Coherence Filtering to Enhance the Mandibular Canal in Cone-Beam CT data

Segmenting the mandibular canal from cone beam CT data, is difficult due to low edge contrast and high image noise. We introduce 3D coherence filtering as a method to close the interrupted edges and denoise the structure of the mandibular canal. Coherence Filtering is an anisotropic non-linear tensor based diffusion algorithm for edge enhancing image filtering. We test different numerical schemes of the tensor diffusion equation, non-negative, standard discretization and also a rotation invariant scheme of Weickert [1]. Only the scheme of Weickert did not blur the high spherical images frequencies on the image diagonals of our test volume. Thus this scheme is chosen to enhance the small curved mandibular canal structure. The best choice of the diffusion equation parameters c1 and c2, depends on the image noise. Coherence filtering on the CBCT-scan works well, the noise in the mandibular canal is gone and the edges are connected. Because the algorithm is tensor based it cannot deal with edge joints or splits, thus is less fit for more complex image structures.