Dear experts,

I use the CONN toolbox to generate ROI-to-ROI correlation matrix for each participant. Subsequently, I wrote a script to calculate various global and nodal topological properties with the BCT toolbox. 

I have a longitudinal dataset (baseline and follow-up) and I normalized both sessions to the same session (e.g. to the baseline T1 images) now (as discussed in this thread: https://www.nitrc.org/forum/message.php?msg_id=39627 ).

I understand that normalizing both sessions to the same t1 template can result in more reliable comparisons between time points. 

However, my primary focus is on graph theory analysis (e.g. clustering coefficient (CC), nodal strength, etc.) rather than longitudinal changes in the brain map. I'm wondering whether normalizing each session to its own t1 images might provide a more accurate ROI-to-ROI correlation matrix, given that my analysis does not directly compare seed-based FC changes over time.

In addition, I recently came across some studies reporting weird changes in small-worldness attributes. For example, one study reported increased CC, decreased characteristic path (Lp), and decreased small-worldness in MCI patients. Since the formula of small-worldness is "normalized CC/normalized Lp", I would have expected small-worldness to increase under such conditions, rather than decrease. Could this discrepancy be related to the algorithm used for generating random networks? I'm curious whether certain algorithms may introduce biases, leading to these unexpected results.

Best regards,

Chih-Hao