摘要
传统诊断方法难以实现儿童龋病的早期发现,深度学习技术主要应用在口腔X光片诊断,缺少可参考的数据集及诊断方法,导致口腔龋齿预防技术进展缓慢;因此提出了一种基于牙齿颌面轮廓与窝沟的儿童龋齿预防算法的医学标准,开发了一套判断牙齿早期是否患龋的算法;采用U-Net网络和注意力机制实现牙齿颌面识别分类,分别对颌面轮廓和窝沟形态的标准进行分类训练并对比效果,进而对两个模型进行加权融合,在α=0.5,γ=0.5时,模型融合效果最佳,AUC(Area Under Curve)达到0.7792、准确率(ACC)达到0.9026、F1-score达到0.9061;实验结果表明:融合模型效果高于单独使用基于颌面轮廓的模型和单独使用基于窝沟的模型,为儿童预防龋齿提供了一种新的解决方案。
Conventional diagnostic techniques face challenges in the early detection of pediatric dental caries.The application of deep learning,predominantly in oral X-ray diagnostics,is hindered by a dearth of relevant datasets and diagnostic methodologies,impeding ad-vancements in dental caries prevention.Hence,this paper introduces a medical standard for a pediatric dental caries prevention algorithm is introduced,focusing on the dental occlusal surface contours and fissure patterns.An algorithm capable of early dental caries detection is This research developed.The U-Net network and an attention mechanism are utilized,the algorithm efficiently identifies and classifies dental occlusal surfaces.Specialized classification training conducted on occlusal surface contours and fissure patterns,followed by a com-parative analysis of the efficacy.These two models are the integrated by using a weighted approach,achieving optimal integration atα=0.5 andγ=0.5,with an area under curve(AUC)of 0.7792,an accuracy(ACC)of 0.9026,and an F1-score of 0.9061.Experimental findings reveal that the integrated model surpasses those based solely on occlusal surface contours or fissure patterns,thereby providing an innova-tive approach to pediatric dental caries prevention.
作者
房禹池
杨世波
施麦克
李宏
FANG Yuchi;YANG Shibo;SHI Maike;LI Hong(Automation School,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China;Automation School,Zhejiang University of Water Resources and Electric Power,Hangzhou Zhejiang 310018,China)
出处
《传感技术学报》
北大核心
2025年第5期817-825,共9页
Chinese Journal of Sensors and Actuators
关键词
龋齿
牙齿颌面图像
U-Net网络
注意力机制
dental caries
occlusal dental imaging
U-Net architecture
attention mechanism
