Hi - yes I'm pretty sure this would be fine. FDR won't give the same
results as the (family-wise-error rate) corrected p-values, but it is
valid.
Cheers, Steve.
On 22 Jan 2009, at 08:10, Eric Kan wrote:
> Hi Steve,
> thanks for your reply. So unfortunately I do need corrected p values
> as you have guessed. And I also need to get around resource issues
> with my particular dataset+ randomise...
>
> would it be ok to take the uncorrected p values (tfce_p_tstatX
> files) that are generated and run them through a fdr to handle the
> multiple comparison correction?
>
> So something like this:
> 1) Chop up image, run randomise tfce
> 2) Recombine uncorrected pvalue volume back to single volume
> 3) Perform fdr correction on volume and threshold accordingly to
> output
>
> Would that work? I notice on the website they using the voxel-based
> thresholding in the example. Would fdr work for tfce as well or is
> it specific to vox analysis? (sorry my statistics is very shaky so
> hopefully this isn't too silly a question)
>
> Thanks!
> Eric
> ________________________________________
> From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf
> Of Steve Smith [[log in to unmask]]
> Sent: Wednesday, January 21, 2009 11:41 PM
> To: [log in to unmask]
> Subject: Re: [FSL] Randomise mask
>
> Hi - if you are just wanting uncorrected voxelwise p-values then
> that's ok. However you are most likely needing p-values that are
> corrected for multiple comparisons across all voxels, in which case
> I'm afraid that chopping up the images like this is not allowed,
> sorry!
>
> Steve.
>
>
> On 22 Jan 2009, at 01:58, Eric Kan wrote:
>
>> Hello,
>>
>> I have some TBM data that I have been trying to pass through FSL's
>> randomise
>> program to do the statistics. My progress is in the original thread
>> was
>> titled "Radomise_parallel with data generated outside FSL"
>>
>> Specifically the issue was that randomise_parallel was not running
>> through
>> to completion due to lack of adequate resources on our computer
>> nodes (each
>> TBM volume is ~20MB in size, 31 volumes total).
>>
>> As a workaround, I tried to chop up the volume into 11 separate rois
>> (via
>> -mask) to compute statistics individually, and then recombining each
>> piece
>> via fslmaths -add to get my whole volume again. This seemed to work
>> well
>> from the standpoint of getting the data to run through to completion
>> on the
>> compute nodes.
>>
>> I'm concerned however, that by arbitrarily chopping up the volume,
>> I'm
>> changing the statistical outcome when compared to analyzing the
>> volume as a
>> whole.
>>
>> Does anyone know if doing what I did changes the statistical outcome
>> of my
>> data?
>>
>> Thanks for any input you can give...
>>
>> I ran a threshold free analysis:
>> randomise_parallel -i <input 4D data> -o OUT -d design.mat -t
>> design.con -m
>> mask_(1-11) -n 1000 -D -T
>>
>> Eric
>>
>
>
> ---------------------------------------------------------------------------
> Stephen M. Smith, Professor of Biomedical Engineering
> Associate Director, Oxford University FMRIB Centre
>
> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> ---------------------------------------------------------------------------
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|