基于深度边界感知的交互式牙齿网格分割方法  

Interactive Tooth Segmentation from Dental Mesh Based on Deep Boundary Sensing

作　　者：邹峥 吴连杰 刘石坚 ZOU Zheng;WU Lianjie;LIU Shijian(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China;Fujian Provincial Key Laboratory of Big Data Mining and Applications,Fuzhou 350118,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福建福州350117 [2]福建省大数据挖掘与应用技术重点实验室,福建福州350118

出　　处：《福建师范大学学报（自然科学版）》2024年第6期30-39,共10页Journal of Fujian Normal University：Natural Science Edition

基　　金：国家自然科学基金资助项目(62172095);福建省自然科学基金资助项目(2022J01932);福建省教育厅科技项目(JAT210283、JAT220052)。

摘　　要：从三维牙颌网格中分割得到单牙数据,是齿科计算机辅助设计系统的重要环节。以网格中的顶点作为样本基本单元,结合友好的人机交互接口设计,提出一种深度边界感知网络(deep boundary sensing network,DBSNet)的方法,可灵活、高效地实现单颗牙齿的交互式分割。以PointNet++为基础,通过基于曲率的自适应采样和基于边界距离感知的监督来提升分割准确度。此外,还提出一种两点式交互接口。自动化方法虽可以完成任务,但在需要个性化分割特定牙齿时,交互式方法更为快速、灵活。为评估DBSNet的性能,通过实验将其与当下最优方法进行对比。测试结果表明,DBSNet能获得边界区域总体准确率92.15%、边界区域平均交并比83.11%的结果,各项指标均优于对比方法。上述结果以及消融实验结果验证了所提方法的有效性。Tooth segmentation from three-dimensional dental mesh is the basis of a dental computer-aided design system.Mesh vertices are selected as the primary data samples,and combined with the design of a friendly human-computer interface,this paper presents a method named Deep Boundary Sensing Network(DBSNet)to flexibly and efficiently realize the interactive segmentation of single teeth.Based on PointNet++,the proposed method improves segmentation accuracy through adaptive sampling based on curvature and supervision based on boundary distance perception.In addition,a two-point interactive interface is presented.Although automated methods can complete tasks,the interactive method is faster and more flexible when it comes to the need for personalized segmentation of specific teeth.To evaluate the performance of the proposed method,extensive comparisons are carried out between DBSNet and SOTAs.The test results show that DBSNet achieves a boundary overall accuracy of 92.15%and a boundary mean Intersection over Union of 83.11%,which are the best results among those compared.The above results and the results from ablative studies demonstrate the effectiveness of the proposed method.

关 键 词：牙齿 三角网格 分割 人机交互 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]