Effectiveness of Artificial Intelligence in Detection of Apical Periodontitis: A Systematic Review

Abstract

ABSTRACT:

Background:
Artificial intelligence (AI), particularly deep learning and convolutional neural networks (CNNs), is increasingly being investigated for the detection of periapical lesions and radiographic manifestations of apical periodontitis. However, reported diagnostic performance varies according to imaging modality, AI architecture, dataset characteristics, reference standards, and validation strategies.

Objective:
To synthesize the available evidence on the diagnostic effectiveness of AI-based systems for detecting periapical lesions and radiographic manifestations of apical periodontitis on periapical, panoramic, and cone-beam computed tomography (CBCT) images, and to summarize methodological limitations and potential sources of bias.

Methods:
This systematic review was structured and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 framework. Published systematic reviews and eligible primary diagnostic-accuracy studies were evaluated, with particular attention to sensitivity, specificity, accuracy, F1 score, area under the receiver operating characteristic curve (AUROC), imaging modality, AI architecture, reference standard, and validation strategy. Study-level methodological quality and risk of bias were considered using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) framework, with additional consideration of AI-specific issues including data leakage, train-test separation, and external validation.

Results:
A total of 12 primary studies were included in the present systematic review, encompassing AI-based detection, classification, localization, segmentation, and AI-assisted clinician assessment across periapical radiographs, panoramic radiographs, and CBCT. Existing systematic reviews consistently indicated promising diagnostic performance but substantial methodological heterogeneity. A 2024 systematic review and meta-analysis included 24 studies and reported pooled sensitivity of 0.94 and specificity of 0.96 from four studies contributing quantitative data. A 2026 systematic review of deep-learning detection of periapical radiolucent lesions on panoramic radiographs included 30 studies, of which six entered meta-analysis, reporting pooled sensitivity of 0.80 (95% CI: 0.49–0.94), specificity of 0.98 (95% CI: 0.87–1.00), diagnostic odds ratio (DOR) of 176.37, and AUROC of 0.93. Risk-of-bias concerns were identified in 16 of 30 studies and applicability concerns in 12. The primary studies included in the present review demonstrated promising diagnostic and detection performance; however, external validation, representative datasets, and clinically relevant reference standards were inconsistently reported.

Conclusion:
AI demonstrates considerable potential as an adjunctive tool for detecting radiographic manifestations of apical periodontitis and periapical lesions. Nevertheless, reported high diagnostic performance should be interpreted cautiously because of substantial methodological heterogeneity, potential spectrum and selection bias, limited external validation, variability in reference standards, and other AI-specific methodological limitations. Further well-designed studies using representative clinical datasets, independent external validation, standardized reference standards, and transparent reporting are required before widespread clinical implementation.