Hi,
Thank you all for your valuable insights. I just had a chance to go back and
try more things. It looks like it is not related to the computer cluster and
parallelization. I did two things:
1) I ran it again on the cluster and looked at each *SEED_tfce_p
and*SEED_tfce_corrp images. They all have constant p values in maybe 90% of
the tracts. All p values are 0.993333.
2) I ran it on a single computer using standard randomise command (without
parallelization). Again, all p values are fixed at 0.99875.
So, would you think that tfce is mishandling this? Both corrected and
uncorrected p values have the same maps. What would I lose/gain if I run it
without tfce?
kind regards,
Tugan
On Sat, 11 Apr 2009 08:59:13 +0100, Steve Smith <[log in to unmask]> wrote:
>Yes - sounds like something's gone wrong - with the TFCE option I
>wouldn't expect you to get the same p-values everywhere.
>If the SGE scripts didn't finish the cluster job on their own then at
>the least something's wrong with the cluster setup - I would suggest
>asking your sysad to look into why it didn't finish.
>Cheers.
>
>
>On 10 Apr 2009, at 23:40, Tugan wrote:
>
>> Hi Yi-Shin,
>>
>> Thanks for sharing your experience. What confused me is that the p
>> values
>> were fixed at 0.99952. So, in the p-value images it is either 0 or
>> 0.99952;
>> no other values in the entire image. And it is the same both in
>> corrected
>> and uncorrected. I had analyzed other data from other populations
>> before and
>> I had quite a range of p-values and they usually drop after FWE
>> correction.
>>
>> best,
>>
>> Tugan
>>
>>
>>
>> On Fri, 10 Apr 2009 18:16:34 -0400, Yi-Shin Sheu <[log in to unmask]
>> >
>> wrote:
>>
>>> Hi Tugan,
>>>
>>> Did you check the subject's FA skeleton in each group (young versus
>>> old)?
>>>
>>> For example, in the same area/voxel, does your FA skeleton appear
>>> to be
>>> uniformly higher/lower in the young group across the whole skeleton
>>> compared
>>> to the old group? Not necessary all subjects in the same group
>>> were that
>>> way, maybe just a few subjects have distinctively different FA
>>> value could
>>> cause the problem you described (significant p value across whole
>>> skeleton). Maybe you could make a testing ROI and extract the FA
>>> value from
>>> each subject's skeleton so that you can take a better look
>>> quantitatively.
>>>
>>> I asked you this because I have experienced similar problem recently.
>>>
>>> Yi-Shin
>>>
>>> ----
>>> Yi-Shin Sheu
>>> Research Assistant
>>> Developmental Biopsychiatry Research Program
>>> McLean Hospital / Harvard Medical School
>>> 115 Mill Street
>>> Belmont, MA 02478
>>> Tel: 617-855-2942
>>> Fax: 617-855-3712
>>>
>>>
>>> On Fri, Apr 10, 2009 at 2:28 PM, Tugan <[log in to unmask]> wrote:
>>>
>>>> Hi,
>>>>
>>>> Thank you for this quick response. Here's the line for randomise:
>>>>
>>>> randomise_parallel -i all_FA_skeletonised.nii.gz -o oldvsyoung -d
>>>> design.mat
>>>> -t design.con -m mean_FA_mask.nii.gz -n 2000 --T2 -V
>>>>
>>>> I did not use any threshold, I just used the --T2 option. I
>>>> actually copied
>>>> the whole randomise command directly from the TBSS page.
>>>>
>>>> I forgot to tell that I ran this on a cluster. All nodes ran fine
>>>> and
>>>> produced the *SEED* outputs, but not the final files. I had to run
>>>> the
>>>> defragment script manually. It seemed to work fine, as t maps
>>>> appear OK.
>>>>
>>>> thank you,
>>>>
>>>> Tugan
>>>>
>>>> On Fri, 10 Apr 2009 17:55:36 +0100, Reza Salimi <[log in to unmask]
>>>> >
>>>> wrote:
>>>>
>>>>> Hi Tugan,
>>>>> could you please send the line which runs *randomise* ?
>>>>> Maybe you are using cluster-based inference with a very small
>>>>> cluster-forming threshold, which results in a huge significant
>>>>> cluster,
>>>>> Cheers
>>>>>
>>>>> On Fri, Apr 10, 2009 at 5:47 PM, Tugan <[log in to unmask]> wrote:
>>>>>
>>>>>> Dear all,
>>>>>>
>>>>>> I am hoping someone can shed a light on what I observed with a
>>>>>> data I
>>>> have
>>>>>> been working on. It is a simple set of DTI-FA maps of old and
>>>>>> young
>>>> people
>>>>>> and I run TBSS on the data. The data is unbalanced, I have 36
>>>>>> olds and
>>>> 18
>>>>>> youngs. After running randomize, p values (both corrected and
>>>> uncorrected
>>>>>> p-images) are constant 0.99952 almost throughout the whole WM
>>>>>> tracts. On
>>>>>> the
>>>>>> other hand, t-statistics show variance across space, so they look
>>>> normal.
>>>>>> Can anyone speculate why I am getting a fixed p value throughout
>>>>>> the
>>>> entire
>>>>>> WM tracts? Moreover, it is the same for both corrected and
>>>>>> uncorrected.
>>>>>>
>>>>>> I browsed through the email discussions but I could not find a
>>>>>> similar
>>>>>> issue.
>>>>>>
>>>>>> Thanks in advance for any insights you might have.
>>>>>>
>>>>>> best,
>>>>>>
>>>>>> Tugan Muftuler
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> G. Salimi-Khorshidi,
>>>>> D.Phil. Student, Dept. of Clinical Neurology, University of Oxford.
>>>>> [log in to unmask]
>> http://www.fmrib.ox.ac.uk/~reza<http://www.fmrib.ox.ac.uk/%7Ereza>
>>>>> 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
>---------------------------------------------------------------------------
|