基于残差网络的口腔全景片分类模型设计与实验  

Design and Experiment of Classification Model for Panoramic Dental Radiographs Based on Improved ResNet-50

作　　者：朱海婷 沈雨晨 李佳岭[3] ZHU Haiting;SHEN Yuchen;LI Jialing(School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Key Laboratory of Computer Network and Information Integration,Ministry of Education,Southeast University,Nanjing 210096,China;Department of Orthodontics,Nanjing Stomatological Hospital Affiliated to Medical School of Nanjing University,Nanjing 210008,China)

机构地区：[1]南京邮电大学物联网学院,南京210003 [2]东南大学计算机网络和信息集成教育部重点实验室,南京210096 [3]南京大学医学院附属口腔医院正畸科,南京210008

出　　处：《实验室研究与探索》2025年第4期103-107,155,共6页Research and Exploration In Laboratory

基　　金：计算机网络和信息集成教育部重点实验室开放课题(K93-9-2023-04);国家卫健委科研项目(A2023-021)。

摘　　要：口腔全景片分类技术能辅助医生快速、准确诊断牙齿生长发育阶段,并支持制定和调整正畸治疗方案。针对残差网络(ResNet)在图像分类中存在的局限性,提出一种新的分类网络ID-ResNet,以提高口腔全景片的分类精度。ID-ResNet结合了双通道注意力机制DANet和Inception模块,旨在解决ResNet在捕捉特征间复杂关系、多尺度特征提取和特征图尺寸不一致等问题。实验结果表明,ID-ResNet在口腔全景片的分类任务上性能优于大多数现有主流模型,显著提升了分类准确性。With technological advancements,panoramic dental radiograph classification becomes increasingly important in orthodontic practice,helping doctors diagnose dental issues quickly and accurately while supporting treatment planning and adjustments.To address the limitations of residual network(ResNet)in image classification,a new network,called ID-ResNet,is designed to enhance classification accuracy.The ID-ResNet incorporates the dual channel attention mechanism(DANet)and the inception module to overcome ResNet shortcomings in capturing complex feature relationships,extracting multi-scale features,and handling inconsistent feature map sizes.Experimental results show that the ID-ResNet outperforms most mainstream models in panoramic radiograph classification,significantly boosting accuracy.

关 键 词：口腔全景片 卷积神经网络 多尺度特征 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]