Hi,
One of my papers is under review showing significant differences in cortical surface area between healthy controls and patients.
In this paper, regarding statistical results, for multiple comparison correction, I ran Monte Carlo simulations using following command:
mri_glmfit-sim \
--cache 4 neg \ --cwp 0.05\
--2spaces In the paper, I reported that multiple comparisons were corrected at p < 0.05 using Monte Carlo simulations. I also reported cluster sizes (number of voxels) of all clusters which survived multiple comparison correction.
I checked literature and found that that's how people report cluster results after running analysis in FreeSurfer.
One of the reviewers asked- Did you consider any constraints on findings (specifically cluster size) before considering the finding significant?
Could you please help me in understanding this question and how can this be answered?
In this paper: http://onlinelibrary.wiley.com/store/10.1002/pbc.25386/asset/pbc25386.pdf?v=..., authors reported that - "The data were tested against an empirical null distribution of maximum cluster size across 10,000 iterations using Z Monte Carlo simulations as implemented in FreeSurfer [31,32] synthesized with a cluster-forming threshold of P < 0.05 (two-sided), yielding clusters fully corrected for multiple comparisons across the surfaces. Clusterwise corrected P < 0.05 (two-sided) was regarded significant."
I assume that that's what reviewer is demanding me to report in my paper.
I would really appreciate any help.
It sounds like you have reported everything correctly. The clusterwise p-value corresponds to a critical cluster size at the given cluster forming threshold (4), sign (neg), and fwhm (fwhm.dat in the glmdir). So the answer to the reviewer is "yes". If you want to report the critical size, then look in (if fwhm=10)
$FREESURFER_HOME/average/mult-comp-cor/fsaverage/lh/cortex/fwhm10/neg/th40/mc-z.cdf
The MaxClustCDF column gives the the clusterwise p-value and the MaxClustBin gives the size of the cluster needed to achieve that p-value. Eg, a p-value of .05 would require a cluster of about 35 mm2.
On 08/08/2017 01:30 PM, Martin Juneja wrote:
Hi,
One of my papers is under review showing significant differences in cortical surface area between healthy controls and patients.
In this paper, regarding statistical results, for multiple comparison correction, I ran Monte Carlo simulations using following command: mri_glmfit-sim \
--cache 4 neg \ --cwp 0.05\ --2spaces In the paper, I reported that multiple comparisons were corrected at p < 0.05 using Monte Carlo simulations. I also reported cluster sizes (number of voxels) of all clusters which survived multiple comparison correction.
I checked literature and found that that's how people report cluster results after running analysis in FreeSurfer.
One of the reviewers asked- Did you consider any constraints on findings (specifically cluster size) before considering the finding significant?
Could you please help me in understanding this question and how can this be answered?
In this paper: http://onlinelibrary.wiley.com/store/10.1002/pbc.25386/asset/pbc25386.pdf?v=..., authors reported that - "The data were tested against an empirical null distribution of maximum cluster size across 10,000 iterations using Z Monte Carlo simulations as implemented in FreeSurfer [31,32] synthesized with a cluster-forming threshold of P < 0.05 (two-sided), yielding clusters fully corrected for multiple comparisons across the surfaces. Clusterwise corrected P < 0.05 (two-sided) was regarded significant."
I assume that that's what reviewer is demanding me to report in my paper.
I would really appreciate any help.
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 -
Martin Juneja