network-based statistic

AFNI Message Board
1

Hi AFNI experts,
I am wondering if AFNI has an implementation of this “network-based statistic”. It is similar to AlphaSim, but applied to “links” rather than voxels.
http://www.sciencedirect.com/science/article/pii/S1053811910008852

2

Zhihao,

Could you elaborate a little bit more about your specific analysis scenario such as response variable, explanatory variables, number of connections, etc? You can shoot me an email if you prefer.

3

Thank you Gang for your prompt response!
My scenario is very simple:

resting-state fMRI
2 groups, control and patient
anatomically derived 200 ROIs
200x200 functional connectivity matrix obtained individually (so 19900 links for each pair of ROI)
(1) link-wise group comparison between controls and patients; or group t-test performed for each one of the 19900 links
(2) link-wise correlation between link strength and a chemical concentration measured individually
Considering corrections of multiple comparison, I want to see which link (actually a group of links) is significantly changed between groups; or which link is significantly correlated with the chemical concentration.

4

I have some rough idea about handling the situation, which is hopefully better than what was presented in the paper you mentioned previously, but I’ll have to iron out a couple of sticky points before it can be applied to your situation.

5

Sure, please let me know when your iron has cooled down …
Thanks!

6

Hi, Dr. Gang Chen,

How about the implementation is going?

I just found this post when I was searching “network-based statistic (NBS) + AFNI”.

I am also looking for an alternatives for the NBS.

Since 2010, NBS was the mostly-commonly used method that provide a easy solution for network links statical comparison.

However, NBS has many limitations. Besides the way it corrected the multiple comparison, a major issue for me is that it didn’t support the “linear mixed model”.

This caused a problem for any longitudinal studies, when we want to model the with-subject variance as the random effect.

Really look forward to your new methods!

Cirong Liu,
cirong.liu@nih.gov
NINDS/NIH