Dear free surfer experts,
I was wondering if someone could give me some general guidance on controlling for multiple comparisons using clustering and FDR. This the situation. I completed an analysis of functional data by first dividing the cortex into parcellations using the icosahedron 5 file then exported the values to matlab. I now have a correlation and significance value for all 10242 parcelations. My uncorrected results made theoretical sense, topologically. However, I attempted to use an FDR procedure on all these parcellations setting the FDR at .05 and no parcellations remained (all values were above the .05 threshold). Because my parcelationas are likely topologically correlated with the nearest neighboring clusters, it is my understanding that it makes more sense to use a clustering method to first identify clusters and then run a correction using FDR as cited in Chumbley, & Friston 2009.
To complete clustering prior to FDR, I have projected the significance values from the parcellations back onto the surface vertices and then run the clustering approach described in the group analysis tutorial (mri_glmfit.sim). I have projected these group results onto the associated vertices on the fsaverage5 surface. However, when I run mri_glmfit.sim, no clusters remain. I believe that I am running into this problem because I am trying to get clusters of vertices when the data was actually analyzed using 10242 ROIs. Please note, I have placed the correlation and significance values (from my analysis in matlab) in a slope directory under GLM prior to running the analysis on the significance values associated with each vertex.
By replacing the significance file generated from a thickness study with the significance file that I created, does freesurfer conduct the same analysis without reference to the type of data; does it only matter that I am using significance values or should I be replacing another file in the directory (I am using a different methodology to obtain my correlation and significance values)? The fsgd file that is in the folder still represents the groups correctly even though the neural data has been changed external to free surfer.
Any suggestions or appropriate literature on how to correct for multiple comparisons using clusters of ROIs on surface would be appreciated?
Thank you, Erik Jahner
Hi Erik, I don't entirely follow what you are doing. It is not appropriate to run mri_glmfit-sim on an fdr corrected map. Why not analyze your functional data on fsaverage5?
On 11/15/2016 12:59 AM, ERIK JAHNER wrote:
Dear free surfer experts,
I was wondering if someone could give me some general guidance on controlling for multiple comparisons using clustering and FDR. This the situation. I completed an analysis of functional data by first dividing the cortex into parcellations using the icosahedron 5 file then exported the values to matlab. I now have a correlation and significance value for all 10242 parcelations. My uncorrected results made theoretical sense, topologically. However, I attempted to use an FDR procedure on all these parcellations setting the FDR at .05 and no parcellations remained (all values were above the .05 threshold). Because my parcelationas are likely topologically correlated with the nearest neighboring clusters, it is my understanding that it makes more sense to use a clustering method to first identify clusters and then run a correction using FDR as cited in Chumbley, & Friston 2009.
To complete clustering prior to FDR, I have projected the significance values from the parcellations back onto the surface vertices and then run the clustering approach described in the group analysis tutorial (mri_glmfit.sim). I have projected these group results onto the associated vertices on the fsaverage5 surface. However, when I run mri_glmfit.sim, no clusters remain. I believe that I am running into this problem because I am trying to get clusters of vertices when the data was actually analyzed using 10242 ROIs. Please note, I have placed the correlation and significance values (from my analysis in matlab) in a slope directory under GLM prior to running the analysis on the significance values associated with each vertex.
By replacing the significance file generated from a thickness study with the significance file that I created, does freesurfer conduct the same analysis without reference to the type of data; does it only matter that I am using significance values or should I be replacing another file in the directory (I am using a different methodology to obtain my correlation and significance values)? The fsgd file that is in the folder still represents the groups correctly even though the neural data has been changed external to free surfer.
Any suggestions or appropriate literature on how to correct for multiple comparisons using clusters of ROIs on surface would be appreciated?
Thank you,
Erik Jahner
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 -
ERIK JAHNER