Network analysis: Correcting for ROI size

[Chris Whelan]

Hi Alexander et al,

I used ExploreDTI to generate a series of connectivity matrices as part of my PhD thesis. One of the major criticisms during my PhD defense was that I only used 'fiber count' as a network weighting in these matrices, without correcting for the volume of each region of interest in the AAL atlas. 

However, I do not believe there is any easy way to correct each connectivity matrix for ROI size... there are no 'standard' volumes associated with the ROIs in the AAL atlas.

Does anyone on these message boards have a solution to this issue?

Many thanks in advance,

Chris Whelan

Post Doc @ University of Southern California 

[stijn m]

Dear Chris,

This is an interesting topic to discuss about and I'd like to hear Alexander's point of view.

It might be wise to correct for volumes associated with the ROIs. One dataset might have a bigger insula then another for instance, having a possibly higher chance having more fiber tracts starting/ending or passing through this ROI. Taking the intracranial volume (ICV) into account, you can correct for the extend of the specific ROI's generated from the AAL atlas. This is some kind of normalization, because the volumes might differ between datasets.

Using "fiber count" does take some network weighting into account, but not really corrects/normalizes your AAL ROIs.

Cheers,
Stijn

[Chris Whelan]

Thanks for your reply, Stijn! 

I have been in touch with Alexander (via email) and thankfully, ExploreDTI already produces a measure that corrects for ROI size and tract length, known as 'connection density / density weight'. 

This measure is based on a paper published by Patricia Hagmann et al (PLoS Biol., 2008), where they define ‘connection density’ as follows…

connection density = 

…this formula corrects for fiber length – l(f) - and also corrects for variability in the size of cortical ROIs – Sv + Sv – before capturing the number of fiber connections per unit surface. Just what I needed!

Thanks,

Chris