External Email - Use Caution
Hi all,
I took the freesurfer course in the recent past, and I am now beginning to run some glm thickness analyses!
Currently I am doing a thickness analysis looking at the difference between two groups while controlling for age, and two continuous measures of depression and anxiety. The general commands I run are as follows:
mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi lh --o lh.[filename].thickness.10.mgh mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi rh --o rh.[filename].thickness.10.mgh mri_glmfit --y lh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage lh --glmdir lh_[filename]_fwhm10.glmdir mri_glmfit --y rh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage rh --glmdir rh_[filename]_fwhm10.glmdir mri_glmfit-sim --glmdir lh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces mri_glmfit-sim --glmdir rh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces
I noticed that in some analyses there would be no significant clusters when the threshold is set at 2.0 (p<0.01) but if you were to run the same analyses with a threshold at 2.3 (p<0.005) there would be a significant cluster, in the lingual for example. If I go one step further and set the threshold to 3.0 there is also a significant cluster - however the cluster is now in the isthmuscingulate.
Am I doing something incorrectly or is this possible? And if so, could you help me understand what is going on here.
Best,
Andrew
External Email - Use Caution
assuming that all preprocessing steps, fsgd file and contrasts are correct, the choice between doss and dods could explain some of the variability you're seeing in your results. Consider running both models and possibly additional validation to see which model better fits your data. Each model assumes a different relationship between your groups and covariates, and they can yield different results because of these assumptions. DOSS: In this model, the assumption is that all groups are affected by the covariates in the same way, i.e., they have the same slope but may have different intercepts. Essentially, it assumes that the effect of your covariates (age, depression, anxiety) is constant across all groups. Any variability among groups will be captured in the intercept term. DODS: This model allows each group to have a different relationship with the covariates. It assumes that both the slopes and the intercepts can differ across groups. In this case, not only can the groups differ in terms of their mean values (offset), but they can also differ in how they relate to the covariates (slope). If you use DOSS but the truth is more akin to DODS, you might underestimate the variability in the relationships between groups and covariates, which could potentially lead to the kinds of inconsistencies you've observed. Conversely, choosing DODS when DOSS would be sufficient could introduce unnecessary complexity and consume degrees of freedom, which might affect the statistical power. Given your focus on nuances like controlling for age, depression, and anxiety, it would be worth scrutinizing how these covariates interact with your groups. If you suspect different groups could indeed have a different relationship with these covariates, DODS might be more appropriate. Finally, The results you are observing appear to be false positives, likely due to the use of a lenient statistical significance threshold in your surface-based analysis. Opting for a more rigorous threshold, such as a cache=4, is generally a solid choice, particularly when you don't have a specific hypothesis about the location of significance of correlation and are conducting exploratory analyses
On Fri, Sep 29, 2023 at 3:38 PM Guthrie, Andrew J. < AJGUTHRIE@mgh.harvard.edu> wrote:
External Email - Use CautionHi all,
I took the freesurfer course in the recent past, and I am now beginning to run some glm thickness analyses!
Currently I am doing a thickness analysis looking at the difference between two groups while controlling for age, and two continuous measures of depression and anxiety. The general commands I run are as follows:
mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi lh --o lh.[filename].thickness.10.mgh
mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi rh --o rh.[filename].thickness.10.mgh
mri_glmfit --y lh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage lh --glmdir lh_[filename]_fwhm10.glmdir
mri_glmfit --y rh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage rh --glmdir rh_[filename]_fwhm10.glmdir
mri_glmfit-sim --glmdir lh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces
mri_glmfit-sim --glmdir rh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces
I noticed that in some analyses there would be no significant clusters when the threshold is set at 2.0 (p<0.01) but if you were to run the same analyses with a threshold at 2.3 (p<0.005) there would be a significant cluster, in the lingual for example. If I go one step further and set the threshold to 3.0 there is also a significant cluster – however the cluster is now in the isthmuscingulate.
Am I doing something incorrectly or is this possible? And if so, could you help me understand what is going on here.
Best,
Andrew
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://secure-web.cisco.com/1jlhzuRUb3xSbK2UnH8Qi7aIgO0uvWW-YTFd1hq09_5PP2O... 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 Mass General Brigham Compliance HelpLine at https://secure-web.cisco.com/1JKgnGjVOUdDsy--X-0PgkrABcNEXPApMWZjV6-9BeozR9F... < https://secure-web.cisco.com/1JKgnGjVOUdDsy--X-0PgkrABcNEXPApMWZjV6-9BeozR9F... .
This is the kind of thing that happens with clusterwise corrections. It is hard to predict what happens. When you raise the threshold, you lose some of the activation in your real map (ie, clusters become smaller). But in the Null map, clusters also become smaller. So, at the lower threshold, the frequency of a cluster of a size in your real data may be high (so you have a high clusterwise p-value). When you raise the threshold, the real data has a new (smaller) cluster size, but in the Null map, the frequency of that clustersize may be low (and so a lower (ie, more sig) clusterwise p-value). It's confusing, I know.
On 9/29/2023 3:35 PM, Guthrie, Andrew J. wrote:
External Email - Use Caution
Hi all,
I took the freesurfer course in the recent past, and I am now beginning to run some glm thickness analyses!
Currently I am doing a thickness analysis looking at the difference between two groups while controlling for age, and two continuous measures of depression and anxiety. The general commands I run are as follows:
mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi lh --o lh.[filename].thickness.10.mgh
mris_preproc --fsgd [fsgd filename] --cache-in thickness.fwhm10.fsaverage --target fsaverage --hemi rh --o rh.[filename].thickness.10.mgh
mri_glmfit --y lh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage lh --glmdir lh_[filename]_fwhm10.glmdir
mri_glmfit --y rh.[filename].thickness.10.mgh --fsgd [fsgd filename] doss --C contrast.mtx --surf fsaverage rh --glmdir rh_[filename]_fwhm10.glmdir
mri_glmfit-sim --glmdir lh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces
mri_glmfit-sim --glmdir rh_[filename]_fwhm10.glmdir --cwpvalthresh .05 --cache 2.0 abs --2spaces
I noticed that in some analyses there would be no significant clusters when the threshold is set at 2.0 (p<0.01) but if you were to run the same analyses with a threshold at 2.3 (p<0.005) there would be a significant cluster, in the lingual for example. If I go one step further and set the threshold to 3.0 there is also a significant cluster – however the cluster is now in the isthmuscingulate.
Am I doing something incorrectly or is this possible? And if so, could you help me understand what is going on here.
Best,
Andrew
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 -
Guthrie, Andrew J. -
John Anderson