Accuracy of artificial intelligence algorithms in detecting periapical lesions on cone-beam computed tomography images: A systematic review and meta-analysis

Abstract Introduction: Periapical lesions (PALs) are common sequelae of pulpal infection, and accurate detection of these lesions is essential for diagnosis, treatment planning, and long-term prognosis. Although cone-beam computed tomography (CBCT) has improved the identification of periapical pathology, its interpretation remains operator-dependent and subject to interobserver variability. Recent advances in artificial intelligence (AI) have shown promise in automating image analysis and supporting diagnostic decision-making. This study was performed to assess the diagnostic performance of AI-based algorithms in detecting PALs on CBCT images. Materials and Methods: This review follows Preferred Reporting Items for Systematic Reviews and Meta-Analysis – Diagnostic Test Accuracy guidelines. A comprehensive search was conducted in PubMed, SCOPUS, EBSCOhost, and Google Scholar until September 2025. Eligible studies compared AI algorithms with expert examiners or reference standards for PALs detection in CBCT imaging. Methodological quality was assessed using the quality assessment of diagnostic accuracy studies (QUADAS)-2 tool, and meta-analysis was performed for calculating pooled sensitivity, specificity, likelihood ratios, diagnostic odds ratios, and overall accuracy in terms of area under the curve (AUC) through summary receiver-operating characteristics (SROC). Results: Eight studies ( n = 2820 CBCT images) met the inclusion criteria. AI algorithms assessed were artificial neural networks, convolutional neural networks, deep learning, Diagnocat, VGG-16, and DenseNet-121. The pooled sensitivity and specificity were 0.76 (95% confidence interval [CI]: 0.37–0.97) and 0.88 (95% CI: 0.48–1.00), respectively, with an AUC of 0.89, indicating excellent diagnostic performance. Most studies demonstrated low-to-moderate risk of bias (ROB). Conclusion: AI algorithms demonstrated good to excellent diagnostic performance for detecting radiographic signs of periapical disease and may serve as adjunctive tools in endodontic diagnosis. However, their current application is limited to image-based interpretation and should be complemented by clinical assessment.