基于卷积神经网络的牙周炎智能诊断  

Periodontitis intelligent diagnosis in convolutional neural networks

作　　者：程子健 黄鹏[1] 戚刚刚 张睿[1] 李晓军[1] Cheng Zi-jian;Huang Peng;Qi Gang-gang;Zhang Rui;Li Xiao-Jun(School of Stomatology,Zhejiang Provincial Clinical Research Center for Oral Diseases,Key Laboratory of Oral Biomedical Research of Zhejiang Province,Cancer Center,Zhejiang University,Hangzhou 310006,China)

机构地区：[1]浙江大学口腔医学院浙江省口腔疾病临床医学研究中心浙江省口腔生物医学研究重点实验室,癌症研究院,浙江杭州310006

出　　处：《兰州大学学报（医学版）》2022年第10期32-35,42,共5页Journal of Lanzhou University（Medical Sciences）

基　　金：中华口腔医学会青年托举人才培育资助项目(2020PYRC001)。

摘　　要：目的 基于卷积神经网络的深度学习技术,迭代升级传统U-net算法,分析判读曲面体层片牙槽骨吸收严重程度,探究在牙周炎诊断中的价值。方法 构建Ali-U-Net算法,分析公共数据集116例脱敏口腔患者的全景X片图像,计算每颗牙的牙槽骨吸收程度,比较U-net和Faster RCNN等传统算法。结果 以高年资牙周专科医师的判读结果为对照,AliU-Net网络曲线下面积为0.93,高于原生U-Net(0.90)和Faster RCNN(0.67)。结论 基于人工智能卷积神经网络成功迭代建立了Ali-U-Net算法,在曲面体层片牙槽骨吸收严重程度判读方面体现了较好的应用价值。将本算法与病历信息的智能读取相结合或可应用于牙周炎的分期及分级智能诊断。Objective Based on the convolutional neural networks, aimed to construct a new U-net algorithm to detect the alveolar bone loss in panoramic radiographs and assist the periodontitis diagnosis. Methods AliU-Net was established and applied to a public dataset containing 116 oral panoramic radiographs, assessed the images for alveolar bone loss. The results were also compared with those of U-net and faster RCNN.Results The interpretation results of well-trained periodontologist were used as control, Ali-U-Net showed a great ability in recognition of alveolar bone loss. The receiver-operating characteristic curve showed that the area under the curve(AUC) of Ali-U-Net was 0.93, which was greater than that of other methods(AUC of U-Net and Faster RCNN was 0.90 and 0.67, respectively). Conclusion Based on the convolutional neural networks of deep learning algorithm, a novel Ali-U-Net algorithm was successfully constructed to detect the alveolar bone loss in panoramic radiographs and displayed a better diagnostic value for periodontitis.

关 键 词：牙周炎 深度学习 智能诊断 口腔曲面体层片 卷积神经网络 

分 类 号：R781.4[医药卫生—口腔医学]