Hello, 

I am new to CONN and as a way of double-checking my understanding of how CONN is calculating my results (specifically the z-matrices) I have been trying to conduct an "apples-to-apples" comparison between using SPM & REX and CONN. These are resting state data only. I preprocessed my data in SPM and used REX to extract the timeseries for 15, 5mm-spherical ROIs. I then ran my correlations and converted my Pearson coefficients to Fisher's Z. I ran the same steps as closely as I could in CONN. I selected the smoothed images instead of unsmoothed as is suggested in order to keep the comparison consistent. In the end, my correlation matrices are quite different. 

When I then go in to the REX output files in CONN to extract the raw timeseries they are similar (but not exact) to what I have extracted using SPM & REX. When I open the data/ROI_Subject*_Session*.mat file and extract the timeseries for each ROI there, they are vastly different (measures are on the order of 5-50 for each timepoint, as compared to 5000-10000 from REX output). I know that the "data/ROI..." file contains denoised data and REX contains raw data, but I'm perplexed as to why these values are so vastly different. Is there an additional step or calculation that I am unaware of? Do you have any recommendations as to how to conduct a more similar comparison between these analytic approaches to help assure we are indeed performing the same operations?

I greatly appreciate any feedback you can offer.

Many thanks,
Jenna

Hello Jenna,

Two quick points which might help:

• For ROI-ROI connectivity, Conn uses unsmoothed images instead of smoothed ones; this is done to prevent signal within an ROI from being affected by signal outside
• Conn converts the time series into percent signal change using the global signal as the reference. This is the default option (user can change it into raw signal in the setup tab); have a look at this thread: https://www.nitrc.org/forum/message.php?... for details on how Conn handles global signal calculation, specifically at the modifications that Dr. Alfonso mentions in his replies

Can you try using REX on unsmoothed images and converting it into PSC?


Best
Pravesh