文献速递口腔修复与种植

基于CBCT的AI深度学习自动检测单牙缺失区与邻近结构

Exploring AI-driven deep learning approaches for optimizing space detection in single gap implantation based on CBCT images.
PubMed 文献 · J Dent · 2026-03-14

本研究提出一种基于nnUNet的DentalSegmentator框架、无需额外手动分割的三步序列自动检测单牙缺失区及邻近结构的方法。自动检测耗时3.3秒,较专家120秒减少37倍,表明该方法可靠、高效,可用于种植术前规划。

要点

  • 三步序列法可自动检测单牙缺失区及邻近结构,无需手动分割。
  • 检测性能F1达0.96,与三位经验牙医标注对照验证。
  • 自动检测耗时3.3秒,较专家120秒减少37倍。

核心结论

  • 所提出的自动检测方法在单牙缺失区检测中表现出可靠性、时间效率和一致性,可作为种植术前规划的有价值工具。

研究设计与样本

  • 解剖结构首先使用DentalSegmentator预训练模型分割,随后通过形态学图像处理、Watershed算法、粒子分析和决策树计算检测。
  • 性能以三位经验牙医的标注为参照,采用准确率、精确率、召回率、特异性和F1-score验证,计算时间用Wilcoxon符号秩检验分析(p < 0.001)。

主要发现

  • 自动检测的准确率、精确率、召回率、特异性和F1-score分别为0.95、0.95、0.96、0.93和0.96。
  • 自动检测时间为3.3秒,专家耗时120秒,处理时间减少37倍。

临床怎么用

  • 该方法可用于单牙缺失种植术前规划,自动检测缺失区及邻近结构,减少对专家手动标注的依赖。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

OBJECTIVE: Computer-assisted implant surgery (CAIS) is a valuable tool for improving implantation accuracy and efficiency, but its clinical implementation remains time consuming and relies on expert experience, particularly for anatomical structure annotation. While AI-driven analysis offers a solution, limited studies have assessed its use for evaluating single edentulous regions. To address this gap, this study introduces a novel, simple, three-step sequential approach for automated detection of single edentulous areas and adjacent structures, leveraging structure segmentation from the nnUNet-based DentalSegmentator framework without requiring further manual segmentation. METHODS: Sixty-six cone-beam computed tomography (CBCT) scans, acquired from four different machines and encompassing a total of 80 single edentulous regions, were divided into parameter tuning and validation cohorts. Anatomical structures were first segmented using the DentalSegmentator with the available pre-trained model. Subsequently, detection was computed using morphological image processing, the Watershed algorithm, particle analysis, and a decision tree. Performance was validated against the annotations of three experienced dentists using accuracy, precision, recall, specificity and F1-score. Finally, computational time was analyzed using the Wilcoxon signed-rank test (p < 0.001). RESULTS: The approach achieved high performance, with accuracy, precision, recall, specificity, and F1-score values of 0.95, 0.95, 0.96, 0.93, and 0.96, respectively. The automated detection time (3.3 s) represents a 37-fold reduction in processing time compared to experts (120 s). CONCLUSIONS: The proposed automated detection approach demonstrated reliability, time efficiency, and consistency, positioning it as a valuable asset for pre-surgical implant planning.

来源:J Dent,查看原文。版权归原作者所有。

引用

  1. Anupuntanun P, Arunjaroensuk S, Narkbuakaew W, et al. Exploring AI-driven deep learning approaches for optimizing space detection in single gap implantation based on CBCT images. J Dent. 2026. PMID: 41833743. doi:10.1016/j.jdent.2026.106631

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。