help > FWE in pickatlas vs SPM
Showing 1-3 of 3 posts
Apr 7, 2015 02:04 PM | Bradley Taber-Thomas
FWE in pickatlas vs SPM
Hello,
First, thanks for the great tool. I have been using it a lot over the past year to look at small volume corrections, primarily using uncorrected p values and a cluster extent from AFNI's 3dClustSim, which has worked great--and I have done some sanity checks to verify that I get the same results looking across a few different result viewers (pickatlas, SPM's results viewer with a mask and SVC, and Aaron Schultz's OrthoView/FIVE). Recently we've run some SVC's for the amygdala trying to use voxelwise FWE correction, but we cannot get the same results in the pickatlas and SPM. Do you have any thoughts on why this might be the case?
I have looked at the results in pickatlas and SPM (8 and 5) installs on several machines and see the same thing each time--several significant clusters for amygdala SVC FWE 0.05 in pickatlas (yay!), but none in SPM (not yay!). I make sure the stats are updated in pickatlas after I apply the mask. And in SPM I set p to FWE 0.05, then click small volume correction and select the image for bilateral amygdala that I saved from the pickatlas.
I noticed there were fewer voxels included in the spm analysis because it is applying the mask.img (which it looks like the pickatlas isn't doing?), so I created a new bilateral amygdala mask in imcalc as the intersection of the mask.img with the bilateral amygdala mask. Then I used that to run SVC in both pickatlas and SPM, which gave the same number of voxels being analyzed in both, but still the same issue of significant results in pickatlas and none in SPM.
Any insight you might have would be a great help!
Thanks,
Brad
----
Bradley C. Taber-Thomas, Ph.D.
Postdoctoral Fellow
Cognition, Affect, and Temperament Lab
bct3@psu.edu
First, thanks for the great tool. I have been using it a lot over the past year to look at small volume corrections, primarily using uncorrected p values and a cluster extent from AFNI's 3dClustSim, which has worked great--and I have done some sanity checks to verify that I get the same results looking across a few different result viewers (pickatlas, SPM's results viewer with a mask and SVC, and Aaron Schultz's OrthoView/FIVE). Recently we've run some SVC's for the amygdala trying to use voxelwise FWE correction, but we cannot get the same results in the pickatlas and SPM. Do you have any thoughts on why this might be the case?
I have looked at the results in pickatlas and SPM (8 and 5) installs on several machines and see the same thing each time--several significant clusters for amygdala SVC FWE 0.05 in pickatlas (yay!), but none in SPM (not yay!). I make sure the stats are updated in pickatlas after I apply the mask. And in SPM I set p to FWE 0.05, then click small volume correction and select the image for bilateral amygdala that I saved from the pickatlas.
I noticed there were fewer voxels included in the spm analysis because it is applying the mask.img (which it looks like the pickatlas isn't doing?), so I created a new bilateral amygdala mask in imcalc as the intersection of the mask.img with the bilateral amygdala mask. Then I used that to run SVC in both pickatlas and SPM, which gave the same number of voxels being analyzed in both, but still the same issue of significant results in pickatlas and none in SPM.
Any insight you might have would be a great help!
Thanks,
Brad
----
Bradley C. Taber-Thomas, Ph.D.
Postdoctoral Fellow
Cognition, Affect, and Temperament Lab
bct3@psu.edu
Apr 8, 2015 01:04 PM | Benjamin Wagner - UT Southwestern Medical Center
RE: FWE in pickatlas vs SPM
As this looked to be a long conversation, Brad and I discussed this
off list. Here are the take away points:
Please let me know if there are further questions/comments/concerns.
Ben
- Using SPM12's "Inclusive Masking" with an ROI is not the same as a ROI analysis. See https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1205&L=spm&D=0&P=394215 and http://www.ncbi.nlm.nih.gov/pubmed/12880848.
- When using SVC, something must survive the initial thresholding.
- SVC is only updating the FWEc p-values listed in the table, but not updating the p-values used to pick which voxels are displayed.
Please let me know if there are further questions/comments/concerns.
Ben
Apr 8, 2015 01:04 PM | Bradley Taber-Thomas
RE: FWE in pickatlas vs SPM
>
Ben,
Thanks for helping us figure this out.
I'll just add the specific way we were doing it in case others encounter
the same thing. In SPM, we were adding our ROI as an implicit mask, setting
FWE p = 0.05, then clicking SVC and selecting our ROI image to correct for
just the ROI search space. If you do this--set FWE first, then run SVC--it
looks like the results (glass brian and table) are only shown for voxels
that survived FWE prior to you adding the SVC mask. I think this is the
"glass brain isn't updated" point that Simon made in the spm listserv post
Ben sent. I hadn't realized the full import of that when I read it a while
ago. Showing only voxels that survived pre-SVC FWE correction is, of
course, not at all what one wants when running SVC, as many (probably most)
voxels will be left out due to not surviving the pre-SVC FWE threshold
(e.g., based on correcting for all voxels in the brain). The take-home I
told our lab was...just use the PickAtlas for SVC!
Thanks again for clarifying this for us and providing the great toolbox.
Best,
Brad
Ben,
Thanks for helping us figure this out.
I'll just add the specific way we were doing it in case others encounter
the same thing. In SPM, we were adding our ROI as an implicit mask, setting
FWE p = 0.05, then clicking SVC and selecting our ROI image to correct for
just the ROI search space. If you do this--set FWE first, then run SVC--it
looks like the results (glass brian and table) are only shown for voxels
that survived FWE prior to you adding the SVC mask. I think this is the
"glass brain isn't updated" point that Simon made in the spm listserv post
Ben sent. I hadn't realized the full import of that when I read it a while
ago. Showing only voxels that survived pre-SVC FWE correction is, of
course, not at all what one wants when running SVC, as many (probably most)
voxels will be left out due to not surviving the pre-SVC FWE threshold
(e.g., based on correcting for all voxels in the brain). The take-home I
told our lab was...just use the PickAtlas for SVC!
Thanks again for clarifying this for us and providing the great toolbox.
Best,
Brad