We have a paper in review using a pipeline consisting of template building via deformable registration, segmentation and back warping the segmentation onto the samples for about 500 mice heads (6 tooth per mouse, so 3000 molars to segment) for the genetic mapping of molar size development.
Compared to manual segmentation (gold standard) similarity is about 0.95 for M1 (using about 80 samples), about 0.9 for M2.
There is no reason for the same concept not to work on human data, if you can keep the age spread in your data under control (either building age specific template, or using a multi-template segmentation paradigm).
From: slicer-users [mailto:[hidden email]] On Behalf Of arumiat
Sent: Thursday, February 02, 2017 8:52 AM
To: SPL Slicer Users <[hidden email]>
Subject: [slicer-users] Dental CT teeth extraction