Evaluation of the potential of automatic segmentation of the mandibular canal using cone-beam computed tomography

Abstract

We aimed to investigate the effectiveness of software for automatically tracing the mandibular canal on data from cone-beam computed tomography (CT). After the data had been collected from one dentate and one edentate fresh cadaver head, both a trained Active Shape Model (ASM) and an Active Appearance Model (AAM) were used to automatically segment the canals from the mandibular to the mental foramen. Semiautomatic segmentation was also evaluated by providing the models with manual annotations of the foramina. To find out if the tracings were in accordance with the actual anatomy, we compared the position of the automatic mandibular canal segmentations, as displayed on cross-sectional cone-beam CT views, with histological sections of exactly the same region. The significance of differences between results were analysed with the help of Fisher's exact test and Pearson's correlation coefficient. When tracings based on AAM and ASM were used, differences between cone-beam CT and histological measurements varied up to 3.45mm and 4.44mm, respectively. Manual marking of the mandibular and mental foramina did not improve the results, and there were no significant differences (p=0.097) among the methods. The accuracy of automatic segmentation of the mandibular canal by the AAM and ASM methods is inadequate for use in clinical practice.