Print

Print
Hi Yacila,

You'd use 0.95 and all the 8 tests.

For merger and splitting, use whichever tool you are more familiar with. You need to merge along with a spatial dimension (x, y or z) not the 4th dim (not t), hence fslroi is probably better suited to separate later (you have to count exactly the number of voxels to start and end).

All the best,

Anderson


On 15 May 2017 at 05:54, Yacila Isabela Deza Araujo <[log in to unmask]> wrote:

Hi Anderson,

Thanks a lot for your answer. Just a short follow-up question:

If I correct for 8 tests globally,  should I merge the uncorrected TFCE maps from the 8 networks in one direction? What about the 2 tests?

Then, is a threshold of  0.95 (1-p) enough? Or should I correct for the tests here (e.g 0.975)?

Finally, instead of fslroi, can I use fslsplit? (I do not have idea how to use fslroi)

 

Sorry for these basic questions and thanks again.

Yacila

 

Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im Auftrag von Anderson M. Winkler
Gesendet: Sonntag, 14. Mai 2017 15:10
An: [log in to unmask]
Betreff: Re: [FSL] FDR and TFCE

 

Hi Yacila,

Please see below:

 

On 12 May 2017 at 20:21, Yacila Isabela Deza Araujo <yacila_isabela.deza_araujo@tu-dresden.de> wrote:

Dear FSL experts,

I was asked to perform multiple corrections for my test and networks. Since the FWE and Bonferroni are a very conservative combination,

 

You wouldn't use FWE and Bonferroni -- that would mean correcting twice. Bonferroni is one type of FWE correction, and it requires that the tests are independent, otherwise it becomes conservative. Not all types of FWE correction are conservative. The one used in randomise, for example, is exact.


 

I was suggested to use FDR instead.
I have never used it, but as far as I read, it should be performed on the T maps obtained from randomise, for example.

 

FDR requires as input the uncorrected p-values.

 

My question is: If I use FDR to correct for 2 test and 4 networks, which threshold should I use and where?

 

You may want to correct each test individually, or correct globally the 2x4=8 tests. To correct individually, simply run the command "fdr", then thresholding the adjusted p-values at the desired level (usually 0.05, or 0.95 if using 1-p maps).

To correct all 8 maps with the command "fdr", the procedure is a bit more involved: merge them all along one spatial dimension (e.g., x), apply the command, then separate using "fslroi".

 

Is is possible to use this threshold together with TFCE?

 

It is possible to use the command "fdr" with p-values for the TFCE statistic.

All the best,

Anderson

 


I would be really happy if you have any suggestion.
Kind regards

Yacila