I’d like to hear Your ideas how to properly normalize the data from longitudinal diffusion scans. Our concern is a bias in deformation/interpolation towards one of the 5 time points in the study. I’d appreciate any hints or already published solutions.
What we already did is a midpoint average from 2 time points according to this paper - https://doi.org/10.1002/hbm.21370. Then we can go on with normalization in eg. FSL TBSS. That’s ok for 2 but would be more tricky with 5 images…
We’ve got another 2 ideas in mind:
- Create a midpoint average from T1w images in Serial Longitudinal Processing in SPM or ANTs and then coregister DWI (FA?) to that template. Question is whether it is better to normalize averaged T1w (SPM?) or rather mean (coregistered and averaged) FAs (TBSS).
- Don’t bother at all and… “simply treat each timepoint as if it were an individual, so that the study-specific template and skeleton is unbiased towards any particular timepoint”
I look forward to hearing from You!