Dear Anderson,
Thanks again for all your great suggestions!
Wright now I am running the different randomise analyses. However, for the first one (voor the global mean contrast) I get the following warning in my .o file:
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
The script that I used was:
fsl_sub -q lowprior*.q randomise –i gfeat/higher_level_feat_amygdala_randomise.gfeat/cope1.feat/filtered_func_data.nii.gz -o randomise_amygdala_GM/cope_1/cope1_randomise_amygdala_GM -d design/higher_level_amygdala_design_randomise_GM.mat -t design/higher_level_amygdala_design_randomise_GM.con -e design/higher_level_amygdala_design_randomise_GM.grp -m masks/whole_brain_mask.nii.gz -n 500 -D –T -1
Based on previous posts on the forum the warning suggests that the contrasts I specified is incorrect (see attachment for design files). Do you have any idea what went wrong?
Best,
Bianca
PhD-student
Department of Child and Adolescent Psychiatry, Curium-LUMC
Brain and Development Lab, Faculty of Social Science, Leiden University
Curium-LUMC
Endegeesterstraatweg 27
2342 AK Oegstgeest
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: dinsdag 22 juli 2014 21:23
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
Hope this helps!In general non-linear produces better alignment, and in your earlier email, you seemed to be planning to use FLIRT, but it's FNIRT that does the non-linear registration. It is possible to run FNIRT manually and get the images aligned, but it's far simpler to just use FEAT to do all the work for you.Hi Bianca,Sorry, I didn't mean to sound cryptic... What happens is that when FEAT runs for higher-level analyses, it already does all the registrations for all subjects automatically, based on settings that you define in the FEAT window (in particular, in the Registration tab). In that window you have the option of using "Non-linear" registration or not. If you select non-linear, FNIRT is used behind the scenes to warp the images to the standard space. If you use non-linear then FLIRT, is used to linearly align the images.
So, my suggestion is: open FEAT, set up the higher-level model, doing everything as if you were going to use FEAT for the actual analysis. Then run it. It will do all the registrations for you, and also merge the files for each subject into a single 4D file. You can ignore the remaining outputs, because for some of your contrasts, with repeated measures, there may be issues in the modelling (in fact, you can even use "fixed effects" to go faster). Then take the file "filtered_func_standard.nii.gz" that will then be present in the *.gfeat directory, and use it as the input for randomise, using the permutations with exchangeability blocks as I mentioned in an earlier email.
All the best,
Anderson
On 22 July 2014 18:55, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Hey Steven,
Zie onderstaande reactie van Anderson.Kun je me helpen om dit te vertalen naar mijn begripsniveau?
Thanks,
BiancaAll the best,Hi Bianca,It's probably better in general to use FNIRT (you may have selected it in the "Non-Linear" option in the FEAT Registration tab). However, it's simpler if you set the whole model using FEAT (i.e., as if you were going to run this as a higher-level FEAT analysis), then just take the filtered_func_data.nii.gz from the *.gfeat directory and use as the input to randomise. The registration will be either linear or non-linear depending on the configurations set in FEAT then.
Anderson
On 22 July 2014 13:08, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Dear Anderson,
Thanks for all these great suggestions.Just to check: the copes from the lower level feat are still in native space. For the randomise analyses I need cope data in standard space. Is it correct to register the cope data to standard space with the use of FLIRT using the example_func2standard.mat from the lower level registration folder.
scripting:flirt [options] -in <inputvol> -ref <refvol> -applyxfm -init <matrix> -out <outputvol>
input=cope dataref= standard space imageinit= registration matrix example_func2standard.matout= file name
Thanks in advance,
Bianca
PhD-student
Department of Child and Adolescent Psychiatry, Curium-LUMC
Brain and Development Lab, Faculty of Social Science, Leiden University
Curium-LUMC
Endegeesterstraatweg 27
2342 AK Oegstgeest
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: donderdag 10 juli 2014 13:17
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
Hi Bianca,
Please, see below:
On 10 July 2014 07:32, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Dear Anderson,
Thanks again for helping me.I talked to a colleague of mine about the randomise analyses. There were two things that we were not able to figure out:
1. what is the 4D input data for the analyses --> for a feat analyses you use the lower level feat directories, but we are not sure about what to use here.
