Problem in Gev values in fitting microstates into individual data

[Raphael Fargier]

Dear Cartoolers,

I am experiencing some problems with the Gev values given by the fitting procedure with the last version of Cartool. The values are very low such as 0.01 on average for nearly 20 participants! This is observed for all maps even those for which we know the Gev should be high (around 30%). This is likely due to the last version of Cartool (3.55 3521) as I got entirely different values when running the same fitting procedure with an older version (3.51). Did anyone experience the same issue?
Is there something different in the fitting procedure between the two versions of Cartool that I have missed? Any help would be really appreciated!
Thank you!

[Jacobien van Peer]

Dear Raphael,

I have the same problem and I know someone else that experienced it as well. Unfortunately, I do not have a solution for it, other than using an older Cartool version, but I do not like that very much. It would be great if someone else has a better solution. Or have you found a solution in the mean time?
Best,
Jacobien van Peer

[Sebastian Moguilner]

I'm having the same issue.

[Denis]

Hi Sebastian,

I finally had the opportunity to investigate this potential GEV issue.

I checked the GEV values generated for both the Segmentation and the Back-Fitting processing, and for both ERPs and Resting States cases. I saw no errors whatsoever in the formulas nor the implementations, having tested different formulas and scenarios... Unless I missed something, everything is clear.

Let's see where potential misunderstandings could come from:

- GEV are values from 0 to 1, or in percentage 0 to 100%. Sometimes it is shown in % (verbose files), or in absolute (.seg files).

- The GEV is computed for all conditions together for a given subject. Ranges for GEV are usually in 50% to 75%.

- The .seg files however show the GEV values SPLIT per condition and per segment. This will significantly shrink the displayed values. Say a subject has a GEV of 0.75, for 4 conditions and 10 maps. This will make an average of about 0.02 GEV per cluster. Throwing in some variations per cluster, more maps, mor conditions, or all the above, we can realistically end up with numbers like 0.01, or even less.

- Finally, the latest version 5.01 has been upgraded to also output the GEV per subjects, numbers which were missing before. These are the real numbers to check, not the per-cluster ones.

- If the total GEV per subject is still very low, then there might be some questionning on the adequacy of the templates vs the data, but this needs the investigate the specificities of each case...

Hope this helps.

Denis

PS: I always recommend using the latest official version, code fixes happen all the time!

TL;DR: found no errors in the GEV computation; improved the verbose output while at it to show the GEV per subjects