On 4/26/14 2:11 PM, Pedro Rosa wrote:
Thanks, Doug!
I looked at the examples on the wiki, and the Paired Analysis wiki page as well.
You mentioned I should include the categorical and continuous variables in the fsgd file for the mri_glmfit. It seems that one of my analyses includes few subjects,  and then one of the class end up lacking a adequate range of continuous variables, and consequently the mri_glmfit finished with errors (this error no longer appears after I merged that problematic group with another).
Therefore I ask whether it would be fine to run the mri_glmfit without some (or all) the continuous  or/an categorical variables, while they would of course be included in the LME ran in MatLab.
Yes, that is fine
Further, I would like to ask what the FMHM outputs represents?
The FWHM refers to the width of a gaussian. It is measured as the average  correlation between the residuals of neighboring vertices (the correaltion coef get transformed into the FWHM). Smoothing with a gaussian of a certain FWHM will induce correlation, but there can be correlation from other sources so the FWHM you get out is not necessarily that which you used to smooth the data with.

doug
Regards,
Pedro Rosa.

On Wednesday, April 16, 2014 at 11:49 AM, Douglas N Greve wrote:


On 04/15/2014 10:34 PM, Pedro Rosa wrote:
Hi, Doug.
Thanks a lot!
The command would look like this: mri_glmfit-sim --glmdir
lh.thickness.Sch.glmdir --sim mc-z 10000 2 teste --sim-sign abs
--overwrite --cache 1.3 abs
It takes only a few seconds to run.
The command is fine, but remove --sim mc-z 10000 2 teste. This tells it
to do the simulation, which you don't need to do. It is overridden by
the --cache flag. If they had occurred in the other order, you would be
waiting days for the simulation to run.
I have a few questions about how to run it:
1) How the single entry for each subject look like in the FSGD file?
I'm not sure what you mean. Have you looked at the examples on the wiki?

2) How to define the voxel/vertex-wise used to define clusters (after
—sim, -log10( p)), and the voxel-wise threshold for the —cache option?
What is the difference between them?
They are the same
doug

Thanks again,
Pedro Rosa.

On Monday, April 14, 2014 at 3:29 PM, Douglas N Greve wrote:


On 04/14/2014 10:35 AM, Pedro Rosa wrote:
Dear Doug and Jorge,
I tried what you suggested and I think it work, although I have some
concerns.
I am working with a longitudinal study with two time-points for all
subjects, three categorical variables (group, substance abuse /
dependence and gender) and three continuous variables (interval
between scans, age and medication intake).
I generated a contrast with intercept + 7 betas for the LME, ran it
without any problem and saved the sig.mgh using
fs_write_fstats(F_lhstats,mri,’sig.mgh’,’sig’).
For the mri_glmfit I entered the same output from mri_surf2surf I used
for the LME (smoothed at 10mm), but I did not know how exactly to
enter the categorical and continuous variables, or which contrast to
use.
I would do it as a paired-test (see the wiki). You may have to re-run
mris_preproc with the --paired-diff flag, then smooth by 10mm. Use this
as the input to mri_glmfit. Set up the FSGD with the categorical and
continuous variables (note that the FSGD file will have only a single
entry per subject). Create your contrasts and run mri_glmfit. Overwrite
the sig.mgh with the one from LME. Then run mri_glmfit-sim with the
--cache option (not permutation)

I don't know what your contrast of interest is, so I can't help you
there. In the end, it does not matter because you are overwriting the
sig map anyway. You just need a contrast as a place holder.

doug

The commmand was: mri_glmfit —y pval.mgh —sim perm 10000 0.05 sch

I just tested creating a matrix with 24 columns
(Nclasses*(Nvariables+1) as suggested for DODS).
Afterwards I ran the mri_glmfit-sim (mri_glmfit-sim --glmdir
Sch-glmdir --sim mc-full 5 2 teste --sim-sign abs, and it finished
apparently without errors.
I attached the logs for both mri_glmfit and mri_glmfit-sim.

That said, I have the following questions:

1) What does the FWHM procedure does?
2) How should I decide which contrast to test if the mri_glmfit does
not consider the longitudinal design?
3) Will the mri_glmfit-sim consider only the FMHM output from
mri_glmfit and sig.mgh from the LME, or also other outputs from the
mri_glmfit?
4) Does the FWHM rely only on the images, and not on variables and
contrasts?

Thank you very much!
Pedro Rosa.

On Monday, March 31, 2014 at 10:53 PM, Pedro Rosa wrote:

Thanks, Doug!
Should I run the mri_preproc and and smooth the output using
mri_surf2surf with, let’s say, 10mm, and than run the LME normally in
MatLab?
Would this be problematic with a different smoothing procedure in
mri_glmfit?
How will mri_glmfit deal with the longitudinal design? Does this
matter, or the FWHM would only be estimated on a average image of all
time-points for all subjects?
Regards,
Pedro Rosa.

