Hi,
That's right.
The low part of the colourbar now represents low p-values which are
*more* significant that the high values.
Also note that you see all p-values (i.e. uncorrected and unthresholded)
with this. If you want to only see those that survive the FDR correction
then you should make sure the range stops at the p-value threshold
reported by surface_fdr. If you didn't take note of this you can either run
it again or do: fslstats filename_pthresh.nii.gz -R
to find out the maximum value.
Finally, if you would like a thresholded p-value image then you can
convert the pthresh "image" (in nifti format, but not a 3D image) to a
surface (vtk format) by the following command (you can see this and
equivalent commands in the surface_fdr script):
run_mesh_utils -m original_va_output.vtk -i filename_pthresh.nii.gz -o newfilename_pthresh.vtk --doVolumeToScalars
However, this often produces worse displays since the p-values that do
not survive the threshold are set to zero and look highly significant on the
image. So I would recommend using the original normally, *unless* the
threshold is very close to some non-significant p-values and so everything
looks the same shade of blue. In that case I would do:
fslmaths filename_pthresh -bin -mul -1 -add 1 -thr 0.5 -bin -add filename_pthresh newname_pthresh
run_mesh_utils -m original_va_output.vtk -i newname_pthresh.nii.gz -o newname_pthresh.vtk --doVolumeToScalars
and then look at "newname_pthresh.vtk" as this will set all p-values that didn't survive
the threshold to values higher than 1 and so you can set your display range to
anything that looks nice (with maximum value of 1.0 or less) and not get non-significant
p-values shown.
All the best,
Mark
On 7 Jul 2011, at 08:15, philipp thomann wrote:
> Dear Mark,
>
> great, that's it! Just one final question on this issue: in terms of significance, the color bar is now inversely than to what it was for the F-statistics (which sounds reasonable to me since lower p "means" higher F). Am I interpreting correctly?
>
> Thanks again and best regards
>
> Philipp
>
>
>
> -------- Original-Nachricht --------
>> Datum: Wed, 6 Jul 2011 17:32:29 +0100
>> Von: Mark Jenkinson <[log in to unmask]>
>> An: [log in to unmask]
>> Betreff: Re: [FSL] FSL-FIRST vertex analysis: converting F into p-value
>
>> Dear Philipp,
>>
>> If you load in your pvals vtk file, and not the F-stat one, then you can
>> see the p-values by adjusting the scalar display range within the surface
>> properties pop-up (the same one that the slider is in). That, together with
>> making a clipping box in the 3D brain (see the FSLView docs), so that you
>> only see certain slices in the background should generate a figure very
>> similar to Fig 8b. We actually used a slightly different piece of software to
>> produce this figure (hence the different colourmap, which is not available
>> in FSLView) but you can get a very similar image out of FSLView.
>>
>> All the best,
>> Mark
>>
>> On 6 Jul 2011, at 16:02, philipp thomann wrote:
>>
>>> Dear Mark,
>>>
>>> thanks for your quick reply. Maybe I was not clear enough: I already ran
>> surface_fdr and loaded the corresponding *Fthresh.vtk file in FSLView
>> (showing that most effects survive correction). Now I would like to
>> "illustrate" the p-values (as in Fig. 8b of the Patenaude paper). Sorry, but I was not
>> able to find any information on how to do this in the course, FSLView or
>> FIRST documentation.
>>>
>>> Thanks again
>>>
>>> Philipp
>>>
>>>
>>>
>>> -------- Original-Nachricht --------
>>>> Datum: Wed, 6 Jul 2011 15:20:42 +0100
>>>> Von: Mark Jenkinson <[log in to unmask]>
>>>> An: [log in to unmask]
>>>> Betreff: Re: [FSL] FSL-FIRST vertex analysis: converting F into p-value
>>>
>>>> Dear Philipp,
>>>>
>>>> The 3D view in FSLView will do this.
>>>> You need to run surface_fdr to get corrected statistics and then
>>>> just follow the directions for display in the practical session from
>>>> the FSL course.
>>>>
>>>> All the best,
>>>> Mark
>>>>
>>>> On 6 Jul 2011, at 14:58, philipp thomann wrote:
>>>>
>>>>> Dear FSL experts,
>>>>>
>>>>> is there any FSLview driven way to show the color-coded p-values of a
>>>> given analysis (e.g. group comparison or correlation) or do you have to
>> use
>>>> fslmaths for this purpose? If so, how? I'd like to illustrate my
>> results as
>>>> shown in the Patenaude paper (Neuroimage 2011).
>>>>>
>>>>> Any advice is highly appreciated
>>>>>
>>>>> Philipp
>>>>>
>>>>>
>>>>> --
>>>>> Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
>>>>> belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de
>>>>>
>>>
>>> --
>>> NEU: FreePhone - kostenlos mobil telefonieren!
>>> Jetzt informieren: http://www.gmx.net/de/go/freephone
>>>
>
> --
> Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
> belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de
>
|