Dear Max,

Yes, that is perfectly valid and reasonable approach. Conceptually this procedures has a little bit of an "asymmetry" in how averaging across all voxels within a mask is performed. In particular, the seed/mask timeseries is going to average the original BOLD signal timeseries across all voxels within your mask (this happens during the Setup step when the BOLD signal within the mask is extracted), then you are going to compute the correlation between this average-BOLD-timeseries and each individual voxel's BOLD-timeseries (this happens during your seed-based analyses), and last you are going to average those correlation coefficients (this happens during the "import values" step; note that here I am disregarding the effect of the Fisher-transformation of correlation coefficients for simplicity). An alternative procedure without this "asymmetry" would be to compute all pairwise correlations between all of the voxels within your mask, and then average the resulting correlation coefficients. It turns out that there is a very nice mathematical identity between these two procedures (assuming that the variances of the BOLD timeseries are equal across all voxels within the mask) which states that r1 (the result in procedure 1) and r2 (the result in procedure 2) are directly related as r1^2 = r2. How this identity breaks in the case of unequal voxel variances also helps clarify what the actual effect of this asymmetry is: mainly that when the BOLD timeseries are not standardized (i.e. if different voxels have unequal variances) then the average BOLD signal computed by the first procedure will be "more-influenced" by the voxels with higher variance compared to voxels with lower variance and the resulting r1^2 will always be lower than r2.

In any case, all of this to say that this your approach and implementation are both perfectly reasonable. If you are concerned about this asymmetry or about the potential of the resulting within-network connectivity measure possibly over-emphasizing voxels with higher BOLD signal variance, you may want to try the second approach instead. If you want to do that in CONN you could use the following syntax to compute those average correlation coefficients (averaged across all pairwise voxel-to-voxel connections):

R = conn_vv2rr('mask.nii');

That will output a [1 x 1 x Nsub x Ncond] matrix containing those average within-network correlation values separately for each subject and for each condition (those conditions defined in Setup.Conditions). Note that this will only work for MNI-space mask files (I have not yet implemented a similar procedure for subject-specific ROIs or ROIs that are defined on a different space from the final functional data imported into your CONN project). 

Hope this helps

Alfonso

Originally posted by max345:

Dear all,

I think I've found a way to calculate functional connectivity using seed-based analysis within a network mask in CONN, and would like to know if you can valide my approach described below:

1.) I imported a mask (based on the ROIs implemented in CONN, for example, for the DMN) as an ROI in SETUP.ROIs.
2.) Then, I proceeded with preprocessing and denoising the data.
3.) Next, I defined a seed-based analysis with this imported mask as the ROI/seed of this analysis in the ANALYSES (1st-level) tab.
4.) In the RESULTS (2nd-level) tab, I opened the second level analysis results window and clicked on "import values". In the newly opened window, I selected the option "others clusters of interest or ROIs (select exported mask/ROI file)" and selected my network mask (the same one I originally imported as an ROI in SETUP.ROIs) again. 
5.) Now, I have one functional connectivity value for each person in SETUP.Covariates (2nd-level), which I can extract for further analysis outside of CONN/SPM.

Can you confirm my procedure described above as a method for extracting functional connectivity values within a network mask?

Many thanks and best,
Max