Differences in group analysis TFCE between CoSMo and FSL

[Jeanne Caron-Guyon]

Hi everyone, 

I have performed searchlight analyses on my dataset in 20 participants, followed by a group analysis using cosmo_montecarlo_cluster_stat, with 'cluster_stat' set to tfce. 

Everything works fine, so thanks for the great toolbox ! 

However, I also used FSL randomize function to perform the group analysis using TFCE as well, and I obtained quite different results. 

Obviously, the bigger clusters are found with both tools. However, there are some differences. Globally, CoSMo seems to yield more significant clusters (even small ones), and bigger clusters. Of course, I am here talking with the same statistics thresholding alpha=0.05.

When looking at the TFCE parameters used in FSL and CoSMo, it seems both use the same parameters, as defined as optimal by Smith and Nichols.. 

Would you happen to know why such differences can be found ? Are there some specifications is CoSMo/FSL leading to this ?

Thank you very much in advance, 

Best, 

Jeanne

[Tamás Andrei Földes]

I would also be interested in this question. Especially because I'm not quite sure how the "uncorrected" and "fwe corrected" TFCE maps FSL/randomise produces relate to the TFCE z-map that COSMO produces... is it accurate to say that COSMO tfce map is similar to the the uncorrected randomise map?