FreeSurfer reconstructs the cortical surface from a T1-weighted scan and derives the
structural measures most anatomical studies report: cortical thickness, surface area,
curvature, subcortical volumes, and a parcellation into named regions. Its single command,
recon-all, runs an hours-long sequence of skull-stripping, intensity normalization,
tissue segmentation, and surface tessellation.
Two practical points matter for anyone here. First, you may already be running it:
fMRIPrep calls FreeSurfer internally for surface reconstruction and
registration, so a fMRIPrep derivatives directory usually contains a full FreeSurfer
output tree you can use without re-running anything. Second, recon-all takes 6–12 hours
per subject, which makes it the step that dominates wall-clock time on a cohort and the
one that most needs job array scheduling rather than a
loop.
Its outputs are also the input to a great deal else — surface-based statistics, normative models of cortical thickness, and most structural biomarkers in the aging and dementia literature.