Understanding intercepts in the longitudinal DBM

[nimzodisaster]

I am conducting a longitudinal DBM analysis using your approach.
In the manuscript at bioRxiv (which I think is yours), the authors wrote, "Voxel-wise statistical modelling was performed using R/3.4.4 and RMINC/1.5.2.2 using mixed-effect linear models, with a fixed effect of (pseudo) timepoint and a random intercept by subject.", and I have a question I'm trying to wrap my head around.

In a normal DBM pipeline, the template space sets the reference point for the Jacobians since all scans are registered to that same image. That is not the case in the longitudinal pipeline, the Jacobians of different subject reference different subject-specific templates. While the mixed model you specified would be accurate because it would (more) accurately index within subject changes in the Jacobian, different subjects would eseentially have different intercepts, I think? So, if we ran a mixed model like Jacobian ~ group + timepoint + group:timepoint + (1|subject), the fixed effect differences of group are kind of meaningless, and the the interaction as well?

Am I wrong in this line of thinking?

[gdevenyi]

Hi, thanks for using twolevel_ants_dbm, you may be interested in https://github.com/CoBrALab/optimized_antsMultivariateTemplateConstruction which is a complete rewrite of all steps of template and DBM to address the many design shortcomings of wrapping the ants template builder. This answer applies equally to both,

So, if we ran a mixed model like Jacobian ~ group + timepoint + group:timepoint + (1|subject), the fixed effect differences of group are kind of meaningless, and the the interaction as well?

You're close.

1. You're correct that the fixed effects are meaningless since the measurements are all measured relative to the subject-specific template.

2. You're incorrect about the interaction:
   An interaction represents the additional slope of timepoint for levels of a fixed predictor (group) vs the reference group. As such you can think about it is a representing a different change over time versus the reference group, i.e. this is probably the most interesting term of a longitudinal model like this.