3D tooth segmentation and classification method based on deep learning

Abstract

The invention discloses a 3D tooth segmentation and classification method based on deep learning. A model framework used in the method is divided into four stages: stage 1: performing sample data acquisition and manufacturing; stage 2, building and training a deep network model based on two-segmentation between teeth and gingiva; stage 3, building and training a deep network model based on sixteen types of segmentation between teeth; and step 4, integrating results of the step 2 and the step 3. The invention is based on a deep convolutional network, and includes performing tooth and gingiva segmentation through a binary classification task, so that the purpose of separating target teeth is achieved; performing sixteen types of tooth interior segmentation tasks on the teeth; due to the difference of the number of patches between single teeth, performing deep learning model training by using different loss weights; performing post-processing by combining tooth and gingiva results and applying a graph cut algorithm; and finally, upsampling original 3D tooth data by using a support vector machine algorithm so as to achieve the purpose of accurately marking actual 3D teeth.

Description

Technical field [0001] The present invention belongs to the field of 3D tooth marks, and more particularly to a 3D dental segmentation and classification method based on deep learning. The present invention is based on the depth convolutionary network, and the results are processed after the graphic algorithm. Background technique [0002] In the field of computer-assisted orthodontic treatment, accurate tooth tags on three-dimensional tooth is an important task, which is premise for dental analysis and rearrangement. [0003] Existing automation or semi-activated methods typically require manual interaction, which is time consuming. In addition, they usually use simple geometric properties as partitioning standards, which does not greatly handle changes in dental appearance of different patients. Recently, in the field of computer vision and computer graphics, several pioneering depth neural networks (such as POINTNET and MESHSNET) have been proposed, and the 3D shape is effecti...

AI Technical Summary

Problems solved by technology

However, there are still several problems in this type of method: 1. The former network ignores the local geometric context, and effective local structure modeling has been proven to be the key to the success of deep neural networks in fine-grained segmentation tasks; 2. Although the latter network considers using grid units instead of points as input to increase local structure information, it only has 14 categories of teeth, and it has the same weight of loss items for each category of teeth during the training process. , and finally lead to not particularly ideal performance in the actual segmentation task process

Images