I'm getting some puzzling discrepancies when comparing the ocn and the cluster outputs. The summary file indicates there should be five significant clusters (see below). The cluster file only shows four of the clusters, although I'm guessing that the supramarginal one is not being displayed because it did not pass the three-space Bonferroni (so the cluster file only includes clusters that pass that additional correction)? What is really confusing me is that the ocn file does display the supramarginal cluster but does not show the rostral anterior cingulate cluster. How does this work? See attached images below.
This is with Freesurfer 5.20 on OS X 10.8.3.
Cheers!
Joe
tksurfer fsaverage lh inflated -aparc -overlay CPA.sm05.lh/CS-v-OTHERS/CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.ocn.nii.gz tksurfer fsaverage lh inflated -aparc -overlay CPA.sm05.lh/CS-v-OTHERS/CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.cluster.nii.gz
# Cluster Growing Summary (mri_surfcluster) # $Id: mri_surfcluster.c,v 1.51.2.3 2012/05/31 22:10:05 greve Exp $ # $Id: mrisurf.c,v 1.693.2.4 2012/11/07 22:38:00 nicks Exp $ # CreationTime 2013/05/26-22:53:59-GMT # cmdline mri_surfcluster --in CS-v-OTHERS.wls/osgm/sig.nii.gz --csd /Applications/freesurfer/average/mult-comp-cor/fsaverage/lh/cortex/fwhm12/pos/th20/mc-z.csd --mask CS-v-OTHERS.wls/mask.nii.gz --cwsig CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.cluster.nii.gz --vwsig CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.voxel.nii.gz --sum CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.cluster.summary --ocn CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.ocn.nii.gz --oannot CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.ocn.annot --annot aparc --csdpdf CS-v-OTHERS.wls/osgm/cache.th20.pos.pdf.dat --cwpvalthresh .05 --o CS-v-OTHERS.wls/osgm/cache.th20.pos.sig.masked.nii.gz --no-fixmni --bonferroni 3 --surf white # cwd /Volumes/Data/CP1/fsfast/secondLevel/CPA.sm05.lh/CS-v-OTHERS # sysname Darwin # hostname megajoe.home # machine x86_64 # FixVertexAreaFlag 1 # FixSurfClusterArea 1 # # Input CS-v-OTHERS.wls/osgm/sig.nii.gz # Frame Number 0 # srcsubj fsaverage # hemi lh # surface white # annot aparc # SUBJECTS_DIR /Applications/freesurfer/subjects # SearchSpace_mm2 65416.6 # SearchSpace_vtx 148649 # Bonferroni 3 # Minimum Threshold 2 # Maximum Threshold infinity # Threshold Sign pos # AdjustThreshWhenOneTail 1 # CW PValue Threshold: 0.05 # Area Threshold 0 mm^2 # CSD thresh 2.000000 # CSD nreps 10000 # CSD simtype null-z # CSD contrast NA # CSD confint 90.000000 # Overall max 5.96965 at vertex 140670 # Overall min -3.06064 at vertex 162717 # NClusters 5 # Total Cortical Surface Area 65416.6 (mm^2) # FixMNI = 0 # # ClusterNo Max VtxMax Size(mm^2) MNIX MNIY MNIZ CWP CWPLow CWPHi NVtxs Annot 1 5.970 140670 2209.93 -5.9 40.9 -0.0 0.00030 0.00000 0.00060 4223 rostralanteriorcingulate 2 4.146 49359 1566.01 -4.0 -19.8 35.8 0.00030 0.00000 0.00060 3668 posteriorcingulate 3 3.860 111375 446.35 -44.6 -37.3 25.1 0.02558 0.02204 0.02911 1037 supramarginal 4 3.711 128614 652.01 -24.4 -26.8 -24.3 0.00150 0.00060 0.00240 1539 parahippocampal 5 3.647 25172 1334.99 -48.7 -50.3 17.3 0.00030 0.00000 0.00060 2700 inferiorparietal
On May 31, 2013, at 4:44 PM, Joseph Dien jdien07@mac.com wrote:
It looks like the corrected vertex p-values (ex: cache.th13.abs.sig.voxel.nii.gz) are only available for the surface-based lh and rh spaces. For the subcortical volume-based analysis I don't see the corresponding corrected voxel p-values being available?
On May 31, 2013, at 2:46 PM, Joseph Dien jdien07@mac.com wrote:
On May 31, 2013, at 12:11 PM, Douglas N Greve greve@nmr.mgh.harvard.edu wrote:
On 05/31/2013 01:49 AM, Joseph Dien wrote:
I was able to make more progress so I'm mostly good at this point but I have a remaining question:
I assume the contents of sig.nii.gz (which I assume are the vertex p-values) are not FWE corrected. Is it possible to get FWE-corrected vertex p-values? Or are only clusterwise corrections available?
