Print

Print
Hi Carole,

I will let someone else more qualified answer the main question, but just wanted to point out one subtle-but-important clarification. I believe TFCE is not an independent stage after randomise, but is performed within randomise on each permutation. This is because it is performing permutation tests on the actual TFCE statistic at each voxel, which itself doesn't have a good closed-form solution or derivability directly from the voxel-wise p/t maps.

I don't THINK there's any reason you can't use FDR on the TFCE voxel-wise p maps (tfce_p_tstat), but TFCE does weird things to the indpendence of voxels, so I would wait for advice from someone more statistically savvy. (e.g. you might need to use the --conservative option to account for correlations).

Best,
Mike



On Mon, Feb 16, 2015 at 1:05 PM, carole Guedj <[log in to unmask]> wrote:
Dear FSL experts,

If I understand well (and please, fell free to correct me if I’m wrong) what’s happened in the last stage of dual-regression (i.e. the ‘randomise’ part), the program use a permutation methods to make inferences (thresholding maps) about hypotheses (matrix design and contrast) => this leads to voxel-wise uncorrected p-values maps (‘_vox_p_tstat’)
Then there are 2 steps:
1/ the TFCE test, enhancing cluster-like structures =>  ‘_tfce_p_tstat’
2/ the Familly-wise error rate correction, correcting maps for multi-comparison  =>  ‘_tfce_corrp_tstat’


In my case, I think that the FWE correction is too conservative for my data, so I would like to replace it by FDR correction, but I’m wondering if it make sense to do it after the TFCE test (instead of vowel-wise images, because I’m interesting in cluster-base structures).

Thank you in advance for any reply!

Best

Carole



--
Michael G. Dwyer, Ph.D.
Assistant Professor of Neurology and Biomedical Informatics
Director of Technical Imaging Development
Buffalo Neuroimaging Analysis Center
University at Buffalo
100 High St. Buffalo NY 14203
[log in to unmask]
(716) 859-7065