Thanks, Franziska. It seems to be a problem with partial voluming. Cerebral white matter (white pixels) mixes with a thin layer of cerebrospinal fluid (black pixels) and the result is a bunch of gray pixels that look like the pulvinar. Partial voluming is one of the biggest problems with these Bayesian segmentation methods, the price you pay for the (huge!) advantage of being agnostic to / robust against changes in MR contrast of the input scan... Unfortunately, there is no easy fix for this at the moment. Normally, the increment in volume due to the leak is not as large as it may seem because segmentations are weighted by their probabilities when computing volumes. In the future, we plan to tackle this with deep learning methods that do a fantastic job with partial voluming. The problems with this approach are: 1. Robustness against contrast changes, and (mostly) 2. Training with ex vivo data is tricky. Cheers, /Eugenio

Juan Eugenio Iglesias Senior research fellow CMIC (UCL), MGH (HMS) and CSAIL (MIT) http://www.jeiglesias.com

On 3/16/21, 03:32, "freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Gronow, Franziska" <freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Franziska.Gronow@uniklinikum-dresden.de> wrote:

External Email - Use Caution

Dear Eugenio,

here are the zipped norm.mgz and automated segmentations for two subjects: https://secure-web.cisco.com/1NFOUrkRWZ6w5iEt9PtCxQNkjFZr9fIybWKJt1Rx3T_PB7D...

Hope this helps!

Best regards, Franziska

_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer