文献速递牙体牙髓病

人工智能在牙科影像中检测根管充填物及评估充填质量:系统综述

Artificial intelligence for detection of root canal fillings and evaluation of obturation quality on dental radiographs: A systematic review.
PubMed 文献 · J Dent · 2026-05-04

本系统综述纳入13项研究、共12,513例检查,评估人工智能模型在根尖片、全景片、咬翼片和CBCT上检测根管充填物及评估充填质量的表现。根管充填物检测的敏感度为0.85–1.00、特异度为0.92–1.00;充填质量评估的敏感度为0.79–1.00、特异度为0.99–1.00。AI模型在根管充填物检测方面有前景,但充填质量评估表现变异较大且依赖成像方式。

要点

  • AI检测根管充填物的敏感度0.85–1.00、特异度0.92–1.00。
  • AI评估充填质量的敏感度0.79–1.00、特异度0.99–1.00,根尖片和CBCT优于全景片。
  • 偏倚风险主要来自患者选择和外验证缺乏,异质性大。

核心结论

  • AI模型在根管充填物检测方面有前景。
  • AI模型对充填质量评估的表现变异较大且依赖成像方式。
  • AI可作为牙髓影像评估的辅助工具。

研究设计与样本

  • 系统综述,检索PubMed、Embase、Scopus、ClinicalTrials.gov、Cochrane Library、Web of Science、Google Scholar和Open Science Framework至2026年3月。
  • 纳入13项研究,共12,513例检查,涵盖根尖片、全景片、咬翼片和CBCT图像。
  • 采用QUADAS-2评估偏倚风险,GRADE评估证据确定性,数据定性合成。

主要发现

  • 根管充填物检测的敏感度为0.85–1.00,特异度为0.92–1.00。
  • 充填质量评估的敏感度为0.79–1.00,特异度为0.99–1.00。
  • 充填质量评估在根尖片和CBCT上的表现优于全景片。
  • 偏倚风险主要与患者选择和外验证缺乏有关,观察到显著异质性。

临床怎么用

  • AI系统可用于自动化影像预筛、图像解读支持和实时第二意见验证。
  • AI工具有助于减少解读时间、提高诊断一致性并支持牙髓病变检出,尤其在高通量场景。
  • 常规临床实施前需进行外验证和前瞻性研究。

局限性

  • 偏倚风险主要与患者选择和外验证缺乏有关。
  • 研究间观察到显著异质性。
  • 证据确定性采用GRADE评估,具体等级未在摘要中报告。

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

OBJECTIVES: To systematically evaluate evidence on artificial intelligence (AI) for root canal filling (RCF) detection and radiographic assessment of obturation quality. DATA: Eligible studies assessed AI-based models for RCF detection and/or obturation quality assessment using dental radiographs and cone-beam computed tomography (CBCT). SOURCES: PubMed, Embase, Scopus, ClinicalTrials.gov, Cochrane Library, Web of Science, Google Scholar, and Open Science Framework were searched up to March 2026. STUDY SELECTION: Studies evaluating AI-based models for RCF detection and/or obturation quality assessment were included. Data were synthesized qualitatively. Risk of bias was assessed with QUADAS-2 and certainty of evidence with GRADE. Thirteen studies were included, comprising 12,513 examinations from periapical, panoramic, bitewing, and CBCT images. For RCF detection, sensitivity ranged from 0.85 to 1.00 and specificity from 0.92 to 1.00. For obturation quality assessment, sensitivity ranged from 0.79 to 1.00 and specificity from 0.99 to 1.00, with better performance in periapical radiographs and CBCT than panoramic radiographs. Risk of bias was mainly related to patient selection and lack of external validation, and substantial heterogeneity was observed. CONCLUSIONS: AI-based models show promise for RCF detection, whereas performance for obturation quality assessment is more variable and modality-dependent. AI may serve as an adjunct in endodontic radiographic evaluation. CLINICAL SIGNIFICANCE: AI-based systems may affect clinical workflows through automated radiographic pre-screening, image interpretation support, and real-time second-opinion validation. These tools may reduce interpretation time, improve diagnostic consistency, and support detection of endodontic findings, particularly in high-volume settings. External validation and prospective studies are required before routine clinical implementation.

来源:J Dent,查看原文。版权归原作者所有。

引用

  1. Guimarães HHDSD, Matos KN. Artificial intelligence for detection of root canal fillings and evaluation of obturation quality on dental radiographs: A systematic review. J Dent. 2026. PMID: 42092623. doi:10.1016/j.jdent.2026.106736

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。