Hi Yang

mri_surf2surf will use the spherical registration to propagate the labels from one subject to another. mris_ca_label on the other hand will do a fully Bayesian segmentation using anisotropic non-stationary Markov Random Fields. In the end they probably aren’t hugely different as the folds are the dominant feature in each, but I would expect ca_label to be a bit better

cheers

Bruce

From: Alex <free_learner@foxmail.com>
Sent: Tuesday, July 14, 2026 10:00 PM
To: freesurfer <freesurfer@nmr.mgh.harvard.edu>
Subject: [Freesurfer] Differences between mri_surf2surf and mris_ca_label for cortical parcellation

        External Email - Use Caution        

Dear experts,

As far as I know, there are two common ways to assign anatomical labels to cortical vertices: using mri_surf2surf (combined with an atlas annot file) and using mris_ca_label (which relies on a pre-trained classifier). My question is: in practice, which method is preferable?

Is mris_ca_label theoretically more accurate, but in reality the two yield nearly identical results? Or are there systematic, non-negligible differences? Also, is mris_ca_label less robust than the simpler projection method, particularly in subjects with imperfect surface reconstruction.

Thank you for your advice.

Yang Hu