I have been made aware that the outcome was different when samples of patients and controls were processed jointly or separately using FreeSurfers’ recon-all and the specialized subfield segmentation algorithms for t1w data. It is stated that FreeSurfer processess all subjects individually and hence then subsequent steps of normalizations etc if influenced by the pool which it is in. I am trying to find literature upon the question of joint or separate processing. Can anyone contribute?

I have not personally tested/compared joint and separate processing, however, it does make sense on a theoretical level to me and I would love to find assessments or statements regarding this.