Dear All,

In this post

http://www.nitrc.org/forum/message.php?m...

Alfonso wrote that "ICC is sign-independent (since it considers r-squared values), while global correlation is not (it averages the actual r correlation-coefficient values, considering their sign, so positive and negative correlations will cancel out)".

I`m interested in using global correlation and had a look at the formula (please see attachment) provided in the paper "Conn: a functional connectivity toolbox for correlated and anticorrelated brain networks". Now I`m confused. Doesn`t this formula use absolute values which would mean that positive and negative values are aggregated and not canceled out?

Thank you very much in advance

Dear Hannes,

Yes, the formula there is the one for ICC (in that manuscript CONN still did not have GCOR implemented so we were using somewhat carelessly "global correlation strength" and "intrinsic connectivity contrast" interchangeably). To clarify, both "intrinsic connectivity" (ICC)  and "global correlation" (GCOR) are measures characterizing aspects of the global pattern of connectivity between each voxel and the rest of the brain. Their mathematical definitions are similar but not identical, where GCOR computes the average of the correlation coefficients while ICC computes the average of the squared correlation coefficients. Because of these similarties their results are actually related by the formula:

   ICC(x) = GCOR(x)^2 + VAR(x)

where ICC(x) = average { r(x,y)^2 }  (the average squared correlation values between the seed-voxel x and the rest of the brain)
            GCOR(x) = average { r(x,y) }   (the average correlation values between the seed-voxel x and the rest of the brain)
     and VAR(x) = variability { r(x,y) }  (the spatial variability in connectivtiy values between the seed-voxel x and the rest of the brain)

Hope this helps clarify
Alfonso 
Originally posted by hannes berg:

That was helpful. Thank you very much!

.