Hello FSL experts,
I've been going through an analysis using FIRST for the first time, and have had a crack at doing a vertex analysis of the thalamus. I've got some previous experience with the FSL GLM from TBSS analyses and the like, but there are parts of the FIRST user guide which indicate that common model set ups for things like TBSS might not be appropriate for FIRST:
"However, the design matrices are treated slightly differently by FIRST compared with FEAT. The main difference is that FIRST does not model separate group variances and so ignores the "Group" column in the Glm gui. Another difference is that only F-tests are needed if you just want to look for any change in vertex position. Therefore it is only necessary to set up an F-contrast, although this needs T-contrasts to be set up (since that is how F-contrasts are specified). Individual T-contrasts can be used to look for signed differences (separating local "expansion" from "contraction" of the structure."
I'm not clear from this paragraph however if these F tests are being highlighted for FIRST just because they can be more simple to put together than a t-test, or if a quirk of the vertex analysis means they're actually more statistically valid. I believe for example that if the "group" variable is set to all 1's (as it can be in common groupwise analyses) then group variances are ignored anyway?
I'm wanting to a typical patient vs. control groupwise analysis of the thalamus, once controlling for age and once without. From previous experience, in the GLM this would mean setting "Group" to all 1's, and then having an EV for each group with 1's indicating membership and 0's for non-membership. Contrasts would then be "1 -1" or a "-1 1" depending on the direction of effect it's examining. Adding in age would mean another "age" EV (with demeaned values), with this then being set to "0" in the contrasts just described.
My question is: is this inappropriate for a FIRST analysis, and should I be using an F test instead?
Many thanks in advance of any help,
All the best,
Iain
|