基于深度学习的点云补全网络修复上颌磨牙钙化根管的研究  

Deep Learning-based Point Cloud Completion Network for Restoration of Calcified Root Canals in Maxillary Molars

作　　者：温佳欢 傅裕杰 任根强 陈宇飞[2] 张旗[1] WEN Jiahuan;FU Yujie;REN Genqiang;CHEN Yufei;ZHANG Qi(Department of Endodontics,School and Hospital of Stomatology,Tongji University,Shanghai Engineering Research Center of Tooth Restoration and Regeneration,Shanghai 200072,China;College of Electronics and Information Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学口腔医学院·同济大学附属口腔医院牙体牙髓病科,上海牙组织修复与再生工程技术中心,上海200072 [2]同济大学电子与信息工程学院,上海201804

出　　处：《口腔医学研究》2023年第5期455-459,共5页Journal of Oral Science Research

基　　金：国家自然科学基金(编号:81870760、82170945);上海市卫生健康委员会卫生行业临床研究专项(编号:202040282);上海申康医院发展中心临床三年行动计划(编号:SHDC2020CR3058B)。

摘　　要：目的:通过模拟钙化根管样本,训练基于深度学习的点云补全网络修复补全上颌磨牙中上段钙化根管,为之后辅助临床治疗提供前期理论基础。方法:收集200颗上颌磨牙(牙髓腔完整,拥有3个根管)并获取其牙髓三维模型。使用Mimics软件编辑牙髓形态以模拟上颌磨牙中上段钙化。将模拟钙化牙髓与完整牙髓配对作为点云补全模型的训练样本。经以上操作,我们制作了一个包含200组样本的钙化牙髓样本数据集,其中180组用于模型训练,20组用于模型测试。在测试阶段,通过逆向工程软件将输出结果网格实体化,再与金标准(完整牙髓)进行形态的对比评估。结果:Dice相似性指数为(86.80±1.32)%、平均对称表面距离为(0.04±0.02)mm和Hausdorff距离为(0.28±0.05)mm。结论:预测结果与金标准相似性高,外形偏差较小。基于深度学习的点云补全网络可以有效修复上颌磨牙中上段钙化。为人工智能辅助中上段钙化根管治疗提供前期理论基础。Objective:To train a deep learning-based point cloud completion network by simulated calcified root canal samples for restoration of calcified canals in upper and middle thirds in maxillary molars.Methods:Two hundred maxillary molars(with complete pulp cavities and three root canals)were collected and scanned by Micro-CT to obtain 200 samples of 3D model of the pulp.The pulp morphology was edited using Mimics software to simulate the calcified root canals in upper and middle thirds of maxillary molars.The paired data of complete pulp and simulated calcified pulp were used as training samples for the point cloud completion network.A dataset containing 200 samples was produced,which was divided into 180 sets as the training set and the remaining 20 sets as test set.In the testing stage,the output results were grid processed through the reverse engineering software,and then compared with the ground truth for morphology evaluation.Results:The DSC between the prediction model and the ground truth was(86.80±1.32)%,the ASSD was(0.04±0.02)mm,and the HD was(0.28±0.05)mm.Conclusion:The predicted results are highly similar with the ground truth and the deviation of shape difference is small.The deep learning-based point cloud completion network can restore the calcified canals in upper and middle thirds in maxillary molars with good completion effect.It can provide a pre-theoretical basis for subsequent artificial intelligence-assisted calcified root canal treatment.

关 键 词：牙髓钙化 点云补全 深度学习 上颌磨牙 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] R783.3[医药卫生—口腔医学]