Yes, that all looks correct. The only other thing you may want to do is to demean (ie, remove the mean from) your continuous variables. When you test for a difference in thickness (as opposed to slope), you are implicitly testing at your variables=0. Eg, if age is one of your variables, then you are testing at age=0 (birth). If there is not an interaction between age and group, then this is not a problem. If there is an interaction, then the difference will depend on what age you compute the difference. So, it's a good idea to test for an interaction between group and your continuous variables. The contrasts would basically look the same (ie, four 0.25s, four -0.25s, and 24 zeros), they will just be rearranged differently.
doug
Habets P (SP) wrote:
Hi Doug,
Just a final check that my fsgd and contrasts are properly defined. This is now my fsgd file
GroupDescriptorFile 1 Title fsgd_pat_con.txt MeasurementName CorticalThickness
Class controls_male_mdeft Class controls_female_mdeft Class controls_male_adni Class controls_female_adni Class patients_male_mdeft Class patients_female_mdeft Class patients_male_adni Class patients_female_adni
Variables Age Education Handedness
Input C001 controls_female_mdeft 25 6 100 Input C002 controls_female_mdeft 27 6 100 Input C003 controls_male_mdeft 46 5 -89.47368622 Input C004 controls_female_mdeft 43 4 100 Input C005 controls_male_mdeft 25 8 100 Input C006 controls_male_mdeft 19 7 100 Input P001 patients_male_mdeft 32 7 -100 Input P002 patients_female_mdeft 46 3 -23.07691956 Input P003 patients_male_mdeft 26 6 100 Input P004 patients_female_mdeft 19 6 60 Input P005 patients_male_mdeft 23 5 100 Input P006 patients_male_mdeft 21 3 100
My hypothesis is than patients differ from controls with regard to cortical thickness, regressing out the effects for gender scantype age educational level and handedness. So my contrast is: 0.25 0.25 0.25 0.25 -0.25 -0.25 -0.25 -0.25 and 24 zero's (together 32 regressors)
Is this properly defined? Thx! Bw Petra
-----Original Message----- From: freesurfer-bounces@nmr.mgh.harvard.edu [mailto:freesurfer-bounces@nmr.mgh.harvard.edu] On Behalf Of Douglas N Greve Sent: Friday, February 11, 2011 5:10 PM To: Habets P (SP) Cc: freesurfer@nmr.mgh.harvard.edu; Gronenschild Ed (NP) Subject: Re: [Freesurfer] Contrasts for GLM DODS with both discrete as continues covariates
Hi Petra, you have mixed up a class with the level of a factor (an easy thing to do). From your FSGD, it looks like you would need 8 classes, each class would be a combination of a level from each factor. So one class would be control-male-adni, etc. Once you have a proper FSGD file, FreeSurfer will create the proper design matrix. After you looked at the FSGD examples on the wiki? Do a search for FSGD, and they will pop up.
doug
Habets P (SP) wrote:
I want to run a GLM DODS in Freesurfer to test whether patients differ from controls with regard to cortical thickness, meanwhile I want to control for several variables, both discrete as continues.
This is my manually made fsgd file
GroupDescriptorFile 1
Title fsgd pat con.txt
MeasurementName Cortical Thickness
Class controls
Class patients
Class female
Class male
Class mdeft
Class adni
Variables Age Education Handedness
input C002 controls female mdeft 27 6 100
input C003 controls male mdeft 46 5 -89.47368622
input C004 controls female mdeft 43 4 100
input P001 patients male mdeft 32 7 -100
input P003 patients male mdeft 26 6 100
input P004 patients female mdeft 19 6 60
The problem I have is how to specify contrasts, on the Freesurfer website I found that :
For DODS, the number of regressors is Nc*(Nv+1)
In my case that would be 6* (3+1) = 24
Would the contrast than be:
1 -1 and 22 zero's
Does Freesurfer automatically create the correct design matrix using the fsgd file? Because it is not clear to me how I should create the design matrix with some discrete variables as confounders.
Bw Petra Habets
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer