摘要：

目的： 评估深度学习（deep learning，DL）在检测牙周放射学骨丧失（radiographic bone loss，RBL）中的准确性。方法： 按检索策略在PubMed、Scopus、Google Scholar、Web of Science数据库中检索相关文献，检索时限为各数据库建库之日起至2024年3月，语言限制为英文。采用诊断准确性研究质量评价工具第2版（Quality Assessment of Diagnostic Accuracy Studies-2，QUADAS-2）对研究质量进行评估。结果： 对纳入的19篇符合标准的文章进行全文筛选，将其中的8篇文章纳入Meta分析。Meta分析结果显示，DL模型在牙周炎分类中的灵敏度为0.84，95%置信区间（confidence interval，CI）=0.79~0.90，特异度为0.83，95%CI=0.75~0.91。综合受试者工作特征曲线下面积（summary receiver operating characteristic-area under the curve，SROC-AUC）为0.92，95%CI=0.89~0.94。结论： DL模型检测牙周RBL有较好的准确率和灵敏度。

关键词: 牙周病, 深度学习, 放射学骨丧失, 卷积神经网络

Abstract:

Objective: To evaluate the accuracy of deep learning (DL) in detecting periodontal radiographic bone loss (RBL). Methods: A systematic search of the literature was conducted in PubMed, Scopus, Google Scholar, and Web of Science for relevant studies, the search covered the period from the inception of each database until March 2024, with language restrictions set to English. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was used to assess the quality of the included studies. Results: A total of 19 articles meeting the criteria were included for full-text screening, with 8 of them being incorporated into the Meta-analysis. The Meta-analysis results showed that the DL models achieved a sensitivity of 0.84 [95% confidence interval (CI)=0.79 to 0.90] and a specificity of 0.83 (95% CI=0.75 to 0.91) in the classification of periodontitis. The summary receiver operating characteristic-area under the curve (SROC-AUC) was 0.92 (95% CI=0.89 to 0.94). Conclusion: DL models demonstrate good accuracy and sensitivity in detecting periodontal RBL.

Key words: periodontal diseases, deep learning, radiographic bone loss, convolutional neural networks

中图分类号: 

• R782