Hi,
I was wondering how the toolbox calculates / defines cluster. And if you implemented any correlation magnitude or clustersize threshold.

Furthermore I am interested in the question how you correlate the timeseries.

I couldn't find anything about that in the Manual or FAQ, so maybe you can tell me where to find these information.

Thanks!

Hi Sarah,

We use the same method used by SPM in standard functional analyses (Friston et al. 1993, Assessing the significance of focal activations using their spatial extent) to define clusters and their associated significance values (this is used to define uncorrected and FWE-corrected cluster-level significance values, in addition we use the method described in Chumbley et al. 2009 to derive the FDR-corrected cluster-level values). Briefly, for example for a seed-to-voxel analyses looking at the functional connectivity between a single seed and every voxel in the brain, we first use a 'height threshold' (e.g. uncorrected p<.001) to select only supra-threshold voxels where the correlation between the seed BOLD timeseries and this voxel is significant at this level. Then we evaluate the size of the resulting clusters (connected sets of voxels) and test those sizes against the distribution of expected sizes from the null hypothesis to compute cluster-level p-values, and use a second threhsold ('extent threshold', e.g. FWE-corrected p<.05) to only keep those clusters that survive this second threshold.

We will be having a manuscript out hopefully soon describing in more detail the analysis steps of the toolbox, in the meantime perhaps you might want to take a look at Chai et al. 2011 (Anticorrelations in resting state networks without global signal regression) for a more detailed description of some of these steps (or check some of the references in http://www.alfnie.com/software/conn/user... for examples of use if you think that might help). 

Hope this helps
Alfonso

Originally posted by Sarah Buetof: