External Email - Use Caution
Ok, well then I do not understand the purpose of the permutation testing if the number of permutation would not change the size of the clusters. What is it doing if not refining what vertices are significant? Perhaps I am just missing a key idea.
Best,
On Thu, Mar 14, 2019 at 5:11 PM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:
The size of the cluster is not going to be affected by the number of iterations (only by the threshold). Why would you think that the cluster size is affected by the number of iterations?
On 3/14/19 5:45 AM, Giuliana Klencklen wrote:
External Email - Use CautionHi Douglas,
According to your suggestion, I used the permutation simulation approach. I chose a cluster forming threshold set at 0.05 and explored how the number of iterations effects the data. For example, I used this command for 1,000 iterations: mri_glmfit-sim \ --glmdir lh.longMRI.glmdir \ --sim perm 1000 1.3 perm.abs.13 \ --sim-sign abs\ --cwpvalthresh 0.05
I did not observe any difference on the size of the cluster between 1,000 vs 5,000 vs 10,000 iterations. As I thought the results were pretty odd, I tried running the command with 5 permutations just to make sure it's not also the same and I did not find any significant clusters.
Is the absence of cluster size difference between different number of iterations expected? Just wanted to make sure with you that this approach is working as it should.
Thanks! Regards,
On Fri, Feb 22, 2019 at 6:50 PM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edu mailto:DGREVE@mgh.harvard.edu> wrote:
You can make the bonferroni correction from of mri_glmfit-sim the same as qdec by not including --2spaces (the bonferroni correction in qdec is actually 1, not 0). Also, if you want to use such a low cluster forming threshold (1.3=p<.05), then you should use permutation and not MC (MC is not valid at such low thresholds). On 2/22/19 7:49 AM, Giuliana Klencklen wrote:External Email - Use Caution Thanks Douglas, that makes sense. I have run mri_glmfit-sim and have that table file. However, the cortical thickness values for each cluster displayed in that table can’t be used because the clusters (something.sig.cluster.mgh opened with tksurfer) do not match exactly (but are very similar to) those previously generated with QDEC. I do not understand the cause of this issue because all the stats I used, i.e., threshold, statistical correction, level of smooting, seem to be the same between both qdec and mri_glmfit-sim. I send you here a typical example of the problematic clusters, as well as the summary for both qdec and mri_glmfit-sim. They appear similar except for the Bonferroni correction that is set at 2 for the mri_glmfit-sim while it is set at 0 for the qdec version? If it is the source of the current issue, how can I configure the Bonferroni correction? If not, do you have any idea how I can resolve the problem? Many thanks in advance. Regards, Giuliana Klencklen QdecGroupComparison.jpg On Fri, Feb 15, 2019 at 5:24 PM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edu <mailto:DGREVE@mgh.harvard.edu>> wrote: This can happen if the label is small and/or you've used a lot of smoothing. It is better to do this kind of thing in fsaverage space rather than moving the labels back to the individual space. If you've run mri_glmfit-sim, then it should have created a table file (something.y.ocn.dat). This file will have a row for each subject and a column for each cluster. The value will be the mean for that subject in that cluster. On 2/15/19 6:23 AM, Giuliana Klencklen wrote:External Email - Use Caution Hi FS experts, I did group-level, surface-based, vertex-wise analysis for baseline and longitudinal data. I used Qdec and do the same work with the fsgd version (mri_glmfit-sim command) to double-check the data. I created label files with tksurfer for each of the clusters showing a significant between-group difference. Then, I used the following command stream to extract the cortical thickness values for each subject and cluster: mri_label2label, mris_anatomical_stats, and aparcstats2table. E.g., mri_label2label --srcsubject fsaverage --srclabel/home/jagust/gklenck/Long_MRI/lh.ac-baseline-rostralmiddlefrontal.label
--trgsubject ${s}_tp1 --trglabel ${s}_tp1/label/lh.ac-baseline-rostralmiddlefrontal.label --hemi lh --regmethod surface mris_anatomical_stats -l lh.ac-baseline-rostralmiddlefrontal.label -t lh.thickness -b -f ${s}_tp1/stats/lh.ac-baseline-rostralmiddlefrontal.stats ${s}_tp1 lh aparcstats2table --subjects ${s}_tp1 --hemi lh --parc ac-baseline-rostralmiddlefrontal --meas thickness --tablefile ${out_dir}/lh.ac-baseline-rostralmiddlefrontal.aparc_stats.txt Subsequently, when I conducted between-group comparisons for each cluster, a couple of clusters (7 out of 16) do not show a significant difference - which does not make sense. This problem seems to appear randomly. Help to the FS archives, I tried to use mri_segstats command but was not able to find a correct combination of arguments. If this is the right track to solve this problem, what combination of arguments should I use? And if this is not, do you have any idea how I can solve this problem? Many thanks in advance. Regards, Giuliana -- Giuliana Klencklen, Ph.D. Helen Wills Neuroscience Institute University of California, Berkeley 118 Barker Hall Berkeley, CA 94720-3190 510-395-0040 giuliana.klencklen@berkeley.edu <mailto:giuliana.klencklen@berkeley.edu> _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer -- Giuliana Klencklen, Ph.D. Helen Wills Neuroscience Institute University of California, Berkeley 118 Barker Hall Berkeley, CA 94720-3190 510-395-0040 giuliana.klencklen@berkeley.edu <mailto:giuliana.klencklen@berkeley.edu> _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer-- Giuliana Klencklen, Ph.D.
Helen Wills Neuroscience Institute University of California, Berkeley 118 Barker Hall Berkeley, CA 94720-3190 510-395-0040 giuliana.klencklen@berkeley.edu mailto:giuliana.klencklen@berkeley.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer