So, which version did you find that ces/sqrt(cesvar) did not equal t? doug
Ben Deen wrote:
Hi Doug,
Thanks for your response. The comments below all regard question #2, is this is the only one that still concerns me.
I actually was using the nowhiten flag in the last analysis I described, which was run in version 4.5. From looking at the ces, cesvar, and t images in the contrast directories, I am finding that the value of t is not equal to ces/sqrt(cesvar), so I believe that this analysis is producing erroneous t-stats.
I've checked results from analyses run by other members of my lab, for which the entire processing stream was performed in fs-fast (my data was preprocessed in FSL and then modeled with FS-FAST, in case that matters). Their results do not have this discrepancy, indicating that something is wrong with the specific commands I'm using. Also, when I model the data using version 5.1, I also do not see this discrepancy.
Anyway, this is no longer an issue for me as I'm fine using version 5.1 for the analyses, but I thought this should be mentioned, in case in reflects a bug in the older modeling scripts.
best,
Ben
On Thu, Sep 8, 2011 at 2:29 PM, Douglas N Greve greve@nmr.mgh.harvard.edu wrote:
Ben Deen wrote:
Hi,
I have an update to this query. I tried using mkanalysis-sess instead of the ".new" version, and now I do seem to be able to get rid of intensity normalization and prewhitening (or at least, adding these flags now has an effect on betas/t-stats, and in the expected directions). The code I'm using is
mkanalysis-sess -analysis ToMLoc-1 -TR 2 -p para.para -dt blocked -fsd bold -funcstem f -runlistfile runlist1 -polyfit 0 -nc 2 -spmhrf 0 -tw 40 -no-inorm -nowhiten -noautostimdur
# mkcontrast stuff here...
selxavg3-sess -sf sessid -df sessdir -analysis ToMLoc-1
However, I have a few remaining questions about the results.
- When I look at the regressors used in the analysis, they have a
surprisingly large magnitude -- the HRF peaks at a value of ~20. Is there any way to alter the analysis such that these peaks are comparable to the values used in FSL/SPM, which is around .8, so that beta values will be directly comparable across analyses? If not, I can just scale the betas by the relevant factor, but it would be nice if this were possible.
No not really. The 5.1 version controls this better (but you still can't set a scaling parameter). You can play with the -TER parameter. By default it is set to .01, I think, which causes the amplitude to be very large. If your events appear on the TR, then try setting it to the TR.
- When I compare the results of this analysis to results obtained
with FSL or SPM, I find that the spatial pattern of beta/t-stats is very similar, but the t-stats have a noticeably reduced magnitude across the brain. Is there anything that may differ about FS-FAST's implementation of the GLM, such that t-stats would differ?
You might try turning off the whitening (--no-whiten I think).
- I saw in a ppt presentation on FS-FAST that the option -hpf can be
used to add a highpass temporal filter to the model. However, when I use this option with mkanalysis-new, the script says that the flag isn't recognized. Is there another way to add such a filter?
It's in the 5.X version. doug
Thanks in advance for any help. Cheers,
Ben
On Mon, Sep 5, 2011 at 12:00 PM, freesurfer-request@nmr.mgh.harvard.edu wrote:
Message: 1 Date: Sun, 4 Sep 2011 13:15:51 -0400 From: Benjamin Matthew Deen bdeen@mit.edu Subject: [Freesurfer] FS-FAST modeling question To: "freesurfer@nmr.mgh.harvard.edu" freesurfer@nmr.mgh.harvard.edu Message-ID:
D0755B22EA5ED64C941ED568237AEC8D1764DCC5E3@EXPO11.exchange.mit.edu Content-Type: text/plain; charset="us-ascii"
Hi,
I'm trying to run a model in fs-fast using no intensity normalization and no autocorrelation correction, for the purpose of comparing results with other software packages, and comparing results with/without autocorrelation correction. However, I haven't been able to get this to work so far. I've tried using the -noinorm and -nowhiten flags for mkanalysis-sess.new, but these don't seem to have any effect on the results: noinorm doesn't change the beta values, and nowhiten doesn't change either betas or t-stats, at all. The commands I'm using are:
mkanalysis-sess.new -analysis ToMLoc-1 -TR 2 -p para.para -dt blocked -fsd bold -funcstem f -runlistfile runlist1 -polyfit 0 -nc 2 -spmhrf 0 -tw 40 -noinorm -nowhiten
mkcontrast-sess -sf sessid -df sessdir -analysis ToMLoc-1 -contrast cope1 -a 1 -c 2 mkcontrast-sess -sf sessid -df sessdir -analysis ToMLoc-1 -contrast cope2 -a 2 -c 1 mkcontrast-sess -sf sessid -df sessdir -analysis ToMLoc-1 -contrast cope3 -a 1 -a 2 -c 0 mkcontrast-sess -sf sessid -df sessdir -analysis ToMLoc-1 -contrast cope4 -a 0 -c 1 -c 2
selxavg3-sess -sf sessid -df sessdir -analysis ToMLoc-1
Any idea what might be going wrong, or how I can get this to work? Thanks,
Ben
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html
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.