Department of Conservative Dentistry and Periodontology, University Hospital, Ludwig-Maximilians University Munich, Goethestraße 70, 80336, Munich, Germany.
Institute for Software Engineering, University of Duisburg-Essen, Essen, Germany.
Clin Oral Investig. 2022 Sep;26(9):5923-5930. doi: 10.1007/s00784-022-04552-4. Epub 2022 May 24.
The aim of this study was to develop and validate a deep learning-based convolutional neural network (CNN) for the automated detection and categorization of teeth affected by molar-incisor-hypomineralization (MIH) on intraoral photographs.
The data set consisted of 3241 intraoral images (767 teeth with no MIH/no intervention, 76 with no MIH/atypical restoration, 742 with no MIH/sealant, 815 with demarcated opacity/no intervention, 158 with demarcated opacity/atypical restoration, 181 with demarcated opacity/sealant, 290 with enamel breakdown/no intervention, 169 with enamel breakdown/atypical restoration, and 43 with enamel breakdown/sealant). These images were divided into a training (N = 2596) and a test sample (N = 649). All images were evaluated by an expert group, and each diagnosis served as a reference standard for cyclic training and evaluation of the CNN (ResNeXt-101-32 × 8d). Statistical analysis included the calculation of contingency tables, areas under the receiver operating characteristic curve (AUCs) and saliency maps.
The developed CNN was able to categorize teeth with MIH correctly with an overall diagnostic accuracy of 95.2%. The overall SE and SP amounted to 78.6% and 97.3%, respectively, which indicate that the CNN performed better in healthy teeth compared to those with MIH. The AUC values ranging from 0.873 (enamel breakdown/sealant) to 0.994 (atypical restoration/no MIH).
It was possible to categorize the majority of clinical photographs automatically by using a trained deep learning-based CNN with an acceptably high diagnostic accuracy.
Artificial intelligence-based dental diagnostics may support dental diagnostics in the future regardless of the need to improve accuracy.
本研究旨在开发和验证一种基于深度学习的卷积神经网络（CNN），用于自动检测和分类口腔内照片中受磨牙-切牙-釉质发育不全（MIH）影响的牙齿。
数据集由 3241 张口腔内图像组成（767 颗无 MIH/无干预的牙齿，76 颗无 MIH/异常修复的牙齿，742 颗无 MIH/密封剂的牙齿，815 颗有界定不透明/无干预的牙齿，158 颗有界定不透明/异常修复的牙齿，181 颗有界定不透明/密封剂的牙齿，290 颗有釉质破坏/无干预的牙齿，169 颗有釉质破坏/异常修复的牙齿，43 颗有釉质破坏/密封剂的牙齿）。这些图像被分为训练集（N=2596）和测试样本（N=649）。所有图像均由专家组评估，每个诊断均作为 CNN 的循环训练和评估参考标准（ResNeXt-101-32×8d）。统计分析包括列联表的计算、接收者操作特征曲线（AUC）和显著图。
开发的 CNN 能够正确分类 MIH 牙齿，总体诊断准确率为 95.2%。总体 SE 和 SP 分别为 78.6%和 97.3%，这表明 CNN 在健康牙齿上的表现优于 MIH 牙齿。AUC 值范围从 0.873（釉质破坏/密封剂）到 0.994（异常修复/无 MIH）。
通过使用经过训练的基于深度学习的 CNN 对大多数临床照片进行自动分类，可获得可接受的高诊断准确性。
无论准确性是否需要提高，基于人工智能的牙科诊断将来都可能支持牙科诊断。
