The only two things that should be different are the number of iterations and the amount of smoothing. The cached simulations used 10k iterations (you have 5k). When you use the cached, it uses the FWHM that is closest to and higher than whatever is in your data. But this should make it more conservative. Maybe run it for 10k and see if you get similar results?

doug

Chris Hyatt wrote:

Dear Freesurfer users,

I am using mri_glmfit-sim to do clusterwise correction for multiple comparisons. I ran the Z Monte Carlo simulation for 5000 iterations for my 43 subjects (vertex wise threshold of 2.0 or p < 0.01), but I received different results than when I ran the correction using the "new" pre-cached simulation (using threshold of 2.0). In general, I get fewer clusters when running the simulation from scratch when compared with the pre-cached simulation. I used the fsaverage subject brain in the original analysis. Here are the commands I ran:

initial glmfit (uncorrected):

mri_glmfit --y lh.HCvsCD.thickness.10.mgh --fsgd HCvsCD.fsgd dods --C group.diff.mtx --surf fsaverage lh --cortex --glmdir lh.HCvsCD.glmdir

Then corrections for multiple comparisons ....

from scratch:

mri_glmfit-sim --glmdir lh.HCvsCD.glmdir --sim mc-z 5000 2.0 mc-z.pos --sim-sign pos --bg 5

using pre-cached:

mri_glmfit-sim --glmdir lh.HCvsCD.glmdir --cache 2.0 pos

Which method is most accurate?

Also, how does one turn off the polling writes to the console when running MC-Z from scratch?

Thanks, Christopher Hyatt Olin Neuropsychiatric Research Center / IOL Hartford, CT

---

---

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