Network features are all same in brain and random networks

Dear Mrtrix3 experts,

I have finished seeding fibers and the tractography was biologically apparent. I did not have any troubles with tck2connectome even. After getting connectivity matrix, I did some thresholding to get a desire network density. Then I compare the brain network to about 100 random networks which were generated from the brain network by preserving degree distribution. However, network features like global efficiency turned out to be same in both brain and random networks which is strange to me. I know this question might not be relevant to mrtrix3 usage; however, I had checked all of my calculations but could not find anything wrong. I wonder if there were any issues in terms of seeding fibers that lead to this situation.
So far, I used ACT method with 5tt image and seed_dynamic to reconstruct fibers.
I am truly appreciated for your help.

Best,
Van

Hi,

What was your network density? My understanding is that if you are using a very dense network, the network itself will be very similar to the random network (this is even worst if you are using binary matrices), so then some of the global metrics will be very similar.

There are a lot of works reporting this kind of analysis, for example here. If you look at the figure 3, you can see how γ = C/Crand and λ = L/Lrand for a densities higher than 0.5 are almost identical and the quotient is 1.

I would like also to hear other opinions in this topic.

I hope this helps,

Best regards,

Manuel

Hi,
I used 0.4 for my network density.
In fact, I measured network features at different level of densities ranging from 0.05-0.5 with step side equals to 0.05. Differences were observed at density 0.1 However, I am afraid that there were many connections lost in the way.
Thank you so much,

Jane

We published not too long ago our thoughts on removal of weak connections from structural connectome data.

I did some thresholding to get a desire network density.

For clarity: are you explicitly thresholding to produce a strictly binary connectome, or are you simply pruning, i.e. setting weak connections to zero but retaining the quantified connection weights for those connections above the threshold?