Use the COPEs from each lower level FEAT directory, then concatenate them into a 4D with fslmerge.
2. Did you specify groups in the first column of your EV matrix? I now have a group column in which all participants from the control group have a 1 and all participants of the clinical have a 2. Is that correct?
It's one group per subject (i.e., one exchangeability block per subject). When you save the design, there will be a warning message about the it not being separable. This message can be ignored in this case, as the design is for use with randomise.
All the best,
Anderson
Bianca
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: woensdag 9 juli 2014 16:41
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
Hi Bianca,
You'll run this in randomise, so the warning message about separable EVs can be ignored.
All the best,
Anderson
On 9 July 2014 15:31, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Dear Anderson,
Thank you so much for all your help! It is very helpfull.Deleting EV5 and EV6 solved the problem concerning the relatedness of EV's.However, I keep getting the warning saying:
design matrix uses different groups (for different variance), but these do not contain seperable EVs for the different groups etc.
Is this related to the EV containing fd values --> there is one EV for the complete sample and not two EVs with the fd values splitted by group. And can it also be related to the voxel dependt EV?
Bianca
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: woensdag 9 juli 2014 13:20
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
Hi Bianca,Thanks for sending. You can remove EV5 and EV6, which it seems are for age and sex, but leave EV7, as well as the last one. These two (EV7 and the last) are for confounds within subject, that differ between sessions, whereas EV5 and EV6 are the same for all sessions of any given subject, and their effect is captured by the subject-specific EVs, so they don't need to be included in the model. Just delete them :)
All the best,
Anderson
On 9 July 2014 11:26, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Dear Anderson,
Sorry for not sending the right files.In the attachment I included the requested files.Hope these are helpful.
Concerning the suggestion made below --> I am not completely sure whether I understand it correctly. Do you mean that cavirates should be included in the lower level feat?
Thanks a lot,
Bianca
PhD-student
Department of Child and Adolescent Psychiatry, Curium-LUMC
Brain and Development Lab, Faculty of Social Science, Leiden University
Curium-LUMC
Endegeesterstraatweg 27
2342 AK Oegstgeest
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: woensdag 9 juli 2014 11:51
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
PS: Btw, there's no need to include age, sex, or any other EV that varies for each subject. These are already accounted for by the subject-specific EVs. Only include as nuisance EVs that vary within subject (if any).
On 9 July 2014 10:47, Anderson M. Winkler <[log in to unmask]> wrote:
All the best,Hi Bianca,Sorry, but these figures don't help. Could you send the .png files that Glm saves? Thanks.
Anderson
On 9 July 2014 10:34, SUBSCRIBE FSL Bianca <[log in to unmask]> wrote:
Dear Anderson,
Sorry for forgetting the attachment. The problem I mentioned before was solved by replacing the (,) with a (.).However, now something else goes wrong and it would be great if you can give me some advice.
I setup the complet model but get an error when I am trying to run the model (see attachment) --> at least one EV is close to a linear combination of the other …
The efficiency matrix indicates that some of my variables are related to each other (probably the covariates (age, gender and number of fd's) and the voxel dependent EV (grey matter).
In total there are 49 EV's --> EV 1-4 represent the two groups and the three measurements, EV 5-7 are the covariates and EV 49 is the voxel dependent EV.
The problem is not only caused by EV49 --> when leaving out this EV the error message still appears.
I read something about orthoganalisation on the mailing list, but I am not sure whether that is the solution for this problem.
All help is very appreciated.
Bianca
Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]Verzonden: dinsdag 8 juli 2014 13:24
Aan: [log in to unmask]
Onderwerp: Re: [FSL] longitudinal resting state analysis: 3 measurements x 2 group
All the best,Hi Bianca,
Did you mean to send an attachment? Nothing came... but the message suggests that in your C1, it's currently all zeros...
Anderson