On 2/21/2020 4:01 AM, Ferraro, Pilar wrote:
External Email - Use Caution
Dear Freesurfer experts,
I’ve just completed a GLM analysis in Freesurfer and I’d like to be sure that all the steps I’ve performed are the right ones.
My analysis is looking at age related changes in cortical thickness in one group of patients, after regressing out the effects of disease duration. I obviously assume an inverse relationship between age and cortical thickness so that older age would be associated with cortical thickness reductions in specific brain regions.
To test this hypothesis with Freesurfer I’ve used a GLM DOSS design. I’ve generated a fsgd file with the first column being the list of patients, the second column the age of patients, and the third column the disease duration. Then I’ve run the mris_preproc command.
Afterward, I’ve generated 2 contrast matrices (one for the left and one for the right hemisphere) with the following structure: 0 -1 0.
1 question. Is the contrast matrices structure I’ve chosen the most appropriate one for the hypothesis I’d like to test?
Yes, but if you are going to use -1, you need to keep track of that when you specify the sign (but you are using abs so it does not matter). Also, you don't need different fsgd and contrast files for each hemisphere.
Then, I’ve run the glm analysis using the mri_glmfit command.
To visualize the results I’ve used the freeview -f command and set the overlay_threshold to 1,5 instead of 4, cause I was interested in looking at all results significant at p < 0.05.
2 question. Is the overlay_threshold I’ve used the correct one?
for visualization, you can use whatever threshold you want. But if you are going to use monte carlo simulations, then you need p<.001. If you want to keep p<.05, then you have to do permutation.
Lastly, I was interested in looking at my results after Montecarlo correction, so I’ve run the simulation using the mri_glmfit-sim command. However, here I’ve changed few options:
mri_glmfit-sim \ --glmdir lh.dur.glmdir \ --cache 4 neg \ (here I’ve changed to 1,5 since I was interested again at a P value < 0,05 and not < 0.0001 and I’ve changed the neg option with abs since I was interested in absolute values identified through the previously run analysis) --cwp 0.05\ --2spaces
3 question. Are the edits I made to the command correct? It is not perfectly clear to me which one should be the preferred sign for the analysis (neg, pos or abs?).
abs is an unsigned test. If you have an aprior hypothesis, then you can set the sign to that. Eg, if you are expecting the age slope to be negative, then you could choose negative (or in your case positive since you flipped the sign in the contrast)
In the end, in order to visualize the Montecarlo corrected results I’ve used again the Freeview -f command and again I’ve sent the overlay threshold to 1,5 since I was interested in all significant clusters at P < 0,05.
4 question. Again, this threshold is the correct one for my purpose?
It will show all the clusters that are at p<.05 corrected.
I have one last question concerning the Montecarlo simulation. Are there any cases in which you would suggest to do not use it since it would not be an appropriate correction?
I'm recommending that everyone use permutation at this point. MC can generate inlfated false positives
Sorry for the long email but I needed to report all the steps in order to get an answer. Many thanks for all the help you’ll be able to provide.
Best,
Pilar Ferraro
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer