Dr. Smith,
Thank you so much for your clarifications and your quick replay.
Best Regards,
Angelica Hotiu
On Tue, 3 Mar 2009 07:25:42 +0000, Steve Smith <[log in to unmask]>
wrote:
>Hi,
>
>On 3 Mar 2009, at 06:19, Angelica Hotiu wrote:
>
>> Dear FSL experts,
>>
>> I was running TBSS analysis using DTI data acquired from 5 control
>> subjects
>> and 3 subjects with mild traumatic brain injuries. I used FSL4.1.2
>> version
>> for all preprocessing of the data including TBSS analysis.I am
>> interested
>> to identify the regions where FA, MD and transversal diffusivity
>> indicate
>> differences between the two groups. For randomisation procedure I
>> used
>> first TFCE option .As a result FWE corrected shows the results only
>> for MD.
>> I couldn't obtain FWE corrected for multiple comparisson for FA and
>> transversal diffusivity using TFCE option. Also I was trying another
>> option,
>> cluster-based thresholding corrected for multiple comparisons using
>> different values for threshold .For FA and threshold c = 4, the p
>> image,
>> tbss_clustere_corrp_tstat2 ,has a p values corrected above 0.928
>> which is
>> not in the range 0.95-1.Also for a threshold of 0.3 the corrected
>> values is
>> in range 0.95-1.
>> 1)I'm wondering if this threshold value of 0.3 is valid or is to low?
>> Is it
>> possible a way to calculate the thershold?Is it a paper or manual, to
>> clarify, I'm sorry for missing .
>
>It is valid, because randomise will generate valid corrected p-values
>regardless of the cluster-forming threshold chosen. However, it is not
>normal to set such a low cluster-forming threshold, and you may find
>that some reviewers don't appreciate that it is still valid, and
>become unhappy! Presumably it results in very large clusters and it
>is hard to interpret any useful spatial localisation of effect - in
>fact you may simply be testing for whether there is a global shift
>between the two groups.
>
>There is no way to "calculate" the threshold - it is an arbitrary user
>choice - see discussion on this in the TFCE paper that's just come out
>in NeuroImage.
>
>> Also I'm wondering if the output of TFCE
>> uncorrected data can be fed into FDR correction.
>
>Yes it can - that is valid.
>
>
>> 2)If yes, could you please take a look to see if it's correct:
>>
>> fslmaths tbss_tfce_p_tstat2 -mul -1 -add 1
>> tbss_inv_tfce_p_tstat2
>> fdr -i tbss_inv_tfce_p_tstat2 -m tbss_tstat2 -q
>> 0.05.
>
>Indeed; see
>http://www.fmrib.ox.ac.uk/fsl/randomise/fdr.html
>
>>
>> 3)I'm not sure what mask image is in -m option.
>
>You should use the same mask that you fed into randomise, i.e.
>mean_FA_skeleton_mask
>
>> 4)Could I apply FDR correction to output from cluster based
>> thresholding ?
>
>No, randomise doesn't give uncorrected p-values for clusters, just
>corrected.
>
>Cheers.
>
>> 5) Is it recommended in this case to randomise using
>> cluster-based thresolding corrected for multiple comparison using
>> cluster
>> mass -C option? What should be the values for C in this case?
>> Any suggestions will be greatly appreciated.
>> Many thanks in advance.
>> Regards,
>>
>> Angelica
>>
>
>
>--------------------------------------------------------------------------
-
>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
>--------------------------------------------------------------------------
-
|