Tudor,
To follow-up on the repeated analysis question: each 'Design' can be named by changing the 'Design Name' text box on the 'Design' tab from something other than 'Untitled', which has the effect of saving all results to a directory of that name, so that you can revisit them later (by loading results in freeview for instance).
Nick
On Thu, 2013-04-18 at 20:12 -0400, Douglas N Greve wrote:
Here's part 1. I'll write until I have to catch my shuttle, answer the rest tomorrow ...
On 04/18/2013 05:27 PM, Tudor Popescu wrote:
Apologies for my previous long email; if anyone gets a chance to look over the questions I'd be really grateful! Many thanks indeed, Tudor
On 16 April 2013 19:40, Tudor Popescu <tudor3@gmail.com mailto:tudor3@gmail.com> wrote:
Thanks Nick, (and thanks Doug too for the answer to question 2.) It must indeed have been a disk-space issue, as running the -qcache again, after clearing up some space, produced all the expected .mgh files If I can follow up on two of my previous questions: 3) Not sure I understand your answer. So it seems discrete variables, such as gender, cannot be taken as covariates or nuisance variables, only as factors. But users might want to take some discrete variables as covariates, rather than as factors, as I might not be interested in their direct effect on the brain measure but simply want to parcel out the variance that they contribute. Are you suggesting that they should be taken as factors even if they aren't of interest?Yes. There is no real distinction between between something that is of interest and something that is not. The way the software is set up, the continuous factors can be listed as nuisance and get around the limitation of only have two covariates.
4) Does the ideal value of FWHM depend on the blob size in the sense that if one expects small blobs in the results (how small?), then one should use small FWHMs in QDEC, and large FWHMs if expecting large blobs?Yes.
I apologise for the amount of questions I keep asking, but I have a few more: A) When trying repeated analyses (designs) in QDEC, do I need to delete the output files of previous analyses, and/or restart QDEC every time? Or are the results of each analysis displayed correctly independently of previously-made analyses in the same QDEC session? I'm asking because I see that, once the "Set using FDR" button is pressed, the corrected t threshold remains in use for subsequent analyses, but after restarting QDEC and redoing the last analysis, the t threshold is no longer the sameNot sure, have to check. I think you can name the output folders differently for each analysis. If you do not change the name, then the output is totally deleted and recreated.
B) Must all QDEC analyses always be done for the two hemispheres separately? Is there no analysis that can be done on the whole brain, such that the t-value thresholds are FDR-corrected at the whole-brain level?There is none.
C) I would like to extract the cortical thickness of several cortical ROIs including the IPS, IFG and SPL; I didn’t know whether the Desikan-Killiany or the Destrieux atlas would be more appropriate, but I tried the command given here <http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/AnatomicalROI#TableoftheaveragethicknessofeachcorticalparcellationintheDestrieuxatlas>, hoping to obtain a table with the thickness of all ROIs from the parcelation corresponding to the Destrieux atlas. However, although the command results in the message " lh.aparc.a2009.thickness.table", I found no such file anywhere in my $SUBJECTS_DIRDid it go into the directory that you ran the command from?
D) How should a regression-type analysis be made in QDEC, i.e. one where I have a continuous predictor such as behavioural score, whose correlation with the brain measure (cortical thickness) I want to compare between my two groups? Because of QDEC's preference for discrete variables as factors, it seems that only ANOVA-type analyses can be done (i.e. effect of discrete factor(s) on brain measure), rather than regression-type (i.e. correlation between continuous factor and brain measure)Enter it in as a covariate (continuous factor). QDEC will automatically produce a contrast testing the difference in thickness/score slopes (ie, an interaction between group factor and continuous factor).
E) The average brain with inflated cortex that results are projected on – is this the same average that is normally used in most papers, or does the inflating algorithm differ? And is the colour-coding the same (dark gray = sulci, light gray = gyri)?It is mostly the same. Some papers may display on the pial or the white or do a custom inflation to keep some of the gyral shape instead of having it so smooth. The colors I assume are the same.
whew! made it through all of them before my shuttle. doug
On 15 April 2013 23:52, Nick Schmansky <nicks@nmr.mgh.harvard.edu <mailto:nicks@nmr.mgh.harvard.edu>> wrote: Tudor, In the recon-all.log, it has this line: ERROR: writing lh.jacobian_white.fwhm15.fsaverage.mgh but earlier in the log it saved lh.jacobian_white.fwhm10.fsaverage.mgh correctly, so this indicates to me that it might have run out of disk space. is that the case? to answer the others: 2. not sure 3. you can select discrete can a regular variate along with your main variate. 'nuisance' variates are like any other. 4. depends on the expected 'blob' size 5. the selection of fwhm in qdec corresponds directly with the values selected by qcache (they are one-to-one related, ie the 10mm fwhm values created by qcache are used by the 10mm fwhm selection in qdec). Nick On Mon, 2013-04-15 at 18:38 +0200, Tudor Popescu wrote: > Dear experts, > > Upgrading to 5.2.0 stopped QDEC (specifically, mri_concat) from > misbehaving, and so after running a first whole-brain group cortical > thickness analysis on my structural data, I have some questions: > > 1. After running recon-all with the –qcache flag (i.e. presmoothing), > files of the type lh.thickness.*.mgh were created for all 38 subjects > (19 in each group), however files of the type rh.thickness.*.mgh were > not created for 5 out of the 19 subjects of the first group. Log files > recon-all-status.log and recon-all.log (attached, for one of those 5 > subjects) both mention that the process ran on Mar22nd ended with > errors, although I can't quite see what that was > > > > 2. When I take age as a continuous factor (covariate), the list of > clusters in my results look dramatically different from the clusters > that I get for the same contrast ran without the covariate. Why is > that, given that normally adding a covariate very rarely changes the > results by a great deal? Also in my case, I had quite a narrow (and > well-balanced between the groups) age range! > > > > 3. I know that discrete factors cannot be taken as nuisance factors, > but it seems they also can't be taken as covariates. How does one, > then, control for the effects of e.g. gender in a group comparison? > > > 4. When should values other than 10 be used for the FWHM parameter of > the smoothing? > > > 5. How come QDEC allows you to set the FWHM parameter, when in fact it > is also set in the qcache stage of recon-all, which precedes QDEC? > > > Many thanks in advance!! > > Tudor > _______________________________________________ > Freesurfer mailing list > Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu> > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.