Hi Freesurfer Experts,
I have used freesurfer smoothed images (lh.thickness.fsaverage.fwhm15.gii) to perform a cortical thickness analysis. I used qdec with a surface-based morphometric analysis to compare two groups. As my discrete fixed factor I used group membership. As Nuisance factors I used age (years), sex (0/1), site(1-10) and years of education(years). I used Threshold min=2, max=5 which I assume refers to a T-stat threshold. I see an FDR correction which and the Monte Carlo Null-Z simulation.
1) Do I run both monte carlo and fdr or just one after my figure apprears? Is the monte-carlo with threshold = 1.3(0.5) similar to a clusterwise threshold with a fixed p value? Are results valid when just setting the threshold min=2, max=5 but they are just uncorrected results?
2) When running Qdec with no nusiance factors and just a discrete fixed factor of group membership, my results look almost identical to AFNI's 3dttest++ output when the tmin=2. However, when I add the nuisance factors the results between the two softare packages are different.
When nuisances covariates are added does Qdec still use a ttest or switch to some other method?
How is centering assumed between groups for covariances? Is the mean from each covariate calculated and removed from each group separately based on group membership or calculated together with all subjects and removed?
Does the analysis assume to have pooled or unpooled variance across the two groups?
3) For the viewer it displaces a logarithmic scale from what I have read in other papers. Is there a way to switch this to displace tvalue?
4) There was a previous post I saw yesterday saying qdec only uses 2 nuisance covariates and so I wanted to see if this is the case or not so I can try to understand why my afni and freesurfer results match with no covariates added but are very different once covariates are added.
Thanks, Ajay
Hello Freesurfer Experts,
I wanted to see if I am setting up qdec correctly. I am using cortical thickness as my measure. When using the Qdec gui in Freesurfer 6.0 i supplied the following files:
qdec.table.dat which contains the following: Subject Group Gender Age Site Yrs_Education
For categorical variables I created the following Group.levels HC PAT
Gender.levels Male Female
Site.levels 1 5 6
For continuous variables such as Age and Years of Education here is a significant difference between both groups.
1) Should I demean age and Years of Education before creating the table? The tutorial uses the actual values (70 73 etc) but a previous user post said that we should deman the values for covariates. What is the preferred method for the qdec gui?
2) For nuisance variables in the design matrix I selected both age and years of education. For Discrete fixed factors, I only selected group as this is the independent variables I was interested in. For categorical nuisance variables such as Gender and Site, are these ignored by the model if they are not highlighted? Although they are not my independent variable are they accounted somehow if they are not highlighted as the independent variable? They do not appear in my nuissance variable box so I guess I am not sure if they are used in the model or not.
3) The dods / doss options are missing from the Qdec gui from Freesurfer v6.0. Which is the default used in this case? Is there a way I can change this?
Thanks, Ajay
On Thu, Apr 21, 2016 at 9:15 AM, Ajay Kurani dr.ajay.kurani@gmail.com wrote:
Hi Freesurfer Experts,
I have used freesurfer smoothed images (lh.thickness.fsaverage.fwhm15.gii) to perform a cortical thickness analysis. I used qdec with a surface-based morphometric analysis to compare two groups. As my discrete fixed factor I used group membership. As Nuisance factors I used age (years), sex (0/1), site(1-10) and years of education(years). I used Threshold min=2, max=5 which I assume refers to a T-stat threshold. I see an FDR correction which and the Monte Carlo Null-Z simulation.
- Do I run both monte carlo and fdr or just one after my figure apprears?
Is the monte-carlo with threshold = 1.3(0.5) similar to a clusterwise threshold with a fixed p value? Are results valid when just setting the threshold min=2, max=5 but they are just uncorrected results?
- When running Qdec with no nusiance factors and just a discrete fixed
factor of group membership, my results look almost identical to AFNI's 3dttest++ output when the tmin=2. However, when I add the nuisance factors the results between the two softare packages are different.
When nuisances covariates are added does Qdec still use a ttest or switch to some other method?
How is centering assumed between groups for covariances? Is the mean from each covariate calculated and removed from each group separately based on group membership or calculated together with all subjects and removed?
Does the analysis assume to have pooled or unpooled variance across the two groups?
- For the viewer it displaces a logarithmic scale from what I have read
in other papers. Is there a way to switch this to displace tvalue?
- There was a previous post I saw yesterday saying qdec only uses 2
nuisance covariates and so I wanted to see if this is the case or not so I can try to understand why my afni and freesurfer results match with no covariates added but are very different once covariates are added.
Thanks, Ajay
In general, qdec should be used to "play" with your data in an interactive way. For many designs, you may need to run the command-line stream. This is probably one of them. If you have continuous nuisance variables, then you should run the analysis using DODS and test for an interaction of the nuisance variables and the factors of interest. If there is no interaction, then use DOSS and test for the difference in your discrete factors. In neither case do you need to demean. If there is an interaction, then you should demean (and talk to a statistician). QDEC uses DODS exclusively; DOSS is not available with QDEC.
