Hi,
Thank you for your answers.
In fact the best way would be that I don't resample in 4mm my fMRI datas after registration with MNI152_2mm (in prestats).
Like that, I stay in the good scale for overlay at the end. But the data would be too large to then run, Melodic / Dual-regression / and randomise (in 26 subjects in 2x2x2mm 91x109x91). What do you think ??
Otherwise, I think I will do the Steve's methods.
Kind regards,
Antoine
________________________________
De : FSL - FMRIB's Software Library [[log in to unmask]] de la part de Stephen Smith [[log in to unmask]]
Envoyé : mercredi 17 juillet 2013 07:03
À : [log in to unmask]
Objet : Re: [FSL] Randomise after dual regression
Hi - I don't really agree with Anderson; the p-values are valid before interpolation, and after interpolation they are still valid, just interpolated. But surely the main point is that you will have thresholded (e.g.) clusters of tstats in a valid way before upsampling, and just want to overlay the upsampled thresholded image onto (eg) a structural.
The practical problem of upsampling before randomise might be in this case that the data is too large to then run randomise.
Cheers, Steve.
On 17 Jul 2013, at 01:07, "Anderson M. Winkler" <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi Antoine,
My advice is that you run the interpolation before randomise, not after. The reason is that, as you observed, the interpolation causes the values in the voxels to change a little bit, and if a p-value changes, even if just slightly, it may no longer be exact or even valid.
On the other hand, if you resample the actual observations that go into randomise, the model will be fit and permuted with the data having already undergone these small changes, which will then be correctly reflected in the p-values. This is particularly true for the corrected p-values, since upsampling involves the introduction of smoothness.
Hope this helps!
All the best,
Anderson
2013/7/16 Bernas, A. <[log in to unmask]<mailto:[log in to unmask]>>
Hi Matthew,
Thanks for your answer. I have just another question.
Now I have my statistic images like that : 4x4x4mm (45x54x45) and I want to resample them in 2x2x2mm (91x109x91) in order to overlay them on the MNI152_T1_2mm standard T1 image
So I do that : flirt -ref $FSLDIR/data/standard/MNI152_T1_2mm_brain -in [my statistic image] -out [my new image resampled] -applyisoxfm 2
It seems to works, but looking at the histograms it changes a little bit. But I think it's due to the interpolation, right?
So is it the best way to do that ?
And do you think it is better to resample my data before or after randomise ?
Kind regards
Antoine
________________________________________
De : FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] de la part de Matthew Webster [[log in to unmask]<mailto:[log in to unmask]>]
Envoyé : mardi 16 juillet 2013 15:05
À : [log in to unmask]<mailto:[log in to unmask]>
Objet : Re: [FSL] Randomise after dual regression
Hello Antoine,
Randomise does not currently preserve the pixdims in the output images when a mask is not supplied with -m option. We will look into changing this in a future patch, but for now you can use fslcpgeom to restore this information to statistic images..
Kind Regards
Matthew
> Hi Steve,
>
> Sorry for the delay, I did not see your answer.
> So, I do my t-test like that (I want to compare 2 groups resting state (ASD et CON) on a IC after a DR) :
>
> test_design.mat :
>
> /NumWaves 2
> /NumPoints 26
> /PPheights 1.000000e+00 1.000000e+00
>
> /Matrix
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 0.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
>
> test_design.con :
>
> /ContrastName1 all_mean
> /ContrastName2 ASD_mean
> /ContrastName3 Con_Mean
> /ContrastName4 CON>ASD
> /ContrastName5 CON<ASD
> /NumWaves 2
> /NumContrasts 5
> /PPheights 5.000000e-01 1.000000e+00 1.000000e+00 1.000000e+00 5.000000e-01
> /RequiredEffect 0.961 1.359 1.359 1.921 0.961
>
> /Matrix
> 1.000000e+00 1.000000e+00
> 0.000000e+00 1.000000e+00
> 1.000000e+00 0.000000e+00
> 1.000000e+00 -1.000000e+00
> -1.000000e+00 1.000000e+00
>
> and I do this command (e.g for the 16th IC) :
>
> $ randomise -i ~/Dr_grp_test3/dr_stage2_ic0015 -o ~/tests/rando_ic16 -d ~/Dr_grp_test3/test_design.mat -t ~/Dr_grp_test3/test_design.con -n 500 -T
>
> And if I check the nifti info of the intput :
>
> $ fslinfo Dr_grp_test3/dr_stage2_ic0015.nii.gz
> data_type FLOAT32
> dim1 45
> dim2 54
> dim3 45
> dim4 26
> datatype 16
> pixdim1 4.0000000000
> pixdim2 4.0000000000
> pixdim3 4.0000000000
> pixdim4 1.0000000000
> cal_max 0.0000
> cal_min 0.0000
> file_type NIFTI-1+
>
> and if I check one of the output I get that for example:
>
> $ fslinfo tests/rando_ic16_tfce_corrp_tstat1.nii.gz
> data_type FLOAT32
> dim1 45
> dim2 54
> dim3 45
> dim4 1
> datatype 16
> pixdim1 1.0000000000
> pixdim2 1.0000000000
> pixdim3 1.0000000000
> pixdim4 1.0000000000
> cal_max 0.0000
> cal_min 0.0000
> file_type NIFTI-1+
>
> So I don't know why randomise change the dimensions of my voxels.
>
> Best regards,
> Antoine
>
>
>
>
> ________________________________
> De : FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] de la part de Stephen Smith [[log in to unmask]<mailto:[log in to unmask]>]
> Envoyé : mercredi 3 juillet 2013 14:57
> À : [log in to unmask]<mailto:[log in to unmask]>
> Objet : Re: [FSL] Randomise after dual regression
>
> How did you do the t-test? It seems like maybe whichever program you used has not kept the header size intact. FSL programs should not cause that.
> Steve.
>
>
> On 3 Jul 2013, at 11:05, Antoine BERNAS <[log in to unmask]<mailto:[log in to unmask]><mailto:[log in to unmask]<mailto:[log in to unmask]>>> wrote:
>
> Hi everybody,
>
> I realised a two sample t-test with randomise on one component `dr_stage2_IC#` after a group-ica and a dual-regression.
> But for each contrast image randomise has changed the size. I get the same number of voxels (45x54x45) but I pass from 4mmx4mmx4mm to 1mmx1mmx1mm.
> So it causes some problems when I want to add a background with a standard image (because 64 times smaller). Is it normal that randomise does that ?
> Otherwise How can I enlarge my contrast images ?
>
> best regards,
>
> Antoine B.
>
>
>
> ---------------------------------------------------------------------------
> 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]<mailto:[log in to unmask]><mailto:[log in to unmask]<mailto:[log in to unmask]>> http://www.fmrib.ox.ac.uk/~steve
> ---------------------------------------------------------------------------
>
> Stop the cultural destruction of Tibet<http://smithinks.net<http://smithinks.net/>>
>
---------------------------------------------------------------------------
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]<mailto:[log in to unmask]> http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
Stop the cultural destruction of Tibet<http://smithinks.net>
|