Dear Freesurfer experts,
We are using mri_glmfit and wanted to compare our results to qdec to decide which method we prefer for group analysis. The results are consistent between the two methods, except for the monte carlo results. When we run mri_glmfit-sim to do the monte carlo simulations there are no significant clusters. We are not surprised by this as when we apply a FDR corrections all of our clusters go away. However, when we are using the monte carlo method in qdec we get a large significant cluster (in the csd summary file there are two large significant clusters, but using the image in qdec we only see one of them). The model is the same for both methods. We are (or think) we are using the exact same parameters as qdec. Is there a reason why the simulation results are so different? Which result should we use?
The following are two of the simulation commands we have tried using v5.1:
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 4 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.01 --overwrite
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 2 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.01 --overwrite
For the following command, we saw three clusters (they are inconsistent with the qdec ones), but none of them significant (p-values=.2):
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 4 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.995 --overwrite
We haven't run 10 000 simulations yet, because we wanted to see if there were any significant clusters there at all before running that many simulations.
Any advice would be very helpful.
Thank you for your time,
Marie
QDEC does not do its own calculations, it uses mri_glmfit and mri_glmfit-sim, so any differences you are seeing are due to differences in the parameters used to run the analysis. QDEC creates a glmdir output folder. You can check if the two methods have the same design by comparing the Xg.dat files. There should also be a ...mri_glmfit-sim.log in each glmdir. Check to see how those are different
doug
On 08/08/2013 11:08 AM, Marie Farstad Høvik wrote:
Dear Freesurfer experts,
We are using mri_glmfit and wanted to compare our results to qdec to decide which method we prefer for group analysis. The results are consistent between the two methods, except for the monte carlo results. When we run mri_glmfit-sim to do the monte carlo simulations there are no significant clusters. We are not surprised by this as when we apply a FDR corrections all of our clusters go away. However, when we are using the monte carlo method in qdec we get a large significant cluster (in the csd summary file there are two large significant clusters, but using the image in qdec we only see one of them). The model is the same for both methods. We are (or think) we are using the exact same parameters as qdec. Is there a reason why the simulation results are so different? Which result should we use?
The following are two of the simulation commands we have tried using v5.1:
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 4 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.01 --overwrite
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 2 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.01 --overwrite
For the following command, we saw three clusters (they are inconsistent with the qdec ones), but none of them significant (p-values=.2):
mri_glmfit-sim --glmdir glm/lh.asrs_sex_age.thickness.glmdir/ --sim mc-z 100 4 mc-z.unsigned --sim-sign abs --cwpvalthresh 0.995 --overwrite
We haven't run 10 000 simulations yet, because we wanted to see if there were any significant clusters there at all before running that many simulations.
Any advice would be very helpful.
Thank you for your time,
Marie
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Marie Farstad Høvik