基于点云处理网络的三维颜面正中矢状面预测模型  

3D facial midsagittal plane prediction model with point cloud processing network

作　　者：刘真光 朱玉佳 王勇[3,4,5] 傅湘玲 赵一姣[3,4,5] 陈晋鹏 LIU Zhenguang;ZHU Yujia;WANG Yong;FU Xiangling;ZHAO Yijiao;CHEN Jinpeng(School of Computer Science(National Pilot Software Engineering School),BUPT,Beijing 100876,China;Key Laboratory of Trustworthy Distributed Computing and Service(BUPT),Ministry of Education,Beijing 100876,China;Peking University School and Hospital of Stomatology,Beijing 100081,China;National Center for Stomatology,Beijing 100081,China;National Engineering Research Center of Oral Biomaterials and Digital Medical Devices,Beijing 100081,China)

机构地区：[1]北京邮电大学计算机学院(国家示范性软件学院),北京100876 [2]可信分布式计算与服务教育部重点实验室(北京邮电大学),北京100876 [3]北京大学口腔医学院口腔医院,北京100081 [4]国家口腔医学中心,北京100081 [5]口腔生物材料和数字诊疗装备国家工程研究中心,北京100081

出　　处：《山东大学学报（工学版）》2024年第3期30-35,共6页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(82071171,82271039);北京大学口腔医院实验室开放课题资助项目(PKUSS20220301);北京邮电大学研究生创新创业资助项目(2023-YC-T030)。

摘　　要：设计一种基于点云处理网络的三维颜面正中矢状面预测模型(facial midsagittal plane prediction network,FSPNet),实现三维颜面正中矢状面端到端自动化预测。FSPNet模型以三维颜面点云数据为输入,利用点云处理网络提高数据处理效率。它包含3个模块:全局特征编码模块从点云整体结构提取全局特征;局部特征编码模块从点云局部空间结构提取局部特征;正中矢状面预测模块聚合全局特征和局部特征,输出正中矢状面平面参数。借助点云编码模块,模型能够从不同角度充分挖掘颜面点云数据空间信息,实现点云特征全面提取。在真实颜面数据集上的试验结果表明,FSPNet模型具有优秀的性能,点云编码模块能够准确提取颜面点云特征,模型预测效果明显优于临床广泛使用的迭代最近点关联法,充分验证了FSPNet模型的有效性。A 3D facial midsagittal plane prediction network(FSPNet)based on point cloud processing network was designed.FSPNet enabled end-to-end automation for predicting the midsagittal plane of 3D facial structures.It took 3D facial point cloud data as input and utilized point cloud processing network to enhance data processing efficiency.The model consisted of three modules:the global feature encoding module extracted global features from the overall structure of the point cloud,the local feature encoding module extracted local features from the local spatial structure of the point cloud,the midsagittal plane prediction module aggregated both global and local features to output the parameters of the midsagittal plane.Through the point cloud encoding module,FSPNet effectively explored the spatial information of facial point cloud data from different perspectives,achieving comprehensive extraction of point cloud features.Experimental results on a real facial dataset showed the excellent performance of FSPNet.The point cloud encoding modules accurately extracted facial point cloud features,and the model achieved significantly better predictive performance compared to the widely used iterative closest point method in clinical practice.These findings provided strong validation for the effectiveness of FSPNet.

关 键 词：颜面正中矢状面 点云处理网络 平面预测 端到端框架 三维颜面数据 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]