Methods: After the manual choice of a (volume of interest) VOI, which is supposed to be inside the investigated root canal. As a first step, the algorithm performs a quick fuzzy c-means clustering based separation of light and dark areas within the selected VOI. This is followed by a three dimensional region growing starting from the manually selected initial point. Thus we obtain the approximate volume of the root canal bounded by light regions in the CBCT records. An adaptive stopping criterion was developed to avoid the inclusion of voxels that are situated outside the root canal. When the set of inner voxels is identified, the corrected marching cube algorithm is employed to extract the triangle mesh surface of the root canal. This is computed using the original intensity values, and subpixel coordinates were obtained. The medial line of the root canal is approximated using the 3D curve skeleton of the identified root canal. The curve skeleton is extracted using the mesh contraction algorithm. The extracted curve skeleton is a collection of points having floating-point coordinates, which can be efficiently approximated as spline curves.
Results: After a manual selection of the VOI, the proposed algorithm can automatically produce an accurate center line in more than 98% of cases with simple root canal, and in 95% of bifurcated root canals. The whole identification process of one root canal took less than 3 seconds.
Conclusions: We have proposed and implemented a complex image processing procedure for the detection of the medial line of root canals from CBCT records. The proposed algorithm proved accurate and efficient.