How you solve this depends a little on what you want to analyze. If CPZ is a nuisance variable, you can run a DOSS (different-offset-same-slope) analysis. In this case, it will fit CPZ to all the data and not to each group separately (use --fsgd fsgdfile doss). If you want to look at the difference in the CPZ slope between Case1 and Case2, then you'll have to set up your own design matrix. You can do this by removing the column of zeros from Xg.dat and changing your contrasts, then running mri_glmfit with --X newX.dat instead of --fsgd
On 2/18/2021 2:41 AM, 김용갑 wrote:
External Email - Use Caution
Deer Freesurfer Experts, We are trying to perform 3 group comaprison analysis in Freesurfer (mri_glmfit). Our fixed factors are "group (Case1, Case2, Healthy)" and our dependent variables are "thickness variables". We want to control "CPZ" continuous variable but CPZs of healthy group are all zero. So, it makes following errors in Freesurfer
- *ill-conditioned problem*, Freesurfer design matrix maybe be
linearly-dependent
- *Badly scaled problem*, CPZ variables are higly different between
groups (Case1, Case2 have big CPZ variables, but Healthy have all zero)
To solve this problem, we are going to make own design matrix using only*two indicator* varibles for three groups. (Freesurfer's design matrix seems to use three indicators, which may make error in this analysis) And we are going to change the contrast to suit my design matrix with two indicators. Is it possible to use two indicator design matrix in Freesurfer to perform 3 group comparison analysis(mri_glmfit)? *mri_glmfit help page* said that the design matrix can be specified in *matlab4 format*, can't in excel?, Shoud it be version 4 in the matlab? Finally, to solve badly scaled problem, can i use z-score calculated on all group samples? Best regards, Yonggab Kim
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer