Mandibular canal segmentation using 3D Active Appearance Models and shape context registration

This paper presents a method for automatic segmentation of mandibular canal from CBCT (cone beam CT) images based on 3D Active Appearance Models (AAM) and shape context registration. The proposed algorithm consists of two stages: Firstly, Shape Context based non-rigid surface registration of the manual segmented images is used to obtain the point correspondence between the given training cases. Subsequently, an AAM is used to segment the mandibular canal on 60 training cases. The method is evaluated using a 5-fold cross validation over 5 repetitions. The mean Dice similarity coefficient and 95% Hausdorff distance are 0.86 and 0.90 mm, respectively.