Extraction of Tooth Preparation Line Based on Sparse Octree Convolutional Neural Network

The traditional tooth preparation margin line extraction requires marking feature points interactively,which is complicated and inefficient.Aiming at the problem,a method is proposed to extract the preparation line by convolutional neural network.Firstly,in order to construct a dental preparation dataset,preparation models are processed as the sparse octree point cloud with labels through the spatial division of the octree.Secondly,the trained network model is used to divide the preparation point cloud into two parts.Thirdly,the dense conditional random field is used to optimize boundary points.Then,the new boundary points can be achieved by the fitting and interpolation.Finally,the corresponding projected points of the boundary points are connected to form a preparation line.Experiments on dental preparation datasets show that the accuracy of label predicted by network model can reach 97.23%,and the validity of the method is verified by comparing the curve deviation between the automatic constructed preparation line and the manual extracted preparation line.