Detecção automática do canal mandibular em tomografias computadorizadas de feixe cônico

Abstract

The mandibular canal is a bony conduit located inside the mandible that carries the inferior alveolar neurovascular bundle. Several methods available in the literature implement channel segmentation, these being manual methods or based on deep learning, which are time consuming or demand a large amount of data, respectively. In this work, we present a semiautomatic computational method for segmentation of the mandibular canal in CT scans based on signal processing. The method begins with an axial cut located 10 mm above the base of the mandible, where the mandible is well delimited, making it possible to define an arc that describes it. Parasagittal slices are obtained, in which the mandibular canal can be detected as a circular structure using the Hough transformation. As the channel is not the only circular structure in such planes, the user selects the structure referring to the channel, and from this seed, the three-dimensional structure of the channel is defined, using the region growth algorithm in each of the oblique slices. In this study, the proposed method was applied to 25 Cone Beam Computed Tomography (CBCT) scans of different patients (21 women and 4 men aged between 11 and 75 years). The resulting volumes were visually evaluated by specialists, who noted how appropriate the volume obtained was. In addition, a statistical evaluation also pointed out the proximity of the volumes obtained by the methods and those defined by specialists (average IoU: 0.79). The specialists' visual evaluation and the statistical evaluation demonstrated the accuracy of the results achieved by the method. The segmentation presented in this work consists of a robust solution, in the sense that the volumes obtained are statistically close to the real volumes, in addition to innovating by overcoming the main challenges of previous methods: time, database size and operator experience.