Towards a Statistical Shape-Aware Deformable Contour Model for Cranial Nerve Identification

This paper presents a cranial nerve segmentation technique that combines a 3D deformable contour and a 3D contour Statistical Shape Model (SSM). A set of training data for the construction of the 3D contour shape model is produced using a 1-simplex based discrete deformable contour model where the centerline identification proceeds by optimizing internal and external forces. Point-correspondence for the training dataset is performed using an entropy-based energy minimization of particles on the centerline curve. The resulting average shape is used as a prior knowledge, which is incorporated into the 1-simplex as a reference shape model, making the approach stable against low resolution and image artifacts during segmentation using MRI data. Shape variability is shown using the first 3 modes of variation. The segmentation result is validated quantitatively, with ground truth provided by an expert.