三维神经网络识别CBCT中上颌磨牙MB2的实验研究  

Identification of maxillary molar MB2 in CBCT using three-dimensional neural network

作　　者：陈盼 申博文 周耕宇 胡靖宇 马净植 CHEN Pan;SHEN Bo-wen;ZHOU Geng-yu;HU Jing-yu;MA Jing-zhi(Center of Stomatology,Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology,Hubei Wuhan 430030,China;School of Stomatology,Tongji Medical College,Huazhong University of Science and Technology,Hubei Wuhan 430030,China;School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Hubei Wuhan 430074,China)

机构地区：[1]华中科技大学同济医学院附属同济医院口腔医学中心,湖北武汉430030 [2]华中科技大学同济医学院口腔医学院,湖北武汉430030 [3]华中科技大学人工智能与自动化学院,湖北武汉430074

出　　处：《临床口腔医学杂志》2025年第3期131-134,共4页Journal of Clinical Stomatology

基　　金：国家自然科学基金面上项目(No.62171193);湖北省重点研发项目(No.2022BCA033)。

摘　　要：目的:比较两种三维神经网络在锥形束CT(cone beam computed tomography,CBCT)中识别上颌磨牙近颊第二根管(second mesiobuccal canals,MB2)的能力。方法:纳入40名患者的CBCT资料,筛选并裁剪出符合实验要求的上颌磨牙近中颊根的CBCT影像,分别训练ViT(Vision Transformer)、DenseNet121两种神经网络,分析比较两者对上颌磨牙MB2的识别能力。结果:测试集中DenseNet121识别MB2的总准确率、灵敏度、精确度、F1评分均高于ViT,分别达到0.9167、1.0、0.9、0.9474;而ViT的受试者工作特征曲线下面积(area under curve,AUC)稍高于DenseNet121达到0.86。结论:三维神经网络在CBCT上颌磨牙MB2的检测中展现出较高的准确率,采用的两种网络中DenseNet121整体表现更佳。Objective:To compare the ability of two three dimentional neural networks in identifying the second mesiobuccal canals(MB2)of the maxillary molars in cone beam computed tomography(CBCT).Methods:Totally 40 patients'CBCT were included in this study,from which the relevant CBCT images of the maxillary molars'mesiobuccal roots were cropped after selecting with experimental requirements.The Vision Transformer(ViT)and DenseNet121 neural networks were trained separately to analyze and compare their abilities to identify MB2 in maxillary molars.Results:In the test set,DenseNet121 demonstrated superior performance compared to ViT,achieving accuracy,sensitivity,precision,and F1 score of 0.9167,1.0,0.9,and 0.9474,respectively.However,ViT had a slightly higher area under the receiver operating characteristic curve(AUC)at 0.86 compared to DenseNet121.Conclusion:Three-dimensional neural networks exhibit high accuracy of detecting MB2 in CBCT images of maxillary molars,with DenseNet121 showing better performance among the two networks.

关 键 词：神经网络 锥形束CT 上颌磨牙 近颊第二根管 

分 类 号：R816.98[医药卫生—放射医学]