On 04/22/2016 09:23 AM, Ajay Kurani wrote:
Hello Freesurfer Experts,
I wanted to see if I am setting up qdec correctly. I am using cortical thickness as my measure. When using the Qdec gui in Freesurfer 6.0 i supplied the following files:
qdec.table.dat which contains the following: Subject Group Gender Age Site Yrs_Education
For categorical variables I created the following Group.levels HC PAT
Gender.levels Male Female
Site.levels 1 5 6
For continuous variables such as Age and Years of Education here is a significant difference between both groups.
- Should I demean age and Years of Education before creating the
table? The tutorial uses the actual values (70 73 etc) but a previous user post said that we should deman the values for covariates. What is the preferred method for the qdec gui?
- For nuisance variables in the design matrix I selected both age and
years of education. For Discrete fixed factors, I only selected group as this is the independent variables I was interested in. For categorical nuisance variables such as Gender and Site, are these ignored by the model if they are not highlighted? Although they are not my independent variable are they accounted somehow if they are not highlighted as the independent variable? They do not appear in my nuissance variable box so I guess I am not sure if they are used in the model or not.
- The dods / doss options are missing from the Qdec gui from
Freesurfer v6.0. Which is the default used in this case? Is there a way I can change this?
Thanks, Ajay
On Thu, Apr 21, 2016 at 9:15 AM, Ajay Kurani <dr.ajay.kurani@gmail.com mailto:dr.ajay.kurani@gmail.com> wrote:
Hi Freesurfer Experts, I have used freesurfer smoothed images (lh.thickness.fsaverage.fwhm15.gii) to perform a cortical thickness analysis. I used qdec with a surface-based morphometric analysis to compare two groups. As my discrete fixed factor I used group membership. As Nuisance factors I used age (years), sex (0/1), site(1-10) and years of education(years). I used Threshold min=2, max=5 which I assume refers to a T-stat threshold. I see an FDR correction which and the Monte Carlo Null-Z simulation. 1) Do I run both monte carlo and fdr or just one after my figure apprears? Is the monte-carlo with threshold = 1.3(0.5) similar to a clusterwise threshold with a fixed p value? Are results valid when just setting the threshold min=2, max=5 but they are just uncorrected results? 2) When running Qdec with no nusiance factors and just a discrete fixed factor of group membership, my results look almost identical to AFNI's 3dttest++ output when the tmin=2. However, when I add the nuisance factors the results between the two softare packages are different. When nuisances covariates are added does Qdec still use a ttest or switch to some other method? How is centering assumed between groups for covariances? Is the mean from each covariate calculated and removed from each group separately based on group membership or calculated together with all subjects and removed? Does the analysis assume to have pooled or unpooled variance across the two groups? 3) For the viewer it displaces a logarithmic scale from what I have read in other papers. Is there a way to switch this to displace tvalue? 4) There was a previous post I saw yesterday saying qdec only uses 2 nuisance covariates and so I wanted to see if this is the case or not so I can try to understand why my afni and freesurfer results match with no covariates added but are very different once covariates are added. Thanks, Ajay
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
On 04/21/2016 10:15 AM, Ajay Kurani wrote:
Hi Freesurfer Experts,
I have used freesurfer smoothed images (lh.thickness.fsaverage.fwhm15.gii) to perform a cortical thickness analysis. I used qdec with a surface-based morphometric analysis to compare two groups. As my discrete fixed factor I used group membership. As Nuisance factors I used age (years), sex (0/1), site(1-10) and years of education(years).
Not sure what you mean here. Are you treating sex and site as a discrete variables or continuous? Continuous does not make much sense.
I used Threshold min=2, max=5 which I assume refers to a T-stat threshold. I see an FDR correction which and the Monte Carlo Null-Z simulation.
- Do I run both monte carlo and fdr or just one after my figure apprears?
You would not run them in any combined way if that is what you mean. You would run them independently
Is the monte-carlo with threshold = 1.3(0.5) similar to a clusterwise threshold with a fixed p value?
It is a clusterwise correction. It uses a monte carlo simulation to determine the chance probability of a cluster of a given size. Other packages use gaussian random field theory to do the same thing (but it makes a lot of assumptions).
Are results valid when just setting the threshold min=2, max=5 but they are just uncorrected results?
what do you mean? you were just talking about cluster correction
- When running Qdec with no nusiance factors and just a discrete
fixed factor of group membership, my results look almost identical to AFNI's 3dttest++ output when the tmin=2. However, when I add the nuisance factors the results between the two softare packages are different.
When nuisances covariates are added does Qdec still use a ttest or switch to some other method?
Still uses t-test. Probably the design matrix is changing between the two packages.
How is centering assumed between groups for covariances? Is the mean from each covariate calculated and removed from each group separately based on group membership or calculated together with all subjects and removed?
No centering is done. You have to do this yourself.
Does the analysis assume to have pooled or unpooled variance across the two groups?
It pools variance across the groups.
- For the viewer it displaces a logarithmic scale from what I have
read in other papers. Is there a way to switch this to displace tvalue?
Not in qdec. you can run freeview outside of qdec to view the tmaps
- There was a previous post I saw yesterday saying qdec only uses 2
nuisance covariates and so I wanted to see if this is the case or not so I can try to understand why my afni and freesurfer results match with no covariates added but are very different once covariates are added.
yes it is. You can also look at the design matrix (Xg.dat) in the output folder and compare that to whatever afni is using
Thanks, Ajay
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Ajay Kurani -
Douglas N Greve