On 05/13/2013 11:11 AM, Fotiadis, Panagiotis wrote:
Thanks for the clarifications, Doug!
I am using FS stable version 5.0.0, so I'm assuming that the --2spaces flag has probably been implemented for the more recent FS versions.
Thanks, that makes a lot more sense now.
In addition, since clusters are formed based on the number of neighboring voxels above the set voxel-based threshold, how is it possible to have reported (from the summary) clusters that have a cluster-wise p value higher than the voxel-based one? Isn't the cluster-wise p value of a cluster the maximum p value of all the contiguous voxels in that cluster?
The cluster-wise p-value is independent from the voxel-wise (ie, cluster forming) p-value. The clusterwise p-value is not the maximum of the p-values inside the cluster.
Thanks again for all your help. Panos
From: freesurfer-bounces@nmr.mgh.harvard.edu [freesurfer-bounces@nmr.mgh.harvard.edu] on behalf of Douglas Greve [greve@nmr.mgh.harvard.edu] Sent: Monday, May 13, 2013 12:22 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] Correction for multiple comparisons question
Hi Panagiotis,
On 5/12/13 9:32 PM, Fotiadis, Panagiotis wrote:
Hi Freesurfer experts,
I was wondering about three things concerning the use of the mri_glmfit-sim command:
- I tried running it with the --2spaces flag in order to Bonferroni correct over the two hemispheres but it tells me that the specific flag is not recognized. Is there another way to correct over the two hemispheres?
What version are you using? You can just adjust the clusterwise threshold by dividing by 2 (eg, if you want a clusterwise threshold of .05, then set it to .025)
- I set the cluster-based threshold to 0.999 in order to see all of the clusters and varied the voxel-based threshold in order to see the differences that it would make on the results. Specifically, i set it to be 1.3, 2.0, 2.3, and 3.0. Between the four different occurring cluster summaries, there was only one common cluster that appeared in all four summaries. As expected, the number of voxels of the specific cluster was decreasing as its voxel-wise threshold was increasing.
a) However, the reported cluster-wise threshold varied between 0.05 to 0.96 in those four cases. Does this sound reasonable? Shouldn't the cluster wise p values of the same cluster-between the four cases-be less dispersed?
I'm not sure what you mean here. At 3.0, the cluster may be very small and so could appear to be very insignificant.
b) If there is a cluster reported in only one of the four summaries (i.e. when the voxel-based threshold is 1.3) that say has a cluster-wise p value of 0.03, do I consider that significant or not, since it does not appear in any other summary with such a low cluster wise p value and since the only published p value is the cluster wise one?
In principle, it is significant. A lot of reviewers balk at such a liberal voxel-wise threshold since gaussian random fields breaks down at liberal thresholds. However, this uses simulations and not GRF. You should make sure the reviewers know this. There still might be some resistance since you could be picking up some low spatial frequency trends (ie, stuff that is not spatially gaussian) in addition to your true activation.
doug
Thank you in advance for your time, Panagiotis Fotiadis _______________________________________________ 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 _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer