wow, that is a big difference. Can you just subtract the y.mgh of the 4.5 version from that of the 5.0 version and see where the differences are? Also, if you take the y.mgh of 4.5 and run it in the mri_glmfit of 5.0, do you see the same FWHM as in the mri_glmfit of 4.5? doug
On 10/01/2012 07:38 AM, Ed Gronenschild wrote:
Doug,
For the PCA residuals we found 37% for the first component. The same analysis has been done for FreeSurfer v4.5.0, and we found a residual FWHM of 15.27 and 30% for the first PCA component. The time courses showed some outliers for both versions, very much dependent on the position on the surface, i.e., each time different subjects were outside the confidence limits. Summarizing: v4.5.0: residualFWHM = 15.27; 1st PCA 30% v5.0.0: residualFWHM = 34.00; 1st PCA 37% (number of subjects is 53).
So, it is a puzzle why the residual FWHM is that large for v5.0.0.
Ed
On 1 Oct 2012, at 9:15, Ed Gronenschild wrote:
On 28 Sep 2012, at 18:00, freesurfer-request@nmr.mgh.harvard.edu mailto:freesurfer-request@nmr.mgh.harvard.edu wrote:
Message: 27 Date: Fri, 28 Sep 2012 10:03:47 -0400 From: Douglas N Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> Subject: Re: [Freesurfer] Fmri_glmfit-sim with pre-cached fails To: freesurfer@nmr.mgh.harvard.edu mailto:freesurfer@nmr.mgh.harvard.edu Message-ID: <5065AE43.7060103@nmr.mgh.harvard.edu mailto:5065AE43.7060103@nmr.mgh.harvard.edu> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hmmm, that's a lot, but I'm very surprised that the final FWHM is over 30. How many subjects do you have? There might be one that is messed up. You can check this by loading the y.mgh as a "time course" in tksurfer. It will give you a graph with each subject. Finally, you can run mri_glmfit with the --pca option to get a PCA of the residuals. This will create a folder called pca-eres. Look at the stats.dat and see how much variance is explained by the first component (first row, 4th column). doug
On 09/28/2012 03:31 AM, Ed Gronenschild wrote:
Hi Doug,
For mri_glmfit I used a cortical thickness smoothed with a FWHM of 20, resultingfrom using the -qcache option in recon-all.
Ed
On 27 Sep 2012, at 18:00, freesurfer-request@nmr.mgh.harvard.edu mailto:freesurfer-request@nmr.mgh.harvard.edu mailto:freesurfer-request@nmr.mgh.harvard.edu wrote:
Message: 16 Date: Thu, 27 Sep 2012 11:29:06 -0400 From: Douglas N Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> Subject: Re: [Freesurfer] mri_glmfit-sim with pre-cached fails To: freesurfer@nmr.mgh.harvard.edu mailto:freesurfer@nmr.mgh.harvard.edu mailto:freesurfer@nmr.mgh.harvard.edu Message-ID: <506470C2.5060906@nmr.mgh.harvard.edu mailto:506470C2.5060906@nmr.mgh.harvard.edu mailto:506470C2.5060906@nmr.mgh.harvard.edu> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hi Ed, when I created the cache, I only went up to 30mm FWHM thinking that no one would ever go above it, or maybe should never go above it. I can create tables at higher smoothing levels, but 34mm is a lot of smoothing. How much did you apply to the data? doug
On 09/27/2012 05:48 AM, Ed Gronenschild wrote:
Hi,
I want to apply a cluster correction with mri_glmfit-sim with the --cache option: mri_glmfit-sim --glmdir my_glmdir --cache 1.3 abs It fails with the error message that it can't find the file: FSDir/average/mult-comp-cor/fsaverage/lh/cortex/fwhm34/abs/th1.3/mc- z.csd (FSDir is the directory where FreeSurfer is located, version 5.0.0, Mac OSX10.5) Looking at the content of the ....fsaverage/lh/cortex directory than indeed, only the directories fwhm01, fwhm02, ..., fwhm30 are present. In the log file from mri_glmfit I can read that the residualFWHM is 34.005014. Since this FWHM is automatically derived by mri_glmfit, does it mean that I can never use the --cache option in this case?
Cheers,
Ed
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer