ABSTRACT
Twomain types of algorithm on automatic recognition have been identified in literature: template matching algorithms (Rueda et al. 2006) and self-sufficient algorithms (e.g. McInerney & Terzopoulos 2000, Lloréns et al. 2009, Li&Yezzi 2007, Li et al. 2009,Mohan et al. 2010, Benmansour & Cohen 2010). An interesting approach regarding the second category is proposed by Yau et al. (2008). The authors presented an immediate and user-friendly methodology to segment the mandibular nerve canal. Their algorithm is based on the common Dentascan multi-view context (axial, panoramic, and cross-sectional views), and consists of automatically analysing, through the successive applications of a statistical segmentation and a region growing filter, each cross-sectional slice along the panoramic curve (to which the cross-sectional planes are orthogonal) between two user-defined extremities. In particular, each new slice region of interest (ROI) is achieved intersecting the current slice ROI with an expected nerve region in the following slice. However, conduct irregularities such as holes, bifurcations, and significant canal interruptions, typical of the mandibular canal, strongly affect the proposed algorithm, leading to possible leakages. Additionally, Yau et al. (2008) claim that this problem can be solved by using the ROI itself as the boundary area, however, this does not actually help to a great extent: even when the ROI is used as a boundary constraint, it would allow the new ROI centre to be in the same slice related position as the previous one at most, which, consequently, would then fall outside the canal, losing its trackwhile scrolling forward as a result. Consequently, inspired by the algorithmproposed byYau et al. (2008), it was decided to work on a new algorithm able to manage the mandibular canal irregularities, while taking advantage of a previously developed functionality that allows run-time custom reconstruction of oblique planes (Chiarelli et al. 2010a).
