Hi Steve,
This makes sense, and would explain why it was hanging at different tstats for different
input datasets. And that's a good point that if your search space is restricted enough to
begin with, then TFCE is not needed. Thanks for the response!
John
On Sat, 8 Nov 2008 10:27:44 +0000, Steve Smith <[log in to unmask]> wrote:
>I think I can guess what's happening - there's a line in randomise
>that estimates the TFCE height increment ("delta") as a one hundredth
>of the maximum tstat value - so it's possible that if you have a very
>restricted input mask, this might possibly end up as zero, in which
>case the TFCE height looping might run forever.
>
>We will check on that and let you know next week; but, for now, I
>suspect that if your mask is VERY small then you don't need TFCE
>anyway, and you could just run with normal corrected voxelwise p-values.
>
>Cheers.
>
>
>On 8 Nov 2008, at 00:10, John Colby wrote:
>
>> Hi Reza,
>>
>> Thanks for the response! By the way I meant to be talking about L1
>> in my original email,
>> and not MD. I checked the mask and it was OK, but I recreated it
>> anyway and tried
>> running with the --T2 options specified through -T like you
>> suggested below. Same
>> thing...Randomise with TFCE gets stuck on permutation 2.
>>
>> Out of curiosity I tried running some other combinations and got
>> some strange results:
>>
>> - MD: Runs fine with the 0.95 thresholded 1-p mask (made from the
>> original FA contrast -
>> see original message).
>> - L1: Does not run with the 0.95 thresholded 1-p mask (hangs at
>> permutation 2 of tstat1).
>> Does run through properly when I threshold the mask at 0.949 or
>> anything lower.
>> - RD (L2+L3)/2: Does not run with the 0.95 thresholded 1-p mask
>> (hangs at permutation
>> 2 of tstat*2*). Does run through properly when the mask is
>> thresholded at 0.89 (but not
>> higher).
>>
>> All of these run fine if I use the full mean_FA_skeleton mask, or
>> try one of the other
>> modes (cluster-mass). I've also tried recreating the data files with
>> the tbss_non_FA
>> script, but no improvement with that either. I do, however, recreate
>> the exact same
>> problems when I try to run the analyses on my local machine (x86
>> 64bit MacPro) instead
>> of our cluster (x86 64bit Linux running SGE).
>>
>> Any ideas? Is there any way I can get additional debugging output
>> from randomise? As it
>> is, things just silently hang on "permutation 2".
>>
>> Thanks again for the help!
>>
>> John
>>
>> On Mon, 27 Oct 2008 22:12:23 +0000, Reza Salimi
>> <[log in to unmask]> wrote:
>>
>>> Hi John,
>>> it shouldn't be like that,
>>> could you please check your mask and if it's allright, just rerun the
>>> analysis with -T option
>>> (with --tfce_H=2 --tfce_E=1 --tfce_C=26)???
>>> thanks, Reza
>>>
>>>
>>>
>>> On Mon, Oct 27, 2008 at 9:51 PM, John Colby <[log in to unmask]>
>>> wrote:
>>>
>>>> Hello all,
>>>>
>>>> I'm doing a 50 subject TBSS project in FSL 4.1.0 and was hoping to
>>>> use a
>>>> TFCE mode
>>>> FWE-corrected p<0.05 significant region from the whole-brain FA
>>>> analysis as
>>>> a mask for a
>>>> subsequent investigation looking at MD. When doing this, randomise
>>>> runs
>>>> properly in the -
>>>> C cluster mass mode, but simply hangs on permutation 2 when I try
>>>> to run it
>>>> in the --T2
>>>> TFCE mode.
>>>>
>>>> When I expand my mask to p<0.06 or higher, however, --T2 TFCE also
>>>> works
>>>> fine. Which
>>>> is why I'm curious, is there a "minimum number of voxels"
>>>> requirement wired
>>>> into TFCE?
>>>>
>>>> Thanks!
>>>>
>>>> John Colby
>>>>
>>>> MD/PhD student
>>>> Developmental Cognitive Neuroimaging Lab
>>>> David Geffen School of Medicine at UCLA
>>>>
>>>> Commands/outputs:
>>>>
>>>> Masks generated like:
>>>> jcolby>> fslmaths FA_tfce_corrp_tstat2 -thr 0.95 -bin
>>>> FA_tfce_corrp_tstat2_95_bin
>>>>
>>>> p<0.05 mask (186 voxels):
>>>> jcolby>> randomise -i all_MD_skeletonised -o MD -m
>>>> FA_tfce_corrp_tstat2_95_bin -d
>>>> design.mat -t design.con -n 1000 --T2 -V
>>>> Loading Data: *****************************************************
>>>> Data loaded
>>>> 7.52118e+21 permutations required for exhaustive test of t-test 1
>>>> Doing 1000 random permutations
>>>> Starting permutation 1 (Unpermuted data)
>>>> Starting permutation 2
>>>> (hangs here forever...)
>>>>
>>>> p<0.06 mask (625 voxels):
>>>> jcolby>> randomise -i all_MD_skeletonised -o MD -m
>>>> FA_tfce_corrp_tstat2_94_bin -d
>>>> design.mat -t design.con -n 1000 --T2 -V
>>>> Loading Data: *****************************************************
>>>> Data loaded
>>>> 7.52118e+21 permutations required for exhaustive test of t-test 1
>>>> Doing 1000 random permutations
>>>> Starting permutation 1 (Unpermuted data)
>>>> Starting permutation 2
>>>> Starting permutation 3
>>>> Starting permutation 4
>>>> Starting permutation 5
>>>> Starting permutation 6
>>>> Starting permutation 7
>>>> Starting permutation 8
>>>> Starting permutation 9
>>>> Starting permutation 10
>>>> Starting permutation 11
>>>> Starting permutation 12
>>>> Starting permutation 13
>>>> (etc..., finishes properly)
>>>>
>>>
>>>
>>>
>>> --
>>> G. Salimi-Khorshidi,
>>> D.Phil. Student, Dept. of Clinical Neurology, University of Oxford.
>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~reza
>>> FMRIB Centre, John Radcliffe Hospital,
>>> Headington, Oxford, OX3 9DU
>>> Tel: +44 (0) 1865 222466 Fax: +44 (0)1865 222717
>>>
>>
>
>
>---------------------------------------------------------------------------
>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
>---------------------------------------------------------------------------