There should be something like cache.th13.abs.sig.voxel.mgh which is corrected on a voxelwise basis (the th13 is just part of the name but it should be the same regardless of the threshold you choose) doug
Excellent! Thanks! :)
Thanks again for your patience!
Joe
On May 30, 2013, at 4:37 PM, Joseph Dien <jdien07@mac.com mailto:jdien07@mac.com> wrote:
Just to make sure I'm doing this right, I'm going to summarize what I've taken away from your answers and to ask some new questions. In order to present the results, I need two things:
- A set of histograms (with error bars) for each cluster figure to
show the % signal change for each of the four contrasts of interest. The cache.th20.pos.y.ocn.dat file only gives it for the condition where the cluster was significant so I can't use that. So I could use mri_label2vol to convert cache.th20.neg.sig.ocn.annot from the group level analysis to generate a mask for each cluster of interest. Then I could extract the value of the voxels from each subject's cespct file for each contrast, average them across the cluster ROI, then average them across each subject, to generate the histogram? This would suffice to give me the %age signal change? I would be doing these computations in Matlab using MRIread.
- A results table with the headings:
Cluster p (FWE corrected) Cluster size Peak Voxel p (FWE corrected) Peak Voxel T Peak Voxel Coords BA Anatomical Landmark
I can get the first two from the cache.th20.pos/neg.sig.cluster.summary files from the group level analysis. I can get the peak voxel coordinates from the summary files as well. I can use this to get the peak voxel p from the group level sig.nii.gz file. Is this FWE corrected? If not, how can I get this information? I can use these coordinates to get the peak voxel T by getting the value from the group level F.nii.gz file and taking its square root. How can I get the sign of the T statistic? I can use the Lancaster transform to convert the MNI305 peak voxel coordinates into the Atlas coordinates to look up the putative BA and landmarks (unless there is a better way with Freesurfer? I'm seeing some references to some BA labels in the forum but it doesn't look like this is a complete set yet?).
Sorry for all these questions! I got some nice results from FSFAST and would like to get them written up.
Cheers!
Joe
On May 29, 2013, at 10:53 PM, Douglas Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> wrote:
On 5/29/13 10:42 PM, Joseph Dien wrote: > > On May 29, 2013, at 11:40 AM, Douglas N Greve > <greve@NMR.MGH.HARVARD.EDU mailto:greve@NMR.MGH.HARVARD.EDU> wrote: > >> Hi Joe, >> >> On 05/29/2013 01:00 AM, Joseph Dien wrote: >>> I need to extract the beta weights from a cluster identified with >>> FS-Fast in order to compute percentage signal change. >>> >>> 1) I see a file called beta.nii.gz that appears to have the beta >>> weight information. It has a four dimensional structure and the >>> fourth dimension appears to be the beta weights. Is there an index >>> somewhere as to which beta weight is which? Or if not, how are they >>> organized? >> For the first level analysis, the first N beta weights correspond >> to the >> N conditions in the paradigm file. The rest are nuisance variables. >>> > > Ah, very good! In order to compute the percent signal change > statistic (I'm following the MarsBaR approach: > http://marsbar.sourceforge.net/faq.html#how-is-the-percent-signal-change-cal...) > I'm also going to need the beta weights for the session mean > regressors. How are the nuisance regressors organized? You can just use the meanfunc.nii.gz. Also, each contrasts is computed as the simple contrast (ces) and as a percent of the baseline at the voxel (cespct, cesvarpct). > >>> 2) In order to extract the cluster, it looks like I would >>> use mri_label2vol to convert cache.th20.neg.sig.ocn.annot into a >>> volume where the voxels are tagged with the number of the >>> corresponding cluster. >> Is that from a group analysis? >>> > > Yes, that's right. > >>> I could then use that to generate masks to extract the information I >>> need for each cluster from beta.nii.gz. >> If this is from a group analysis, then there should already be a file >> there (something.y.ocn.dat) that has a value for each subject in the >> rows and a value for each cluster in the columns. >>> > > I see it. Are these values already scaled as percent signal > change? If so, that would be wonderful! :) Only if you specified it when you ran isxconcat-sess. Note that the "non-scaled" values are actually scaled to percent of grand mean intensity. > >>> Is that correct? >>> >>> 3) The final information that I would need is the canonical hrf >>> shape >>> generated by FSFAST for a single event. I guess I could generate >>> that >>> by setting up a dummy analysis run with a single event of the >>> desired >>> duration and then look in the X variable in the resulting X.mat file? >> try this >> plot(X.runflac(1).flac.ev(2).tirf, X.runflac(1).flac.ev(2).Xirf) >>> > > Perfect! :) > >>> Sorry for all the questions! >>> >>> Joe >>> >>> >>> >>>
--------------------------------------------------------------------------------
Joseph Dien, Senior Research Scientist University of Maryland
E-mail: jdien07@mac.com Phone: 301-226-8848 Fax: 301-226-8811 http://joedien.com//
