Margin Line Deformation And Segmentation Prediction Based On Deep Learning

In the process of prosthodontics,the margin line of teeth plays a vital role in the tightness between the user’s gums and teeth,which determines the success of prosthodontics.This paper mainly aims at the problems that the artificial method will introduce subjective errors in the dental restoration deformation design of the margin line,and the current extraction method of the dental margin line requires complicated manual interaction,realizing the deformation estimation and automatic extraction of the dental margin line.Aiming at the problems of high network computational complexity in the surface deformation prediction method,the feature processing steps in the curve deformation prediction method ignore the correlation of non-local information,and ignore the geometric consistency relationship between the position and the normal direction,this paper proposes a margin line deformation prediction network which based on graph convolution and combined with geometric consistency.This method builds a graph convolution layer that can model local and non-local information at the same time,and designs a geometrically assisted prediction module which uses geometric constraints between position and normal.First,the graph convolution layer which combines the k-nearest neighbor algorithm and the self-attention mechanism is used to mine the model’s local and non-local information at the same time to predict the position and normal information of the points;then,input the position of the predicted point into the geometric auxiliary prediction module,using the geometric consistency constraint to calculate the geometric normal information;finally,the obtained geometric normal information is fused with the normal information predicted by the graph convolutional network,and the graph convolutional network is used again for deformation prediction to obtain the final deformation prediction result.We verify the effectiveness of our method on the self-built margin line database.Experiments show that,compared with the mainstream algorithms,our method can improve margin line prediction accuracy of deformation results effectively,and the mean square error measurement is reduced by 16.7%.In addition,aiming at the problem of manual extraction of the margin line in the current dental restoration work,inspired by the recent progress of point cloud semantic segmentation,we propose a margin line automatic extraction method based on point cloud segmentation.The method uses a point cloud segmentation network to predict the labels of the teeth and gums,and uses a data-driven approach to automatically extract the margin line.On this basis,a conditional random field was introduced to refine the segmentation results,and the correlation between points was used to transmit spatial information of the segmentation results,so as to avoid rough zigzag distribution in the boundary area of the segmentation results.The experimental results on the established dental database show that this method can automatically extract the margin line by using the deep learning network,and can ensure the extraction accuracy of the margin line.The average absolute error of the extraction result of the neck edge line is 0.016.