On Sunday, March 30, 2014 at 3:51 PM, Douglas Greve wrote:


I think I would just run mri_glmfit on your data to get the proper
directly structure and estimate of FWHM, then copy the sig file from
the mixed fx analysis into the glmfit folder for one of the
contrasts. Then run mri_glmfit-sim.

doug


On 3/29/14 10:29 AM, Pedro Rosa wrote:
Dear Doug and Jorge,
Thank you very much for your help.
I found another message in the list

in which you suggested a way of using MC in mri_glmfit-sim by
creating “fake files”, which would not be read by the script. In
this case, only the simulation would be run, and not the full
statistics. The command would be something like this:
- mri_glmfit-sim --glmdir $SUBJECTS_DIR --sim mc-full 5 2 teste
--sim-sign abs
I created a “fake” mri_glmfit.log, fwhm.dat and mask.mgh files as
suggested by the older post. This would be fine, I believe, if only
sig.mgh is read by the script.
However, I get this message after running the command:

[server:Long-T0-T2-Posproc/Vertex/Sch] pedrogomesrosa%
mri_glmfit-sim --glmdir $SUBJECTS_DIR --sim mc-full 5 2 teste
--sim-sign abs

if: Expression Syntax.


Is it possible to do what I am trying to do? Does the residual
errors at each location included in the sig.mgh, and, if necessary,
how to compute it into image FWHM?

Regards,

Pedro Rosa.

On Friday, March 28, 2014 at 2:38 PM, Douglas N Greve wrote:

Jorge, do you output the FWHM?
doug

On 03/27/2014 03:14 PM, jorge luis wrote:
Hi Pedro

Sorry, right now the only multiple comparisons corrections
implemented
in lme are the original Benjamini and Hochberg (1995) FDR procedure
(lme_mass_FDR) and a more recent and powerful two-stage FDR
procedure
(lme_mass_FDR2):

Benjamini, Y., Krieger, A.M., Yekutieli, D. (2006). Adaptive linear
step-up procedures that control the false discovery rate.
Biometrika,
93, 491-507.

In my experience, this procedure is as powerful to detect
effects in
neuroimage data as alternative corrections with strong control of
the
family-wise error rate (FWE). However it would be great if we could
use an implementation of any multiple comparisons correction with
strong control of the FWE (MC, RFT, ect...) for lme (FDR procedures
only provide weak control). The residual errors at each location
required to compute an estimate of the image FWHM can be obtained
from
the lme output. But an actual FWHM estimate is not currently saved.

Best
-Jorge


El Martes 25 de marzo de 2014 8:15, Pedro Rosa
escribió:

Dear Doug,
Thank you very much!
I will try what you suggested, although I am not sure if Jorge's
stream outputs the FMHM, or if I would need to run the statistics
from the beggining using in the terminal, and not in MatLab.
Do you think Jorge could comment on this issue?
Regards,
Pedro Rosa.

On Mar 24, 2014, at 12:44 PM, Douglas Greve


In theory, it should be possible. I have not used Jorge's stream,
so I
don't know that much about it. Does it save an estimate of the
FWHM? If
so, then you can run mri_surfcluster passing it the p-value (ie,
-log10(p)) map, the FWHM, the mask, and a voxel-wise threshold.
This is
what mri_glmfit-sim does, so you might check that script for
mri_surfcluster command line options

doug


On 3/22/14 11:03 PM, Pedro Rosa wrote:
Dear list,
I ran the recon-all and the Freesurfer 5.1 longitudinal pipeline
in a structural MRI dataset and I would like to use Monte Carlo as
the method for correction for multiple comparisons. However, the
longitudinal LME tutorial includes only FDR correction
(lme_mass_FDR2).
Is it possible to use Monte Carlo correction for longitudinal
data? Can I input the outputs from MatLab (fstats =
lme_mass_F(?h,CM): stats.F / pval / sgn / df) into mri_glmfit and
then run Monte Carlo?
If not, do you have any other suggestions of how I use Monte
Carlo in longitudinal analyses?
Thanks in advance,

_______________________________________________
Freesurfer mailing list


The information in this e-mail is intended only for the person to
whom it is
addressed. If you believe this e-mail was sent to you in error and
the e-mail
contains patient information, please contact the Partners
Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to
you in error
but does not contain patient information, please contact the
sender and properly
dispose of the e-mail.



_______________________________________________
Freesurfer mailing list




_______________________________________________
Freesurfer mailing list

--
Douglas N. Greve, Ph.D.
MGH-NMR Center
Phone Number: 617-724-2358
Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
Outgoing:

_______________________________________________
Freesurfer mailing list



_______________________________________________
Freesurfer mailing list

_______________________________________________
Freesurfer mailing list


The information in this e-mail is intended only for the person to
whom it is
addressed. If you believe this e-mail was sent to you in error and
the e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to
you in error
but does not contain patient information, please contact the sender
and properly
dispose of the e-mail.







_______________________________________________
Freesurfer mailing list

--
Douglas N. Greve, Ph.D.
MGH-NMR Center
Phone Number: 617-724-2358
Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting

_______________________________________________
Freesurfer mailing list



_______________________________________________
Freesurfer mailing list

--
Douglas N. Greve, Ph.D.
MGH-NMR Center
Phone Number: 617-724-2358
Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting

_______________________________________________
Freesurfer mailing list
 



_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer