:20 ratio. So maybe in my case assuming that effects are weak is unwarranted unless I am missing something (e.g., even though activations might be stronger during rest, correlations might not necessarily also be stronger).
My main question (assuming that I have used appropriate thresholding criteria) relates to whether I should report ALL anatomical areas within a significantly identified cluster when writing a paper. The reason I am asking is because the most significant cluster (out 16) in my analysis spans over thirty atlas-defined anatomical areas and its size is 50 574 voxels, covering a huge chunk of the neocortex.
Is there any convention about reporting areas within a cluster - do I report all or just, for example, the first five? I did come across a recommendation only to report the general area (e.g., left hemisphere) when the cluster is large because it is not clear which anatomical regions are truly correlated and which are false positives (2nd reference below).
The problem with that is that I do not really know how to call this huge chunk of the neocortex (e.g., just forebrain?) and from prior literature, we know which regions should be significantly negative/positively correlated with the PCC, which makes me think that my threshold is too liberal.
Do I just report them like this: cluster one (MNI of most active voxel, size, cluster-level extent p-value), etc., without referring to the anatomical location of the clusters themselves?
Smitha KA, Akhil Raja K, Arun KM, et al. Resting state fMRI: A review on methods in resting state connectivity analysis and resting state networks. Neuroradiol J. 2017;30(4):305‐317. doi:10.1177/1971400917697342
Woo, C.-W., Krishnan, A., & Wager, T. D. (2014). Cluster-extent based thresholding in fMRI analyses: Pitfalls and recommendations. NeuroImage, 91, 412–419. doi:10.1016/j.neuroimage.2013.12.058