Hi,
I have a couple of specific questions regarding cluster analysis with mri_glmfit that weren't answered at the course the other week and on which the documentation is slightly vauge.
My data includes continuous variables (age and IQ, matching the documentaion!) so I should use MC simulation instead of permutation. I have 70 subjects in each group (case/control). The course notes say I can use the mc-z if I have O(80) subjects. How can I tell if I have enough subjects; is there a statistical test that would show if the outputs are reasonable? I would think 70 is close enough to 80, but would like to be able to test it as I assume it is going to depend to a certain extent on the data.
The notes say I need to supply the fwhm of the residual. This value is easy to find in the y file, but it is not clear how to specify it to mri_glmfit. Do we use the --fwhm flag as in the example on page 104/202 in the coursebook? Note that the example uses mc-full, not mc-z, and that the --fwhm flag an arguement to mri_glmfit, not a part of the --sim flag.
Finally, with regard to the clustering. The commmand as given in the course notes (pg 104/202) works, but the example does not match mri_surfcluster --help. Specifically, the --help text does not mention (that I could see) the --src flag or the --ocp flag. We are using the example from the course as our template. What is the output of the --ocp flag?
Thanks!
d.g.lawyer@medisin.uio.no wrote:
Hi,
I have a couple of specific questions regarding cluster analysis with mri_glmfit that weren't answered at the course the other week and on which the documentation is slightly vauge.
My data includes continuous variables (age and IQ, matching the documentaion!) so I should use MC simulation instead of permutation. I have 70 subjects in each group (case/control). The course notes say I can use the mc-z if I have O(80) subjects. How can I tell if I have enough subjects; is there a statistical test that would show if the outputs are reasonable? I would think 70 is close enough to 80, but would like to be able to test it as I assume it is going to depend to a certain extent on the data.
70 is probably fine. In any event, mc-z should be conservative relative to mc-full.
The notes say I need to supply the fwhm of the residual. This value is easy to find in the y file, but it is not clear how to specify it to mri_glmfit. Do we use the --fwhm flag as in the example on page 104/202 in the coursebook? Note that the example uses mc-full, not mc-z, and that the --fwhm flag an arguement to mri_glmfit, not a part of the --sim flag.
Yes, use the --fwhm flag. This flag can be used for either real data or simulation. Ie, you could use this to smooth your data instead of mri_surf2surf. Note that you will probably use a different value for smoothing your real data vs the simulation as the simulation usually needs to be smoothed more because the data often start a bit smooth.
Finally, with regard to the clustering. The commmand as given in the course notes (pg 104/202) works, but the example does not match mri_surfcluster --help. Specifically, the --help text does not mention (that I could see) the --src flag or the --ocp flag. We are using the example from the course as our template. What is the output of the --ocp flag?
sorry, I recently changed the names of some of the flags. --src and --ocp should still work, but they have been replaced in the help with --cwsig and --in. --cwsig/--ocp output the cluster-wise significance. Ie, if a vertex meets all the criteria to be in a cluster, then it's value will be -log10(p), where p is the cluster-wise significance (ie, the pvalue after correction for mult comparisons). This means that every vertex in the cluster will have the same value.
doug
Thanks!
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
d.g.lawyer＠medisin.uio.no -
Doug Greve