 |
 |
Hello all, I am trying to conduct a longitudinal volumetric analysis of three time points, each in a different condition, 1) baseline, 2) treatment A, 3) treatment B (A and B are counterbalanced). I am mainly interested in the differences: Treatment A - Baseline; Treatment B - Baseline; Treatment A - Treatment B. In my first pass at the data, I did a longitudinal registration of all three scans (after skull stripping) for each individual. I used '0' as the age for each condition, as I am not interested in the weighted differences based on treatment duration. I then segmented the subject averages (in this new average space) and used the rc1's and rc2's in the dartel pipeline to create a template and flow fields for the dataset. In the end I combined the jacobians from the longitudinal registration (3 per subject) and the dartel flow fields (1 per subject) to create a value for each timepoint for each subject (again, 3 per subject). I then used these in a repeated measures analysis with total brain volume (a scalar value) as a covariate. I was curious to see what your thoughts were on the above procedures. I am trying to compare my methods to those in Eshaghi et al., 2014 (NeuroImage). They add in a step where they take the average subject template, and register each condition to it again to produce their within-subject jacobians (in a pair-wise fashion). This, however, seem to be redundant step, as I had assumed the original longitudinal registration would produce those same jacobians for each image used in the longitudinal registration procedure. How comparable are these two versions of calculating the jacobians? This same paper also multiplied these 'pair-wise' jacobians by the subject's gray matter segmentation of the average image to create a 'pseudo-time-point' which they then flowed to their group specific template. How does this method compare to creating the subject flow fields independently and then combining them to the initial longitudinal jacobians? Best, Kathy