Hello Sir/Madam,
I am currently running a statistical test using the FSGD method. I am encountering a problem in relation to the creation of the glmdir folder when I run the mri_glmfit function. I recieve an ERROR message that tells me that the design matrix (automatically generated) is ill conditioned or badly scaled, condo = 10565.7. All the steps before creating the glmdir folder have been done in accordance with the FSGD group analysis tutorial on the freesurfer wiki.
The command line is : mri_glmfit --y lh.thomas4.10.mgh --fsgd thomas4.txt dods --C t4.txt --surf fsaverage lh --cortex --glmdir lh.thomas4.glmdir
The FSGD file = thomas4.txt ( contains 2 groups and 5 variables) The data matrix = Xg.dat The contrast matrix = t4.txt
There are 3 columns of variables that have discreet values of either 1 or 2. However I have tried this with another FSGD file (thomas3.txt) that contained 4 variables of which 2 were continous and the other 2 were in 1s' and 2s'.
The contrast matrix of thomas3.txt = C.dat The design matrix of thomas3.txt = Xg1.dat
I was able to generate a glmdir using mri_glmfit with thomas3.txt and generate a surface for tksurfer to use. The file created was the sig.mgh file which was then compared with the fsaverage files. However once I added another variable titled "voxel" in thomas4.txt, I received the error message.
I did a correlation of the columns containing 1s and 2s in the FSGD file thomas4.txt using Matlab and have not found any significant correlation with each other.
Please do advise on how to solve this issue and also from where the problem could stem from as I am unsure as to the nature of the problem.
Thank You for your patience and assistance
Joshua
Hi Joshua, you should start by not coding gender and handedness as continuous variables. In doing this, you are saying that you expect right handers to have 2x the thickness as left handers (or females to have 2x the thickness as males). You need 8 classes. If your "voxel" variable is discrete, then you will need 16 classes. doug
On 04/04/2013 05:22 AM, Joshua deSouza wrote:
Hello Sir/Madam,
I am currently running a statistical test using the FSGD method. I am encountering a problem in relation to the creation of the glmdir folder when I run the mri_glmfit function. I recieve an ERROR message that tells me that the design matrix (automatically generated) is ill conditioned or badly scaled, condo = 10565.7. All the steps before creating the glmdir folder have been done in accordance with the FSGD group analysis tutorial on the freesurfer wiki.
The command line is : mri_glmfit --y lh.thomas4.10.mgh --fsgd thomas4.txt dods --C t4.txt --surf fsaverage lh --cortex --glmdir lh.thomas4.glmdir
The FSGD file = thomas4.txt ( contains 2 groups and 5 variables) The data matrix = Xg.dat The contrast matrix = t4.txt
There are 3 columns of variables that have discreet values of either 1 or 2. However I have tried this with another FSGD file (thomas3.txt) that contained 4 variables of which 2 were continous and the other 2 were in 1s' and 2s'.
The contrast matrix of thomas3.txt = C.dat The design matrix of thomas3.txt = Xg1.dat
I was able to generate a glmdir using mri_glmfit with thomas3.txt and generate a surface for tksurfer to use. The file created was the sig.mgh file which was then compared with the fsaverage files. However once I added another variable titled "voxel" in thomas4.txt, I received the error message.
I did a correlation of the columns containing 1s and 2s in the FSGD file thomas4.txt using Matlab and have not found any significant correlation with each other.
Please do advise on how to solve this issue and also from where the problem could stem from as I am unsure as to the nature of the problem.
Thank You for your patience and assistance
Joshua
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Joshua deSouza