首页 > 资讯 > 基于口内数码照图像深度学习的牙周病早期筛查研究

基于口内数码照图像深度学习的牙周病早期筛查研究

泰然健康网
2025-07-10 03:33

摘要： 目的: 基于卷积神经网络(convolutional neural network,CNN)的深度学习技术构建人工智能(artificial intelligence,AI)牙周病早筛模型,辅助非牙周医生对牙周病进行早期筛查。方法: 收集南昌大学第二附属医院口腔医学诊疗中心就诊的牙周非健康人群以及牙周健康人群的口内数码照和临床资料。基于VGG-16结构对口内数码照图像进行训练和测试,建立口腔九宫格、正位咬合、正位咬合(剔除无效背景)3种训练集模型。结果: 共收集到578位研究对象的3869张口内数码照图像,其中牙周健康图像2230张,牙周非健康图像1639张。采用VGG-16结构建立3种训练集模型,对九宫格口内数码照、正位咬合口内数码照、正位咬合(剔除无效背景)口内数码照预测的准确度分别为66.62%、64.66%、77.44%,曲线下面积(area under curve,AUC)值分别为0.651、0.767、0.784。结论: 本研究构建的VGG-16模型能有效通过对口内数码照图像识别,辅助非牙周医生对牙周病进行早筛。

关键词: 卷积神经网络, 牙周病, 深度学习, 人工智能

Abstract: Objective: To construct an artificial intelligence (AI) early screening model of periodontal disease based on convolutional neural network (CNN) deep learning technology, and to assist non-periodontal doctors in early screening of periodontal disease. Methods: The oral digital photos and clinical data of periodontal non-healthy people and periodontal healthy people were collected from the Second Affiliated Hospital of Nanchang University. Vgg-16 was used to train and test intra oral digital images. Three training models, i.e. nine grid mouth, orthotopic occlusal, and orthotopic occlusal excluding invalid background, were established. Results: A total of 3869 oral digital images of 578 subjects were collected, including 2230 periodontal healthy images and 1639 periodontal unhealthy images. Vgg-16 was used to establish three kinds of training set models. The accuracy of prediction of digital image in nine grid mouth, digital image in orthotopic occlusal mouth, and digital image in orthotopic occlusal mouth excluding invalid background were 66.62%, 64.66%, and 77.44%, respectively. AUC values were 0.651, 0.767, and 0.784, respectively. Conclusion: The VGG-16 model constructed in this study can effectively assist non-periodontal doctors in early screening of periodontal disease through intra-oral digital image recognition.

Key words: convolutional neural network, periodontal disease, deep learning, artificial intelligence