Related Experiment Video
Updated: Mar 17, 2026
Accuracy in Dental Medicine, A New Way to Measure Trueness and Precision
Published on: April 29, 2014
DIAGNOSTIC PERFORMANCE OF MACHINE LEARNING-AIDED PROXIMAL CARIES DETECTION: A SYSTEMATIC REVIEW AND META-ANALYSIS
Kaixin Guo1, Madeline Jun Yu Yon2, Yanqi Yang3
1Paediatric Dentistry, Faculty of Dentistry, The University of Hong Kong, Pokfulam, Hong Kong SAR, Hong Kong.
Objectives:
The systematic review aimed to evaluate the diagnostic performance of machine learning in detecting proximal caries through radiographic modalities compared to expert consensus.
Materials And Methods:
A strategic literature search was carried out across 4 electronic databases, with 2 independent reviewers for screening and data extraction. Measures as sensitivity, specificity, positive (LR+) and negative likelihood ratios (LR-), area under the receiver operating characteristic curve (AUC-ROC) and diagnostic odds ratio (DOR) were calculated. Risk of bias was assessed using the Quality Assessment tool for Diagnostic Accuracy Studies (QUADAS-2). Subgroup analysis was performed based on the data modalities and image augmentation. The certainty of evidence was evaluated according to the Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) framework.
Results:
Following a thorough review of 2439 studies, 28 studies met the inclusion criteria. Eleven studies were included in the meta-analysis. The pooled sensitivity and specificity were 0.84 (95% Confidence Interval [95% CI] 0.78-0.88) and 0.93 (95% CI 0.90-0.97) respectively. The LR+ and LR- were 14.00 (95% CI 8.30-23.40) and 0.17 (95% CI 0.12-0.23) respectively. The pooled AUC-ROC was 0.95 (95% CI 0.17-1.00), and DOR was 84.08 (95% CI 40.54-174.40). The limited certainty of evidence indicated that all radiographic modalities have comparable specificity in detecting proximal caries, while cone beam computer tomography demonstrates significantly higher sensitivity. No difference between the presence or absence of image augmentation prior to machine learning model training was found. Most studies (22/28) showed high risk of bias. Certainty of evidence was low.
Conclusions:
Machine learning models demonstrated acceptable diagnostic performance for proximal caries detection, with cone beam computed tomography (CBCT) potentially offering higher sensitivity. However, future studies to investigate the AI performance across various depths and types of dentitions are warranted due to limited evidence certainty.
