Thanks a lot for the help and confirmation, Mark!
Gang
On Jul 18, 2011, at 12:09 PM, Mark Woolrich wrote:
> Hi Gang Chen,
>
>> Hi, I've run FLAME 1+2 using flameo on the terminal with and without outlier modeling for one group of subjects. The commands I used are:
>>
>> 1) No outlier modeling
>> flameo --cope=Beta.nii.gz --varcope=Var.nii.gz --mask=Mask.nii --dm=test.mat --tc=test.con --cs=test.grp --runmode=flame12
>> 2) Outlier modeling
>> flameo --cope=Beta.nii.gz --varcope=Var.nii.gz --mask=Mask.nii --dm=testOutliers.mat --tc=testOutliers.con --cs=testOutliers.grp --runmode=flame12 --inferoutliers
>>
>> The three files for option --dm, --tc, and --cs were generated with the FSL GUI. I have two questions:
>>
>> (A) What are the following two output files? Are the results from Stage 1? Why lower and upper?
>>
>> zflame1lowertstat1.nii.gz
>> zflame1uppertstat1.nii.gz
>
> These are the results of flame stage 1 and determine which voxels are passed to flame stage 2 (the MCMC sampling), ie. any voxels with zflame1lowertstat1<z_threshold< zflame1uppertstat1 are passed to stage 2.
>
>> (B) The analysis with outlier modeling took more than two times longer time than the one without outlier modeling, but the results without and with outlier modeling are identical. Does it mean that no outliers were identified in the analysis? Or something is wrong with my commands?
>
> The commands look fine. So yes it does look as though no outliers are identified. The approach is designed to be conservative, so that it does not deviate from the comfort of the standard non-outlier GLM without strong evidence. So it is not uncommmon for that to happen.
>
> Best, Mark.
|