Yea, the real hard part is figuring out how to model the twins that are split between groups. You might be able to create a bunch of groups. If you've already created a design in FSL that you like, you can extract the design matrix and use it in mri_glmfit with the --X option. Not sure about correcting for multiple comparisons though



On 3/8/19 3:13 AM, Julian Macoveanu wrote:

        External Email - Use Caution        

Thank you for the reply.

The original idea was to have an ANOVA with a group factor “diagnosis” with three levels. We could use an F-test to check for an effect of diagnosis and then perform 3 pairwise group tests (LR vs HR, LR vs ILL, HR vs ILL).

 

Would it help to only implement the two-sample group tests? However, the critical issue to account for correlated data remains since about 80% of the subjects have their co-twin in the same group. The idea behind the model that has a variable with 1s for each twin pair is to model out each pair’s mean. I implemented the model I described in the FSL GLM and estimated with FEAT for the functional data of the group. The second option I had was to perform a permutation based estimation as implemented in PALM and restrict the permutations to between pairs and single subjects.

 

Would any of these options work in FS? Maybe clustering the subjects somehow?

 

Julian