Hi Clara,

you would do the analysis of a single time point the same way you'd run a regular glm or qdec analysis on cross sectional data. You cannot use LME as that is for longitudinal data (more than 1 time point per subject, at least for most subjects). Also the two stage model, first reduces to multiple measures in each subject to a single measure (slope of the linear fit). It would not be the right tool for selecting a single time point.

You could for example create a qdec table with the first row fsid timepoint1.long.base for each subject and then use qdec.

If you want to make use of the improved data from the longitudinal image processing, you need to specify the *.long.base directories, so that data is taken from there.

Good luck, Martin

On 12/08/2015 04:24 AM, Clara Kühn wrote:

Dear FreeSurfer experts,

I have a subsample of my longitudinal data with 3 scans per subjects fully preprocessed and would like to start with some simple correlations of thickness at time 1 with behavioral measures and other factors like age.

Is there a way to do that with LME (since I was planning on using that as a method of choice) or is it possible to do that with QDEC? I've prepared my data for both methods (LME and linear two stage model) but if I use the cross.qdec.table derived from the longitudinal data, it uses the registered and smoothed (with mris_slopes) surfaces from the Base, right? So, I would like to know how I can analyzse a single time point within my longitudinal design.

Cheers, Clara

Hi Freesurfer experts,

I eventually used the recon-all -s subjectname -qcache -target my_template to get the smoothed surfaces to use in qdec and therefore do cross-sectional analyses on the single tps of my longitudinal data. I did the analyses on the left hemisphere so far. Now that I tried using the right hemisphere I get the error message "multiple surfaces must have the same number of vertices". What does that mean? As far as I understood, the recon-all -qcache created smoothed surfaces for the right and left hemisphere. At least everything is in the respective folders...

Cheers and thanks in advance! Clara

----- Ursprüngliche Mail ----- Von: "ckuehn" ckuehn@cbs.mpg.de An: "Freesurfer support list" freesurfer@nmr.mgh.harvard.edu Gesendet: Mittwoch, 9. Dezember 2015 10:27:40 Betreff: Re: [Freesurfer] correlations with longitudinal data

Hi Martin,

thanks for your reply. I have created a qdec file with the fsidtp1.long.fsid.base as first column and specified the long directories with /path_to_data/*.long.*. And each subject has a ?h.thickness file from the recon-process. However, Qdec is asking for a very specific file ?h.thickness.fwhm10.nameoftemplate.mgz or .mgh (even when I pick 0 smoothing).

If I understand correctly, that is the file that I get when running the mris_slope command. But you said I wouldn't need that. Is there a way to get these files for a cross-sectional analysis of the *.long.*.base timepoints?

My qdec file currently looks like this:

fsid group sex age (in months) ... (behavioral measures) XXXX_1.long.XXXX.base 1 0 52 ... YYYY_1.long.YYYY.base 2 1 48 ... ...

Thank you, Clara

----- Ursprüngliche Mail ----- Von: "mreuter" mreuter@nmr.mgh.harvard.edu An: freesurfer@nmr.mgh.harvard.edu Gesendet: Dienstag, 8. Dezember 2015 18:14:40 Betreff: Re: [Freesurfer] correlations with longitudinal data

Hi Clara,

you would do the analysis of a single time point the same way you'd run a regular glm or qdec analysis on cross sectional data. You cannot use LME as that is for longitudinal data (more than 1 time point per subject, at least for most subjects). Also the two stage model, first reduces to multiple measures in each subject to a single measure (slope of the linear fit). It would not be the right tool for selecting a single time point.

You could for example create a qdec table with the first row fsid timepoint1.long.base for each subject and then use qdec.

If you want to make use of the improved data from the longitudinal image processing, you need to specify the *.long.base directories, so that data is taken from there.

Good luck, Martin

On 12/08/2015 04:24 AM, Clara Kühn wrote:

Dear FreeSurfer experts,

I have a subsample of my longitudinal data with 3 scans per subjects fully preprocessed and would like to start with some simple correlations of thickness at time 1 with behavioral measures and other factors like age.

Is there a way to do that with LME (since I was planning on using that as a method of choice) or is it possible to do that with QDEC? I've prepared my data for both methods (LME and linear two stage model) but if I use the cross.qdec.table derived from the longitudinal data, it uses the registered and smoothed (with mris_slopes) surfaces from the Base, right? So, I would like to know how I can analyzse a single time point within my longitudinal design.

Cheers, Clara

Hi Martin,

thanks for your reply. It turns out that QDEC just doesn't like working on different hemispheres in the same instance. If I close it, restart it and do the analyses on the other hemisphere only, it works like a charm. Only when you try to change the hemisphere in the design tab, it gives out that error.

I also wanted to tell you that I'm learning to do the LMEs with Matlab and I run into an error with matlabpool. When I try to fit the model with lme_mass_fit_EMinit (which calls the function matlabpool). I get an error that matlabpool doesn't exist. It turns out that matlabpool is now called parpool. I worked with matlab 8.6. I went back to check which matlab version still uses matlabpool and I find that up to 8.2 matlabpool works. So maybe you could put that as a warning into the tutorial or something like that.

As for the qcache problem: I ended up doing exactly what you said. Good to hear that I did it right ;)

Thank you and Frohe Weihnachten! Clara

----- Ursprüngliche Mail ----- Von: "mreuter" mreuter@nmr.mgh.harvard.edu An: "Freesurfer support list" freesurfer@nmr.mgh.harvard.edu Gesendet: Donnerstag, 17. Dezember 2015 17:40:40 Betreff: Re: [Freesurfer] correlations with longitudinal data

Hi Clara,

so are you creating the qcache files in the cross sectional directories? I would recommend to use the longitudinal directories, for that you can simply add the -qcache flag to the longitudinal recon-all like recon-all -long <tpid> <baseid> -qcache -target <template>

I don't know why you are getting an error with the other hemisphere. Maybe you use a different template for qcache and then fsaverage later?

Best, Martin

On 12/11/2015 11:36 AM, Clara Kühn wrote:

Hi Freesurfer experts,

I eventually used the recon-all -s subjectname -qcache -target my_template to get the smoothed surfaces to use in qdec and therefore do cross-sectional analyses on the single tps of my longitudinal data. I did the analyses on the left hemisphere so far. Now that I tried using the right hemisphere I get the error message "multiple surfaces must have the same number of vertices". What does that mean? As far as I understood, the recon-all -qcache created smoothed surfaces for the right and left hemisphere. At least everything is in the respective folders...

Cheers and thanks in advance! Clara

----- Ursprüngliche Mail ----- Von: "ckuehn" ckuehn@cbs.mpg.de An: "Freesurfer support list" freesurfer@nmr.mgh.harvard.edu Gesendet: Mittwoch, 9. Dezember 2015 10:27:40 Betreff: Re: [Freesurfer] correlations with longitudinal data

Hi Martin,

thanks for your reply. I have created a qdec file with the fsidtp1.long.fsid.base as first column and specified the long directories with /path_to_data/*.long.*. And each subject has a ?h.thickness file from the recon-process. However, Qdec is asking for a very specific file ?h.thickness.fwhm10.nameoftemplate.mgz or .mgh (even when I pick 0 smoothing).

If I understand correctly, that is the file that I get when running the mris_slope command. But you said I wouldn't need that. Is there a way to get these files for a cross-sectional analysis of the *.long.*.base timepoints?

My qdec file currently looks like this:

fsid group sex age (in months) ... (behavioral measures) XXXX_1.long.XXXX.base 1 0 52 ... YYYY_1.long.YYYY.base 2 1 48 ... ...

Thank you, Clara

----- Ursprüngliche Mail ----- Von: "mreuter" mreuter@nmr.mgh.harvard.edu An: freesurfer@nmr.mgh.harvard.edu Gesendet: Dienstag, 8. Dezember 2015 18:14:40 Betreff: Re: [Freesurfer] correlations with longitudinal data

Hi Clara,

you would do the analysis of a single time point the same way you'd run a regular glm or qdec analysis on cross sectional data. You cannot use LME as that is for longitudinal data (more than 1 time point per subject, at least for most subjects). Also the two stage model, first reduces to multiple measures in each subject to a single measure (slope of the linear fit). It would not be the right tool for selecting a single time point.

You could for example create a qdec table with the first row fsid timepoint1.long.base for each subject and then use qdec.

If you want to make use of the improved data from the longitudinal image processing, you need to specify the *.long.base directories, so that data is taken from there.

Good luck, Martin

On 12/08/2015 04:24 AM, Clara Kühn wrote:

...

Dear FreeSurfer experts,

I have a subsample of my longitudinal data with 3 scans per subjects fully preprocessed and would like to start with some simple correlations of thickness at time 1 with behavioral measures and other factors like age.

Is there a way to do that with LME (since I was planning on using that as a method of choice) or is it possible to do that with QDEC? I've prepared my data for both methods (LME and linear two stage model) but if I use the cross.qdec.table derived from the longitudinal data, it uses the registered and smoothed (with mris_slopes) surfaces from the Base, right? So, I would like to know how I can analyzse a single time point within my longitudinal design.

Cheers, Clara