Artificial intelligence for detecting C-shaped canals on dental radiographs: A systematic review and bayesian diagnostic accuracy meta-analysis.
Authors
Affiliations (6)
Affiliations (6)
- Department of Dentistry, University of São Paulo, São Paulo, Brazil. Electronic address: [email protected].
- Department of Restorative Dentistry, Federal University of Minas Gerais, Minas Gerais, Brazil.
- Department of Restorative Dentistry, Catholic University of Brasília, Brasília, Brazil.
- Department of Restorative Dentistry, São Judas Tadeu University, São Paulo, Brazil.
- Department of Restorative Dentistry, São Leopoldo Mandic School of Dentistry, Campinas, Brazil.
- Department of Dentistry, University of São Paulo, São Paulo, Brazil.
Abstract
To evaluate AI-based identification of C-shaped canals across dental radiographic modalities, with an exploratory synthesis of conventional panoramic detection using cone-beam computed tomography (CBCT) reference labels. Seven sources were searched through August 12, 2026. Peer-reviewed original studies were appraised using QUADAS-3, STARD-AI, and GRADE. Binary panoramic estimates with CBCT reference and auditable counts were pooled using a Bayesian bivariate model nesting external centers within parent studies. CRD420251046384. Nine studies were included; three independent parent studies supplied five cohorts (899C-shaped and 953 non-C-shaped teeth). Pooled sensitivity was 0.918 (95 % credible interval [CrI], 0.837-0.962) and specificity 0.885 (0.588-0.969). Marginal model-implied new-parent-study prediction intervals were 0.716-0.981 and 0.219-0.993, respectively. Only one program provided multicenter external evaluation, with center-specific optimized thresholds. All selected estimates had high risk of bias. Certainty was low for sensitivity and very low for specificity. The sparse evidence does not establish dependable clinical performance. Predictive uncertainty and workflow limitations constrain transportability. Prospective evaluation of prespecified systems and clinical consequences is needed before implementation.