Hi Doug,
Thanks for looking into it. I'm still at newbie at this, so could you please clarify -- which matrix is badly conditioned? And what do the columns (e.g. 5th column that you mentioned) refer to? You don't mean the qdec.table.dat file that I input into qdec, do you?
I'm just trying to understand what makes the matrix badly conditioned in this case.
Thanks again! Maria
On Jul 8, 2013, at 3:47 PM, Douglas N Greve wrote:
Hi Maria, I've tracked down the source of the difference. In 5.0, mri_glmfit simply takes whatever matrix you give it regardless of how well or badly conditioned it is. In your case, your matrix is badly conditioned because the scaling between the columns is very different. Eg, the 5th column is almost 2000 time that of the 3rd column. This creates difficulties when trying to invert the matrix. In 5.1, I introduced code to rescale each column so that all columns are about the same magnitude. This makes the inversion better behaved and more accurate. The scale is then reapplied once the beta's are computed. So, in your case the 5.0 was not inverting the matrix very accurately. The 5.3 version is the one you want.
doug
On 07/08/2013 09:48 AM, Maria Kharitonova wrote:
Hi, sorry I must have missed the response to this a few weeks back. Yes, I'm doing an identical analysis: comparing 2 groups of kids (controls and ADHD) in cortical thickness, controlling for age and demeaned ICV (nuisance). So all the parameters are the same. I'm attaching 3 screen shots: group differences using 5.0 before using Monte Carlo correction (lots of clusters), group differences before correction using 5.3 (few small clusters), and group differences using 5.0 after correction (frontal cluster). there was nothing sig. remaining after correcting for multi. comparisons using 5.3. I'm puzzled about why the differences are so large. I selected DODS in 5.0, so everything should be the same? I triple-checked to make sure the other differences (e.g. FWHM, etc.) were identical (left at default).
Thanks! Maria
Message: 10 Date: Tue, 18 Jun 2013 23:34:10 -0700 From: Douglas Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> Subject: Re: [Freesurfer] qdec differences in FS 5.0 vs 5.3 To: freesurfer@nmr.mgh.harvard.edu mailto:freesurfer@nmr.mgh.harvard.edu Message-ID: <51C150E2.1080604@nmr.mgh.harvard.edu mailto:51C150E2.1080604@nmr.mgh.harvard.edu> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hi Maria, can you elaborate on exactly what you have done? Is the data the same and you are only changing the version of QDEC? What is the nature of the difference? Can you send pics or stats? doug
On 6/18/13 6:06 AM, Maria Kharitonova wrote:
Hi,
I'm learning how to use QDEC and was comparing results that it produces under FresSurfer 5.0 vs 5.3. I know the 5.0 has an option for either DODS or DOSS method of producing a design matrix, while the 5.3 only has DODS. This suggests that results should be identical (or at least similar) across versions if I select DODS. However, I'm getting very different results based on the version I select: complete lack of differences across groups in 5.3 and a predicted pattern of differences in 5.0 using DODS. I'd like to understand where the differences are stemming from -- any ideas?
Thanks in advance! Maria
Maria Kharitonova, Ph.D. Postdoctoral Research Fellow Laboratories of Cognitive Neuroscience Boston Children's Hospital
-- 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: https://gate.nmr.mgh.harvard.edu/filedrop2 www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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.