文献速递牙体牙髓病

计算机辅助测量检测锥形束CT图像中牙邻面龋的准确性

[Accuracy of computer aided measurement for detecting dental proximal caries lesions in images of cone-beam computed tomography].
PubMed 文献 · Zhonghua Kou Qiang Yi Xue Za Zhi · 2017-01-01

研究建立并验证用于检测锥形束CT图像中牙邻面龋的计算机程序。以组织学检测为金标准,对北京大学口腔医学院·口腔医院45颗离体牙的90个邻面进行检测,程序诊断的ROC曲线下面积为0.703,高于评估者肉眼影像诊断的0.632,差异有统计学意义(P=0.023)。

要点

  • 程序检测邻面龋的ROC曲线下面积为0.703,评估者肉眼影像诊断为0.632(P=0.023)。
  • 程序对龋深度的评价与金标准的相关系数r(s)=0.525,高于评估者的r(s)=0.457(P=0.000)。
  • 建立的图像处理与分析系统能提高锥形束CT图像中邻面龋的诊断准确性。

核心结论

  • 建立的图像处理与分析系统能提高锥形束CT图像中邻面龋的诊断准确性。

研究设计与样本

  • 建立龋病锥形束CT图像处理与分析系统,由1名评估者和图像处理分析系统分别对北京大学口腔医学院·口腔医院45颗离体牙锥形束CT图像中的邻面龋进行评估,并以组织学检测作为金标准进行受试者工作特征曲线分析、Spearman相关关系检验。
  • 共检查45颗离体牙的90个邻面,牙齿经锥形束CT扫描仪(Promax 3D)扫描。

主要发现

  • 对锥形束CT图像中邻面龋诊断,评估者肉眼影像诊断与图像处理分析系统的受试者工作特征曲线下面积分别为0.632和0.703,二者差异有统计学意义(P=0.023)。
  • 图像处理与分析系统对邻面龋深度的评价与金标准的相关关系(相关系数r(s)=0.525)显著高于观察者肉眼影像诊断(r(s)=0.457),差异有统计学意义(P=0.000)。

局限性

  • 原文未报告局限性。

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: To establish and validate a computer program used to aid the detection of dental proximal caries in the images cone beam computed tomography (CBCT) images. Methods: According to the characteristics of caries lesions in X-ray images, a computer aided detection program for proximal caries was established with Matlab and Visual C++. The whole process for caries lesion detection included image import and preprocessing, measuring average gray value of air area, choosing region of interest and calculating gray value, defining the caries areas. The program was used to examine 90 proximal surfaces from 45 extracted human teeth collected from Peking University School and Hospital of Stomatology. The teeth were then scanned with a CBCT scanner (Promax 3D). The proximal surfaces of the teeth were respectively detected by caries detection program and scored by human observer for the extent of lesions with 6-level-scale. With histologic examination serving as the reference standard, the caries detection program and the human observer performances were assessed with receiver operating characteristic (ROC) curves. Student t-test was used to analyze the areas under the ROC curves (AUC) for the differences between caries detection program and human observer. Spearman correlation coefficient was used to analyze the detection accuracy of caries depth. Results: For the diagnosis of proximal caries in CBCT images, the AUC values of human observers and caries detection program were 0.632 and 0.703, respectively. There was a statistically significant difference between the AUC values (P=0.023). The correlation between program performance and gold standard (correlation coefficient r(s)=0.525) was higher than that of observer performance and gold standard (r(s)=0.457) and there was a statistically significant difference between the correlation coefficients (P=0.000). Conclusions: The program that automatically detects dental proximal caries lesions could improve the diagnostic value of CBCT images. 目的: 建立图像处理与分析系统并以此检测口腔颌面锥形束CT图像中的邻面龋,评价其检测准确性,为邻面龋的早期诊断提供参考。 方法: 建立龋病锥形束CT图像处理与分析系统。由1名评估者和图像处理分析系统分别对北京大学口腔医学院·口腔医院45颗离体牙锥形束CT图像中的邻面龋进行评估,并以组织学检测作为金标准进行受试者工作特征曲线分析、Spearman相关关系检验。 结果: 对锥形束CT图像中邻面龋诊断,评价者肉眼影像诊断与图像处理分析系统诊断的受试者工作特征曲线下面积分别为0.632和0.703,图像处理与分析系统较评价者肉眼影像诊断更准确,二者差异有统计学意义(P=0.023)。图像处理与分析系统对邻面龋深度的评价与金标准的相关关系(相关系数r(s)=0.525)显著高于观察者肉眼影像诊断(r(s)=0.457),差异有统计学意义(P=0.000)。 结论: 建立的图像处理与分析系统能提高锥形束CT图像中邻面龋的诊断准确性。.

来源:Zhonghua Kou Qiang Yi Xue Za Zhi,查看原文。版权归原作者所有。

引用

  1. Zhang ZL, Li JP, Li G, et al. [Accuracy of computer aided measurement for detecting dental proximal caries lesions in images of cone-beam computed tomography]. Zhonghua Kou Qiang Yi Xue Za Zhi. 2017. PMID: 28253585. doi:10.3760/cma.j.issn.1002-0098.2017.02.010

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。