Abstract Objectives This review examines how contemporary artificial intelligence (AI) systems primarily convolutional and artificial neural networks support diagnosis and decision-making across core endodontic imaging tasks, and synthesizes where these tools add clinical value, where evidence is limited, and what is needed for responsible integration into practice. Materials and methods A narrative review with predefined, transparent elements was undertaken. Searches spanned Google Scholar, PubMed, NCBI, and ResearchGate using controlled terms related to AI and endodontics; eligibility focused on studies reporting diagnostic or algorithmic performance on periapical, panoramic, or CBCT imaging for tasks such as periapical-lesion detection, vertical root fracture (VRF) identification, working-length/apical-foramen estimation, canal-morphology assessment, and separated-instrument detection. Results Nine studies meeting the eligibility criteria were included in the synthesis., AI models achieved high diagnostic performance across all tasks: accuracy for periapical lesion detection ranged from 93 to 97%, while vertical root fracture (VRF) detection on CBCT reached 96.6% to 97.8%. For working length estimation, ANN models demonstrated 93–96% accuracy, and canal morphology identification showed performance between 91% and 95%. Performance for separated instrument detection ranged from 88 to 94%. Model efficacy was significantly influenced by the imaging modality (3D CBCT vs. 2D radiographs) and the robustness of the reference standard used for ground-truthing. Limitations The evidence base is constrained by methodological heterogeneity (small or imbalanced samples, inconsistent ground truths, limited external validation) and model-level issues (black-box behaviour, bias), as well as system-level barriers involving workflow integration, user adoption, and data privacy/security. Conclusions AI presently functions best as an adjunct that can standardize detection, reduce observer variability, and support measurements when radiographic cues are subtle rather than as a replacement for clinician judgment. Future progress hinges on community datasets that reflect real-world diversity, standardized reporting beyond accuracy (e.g., F1, AUC/PR-AUC), external/prospective validation, and explainable, privacy-by-design systems to enable safe clinical translation. Clinical relevance With careful attention to methods, explainability, and data governance, AI can help clinicians see earlier and measure more consistently, improving confidence and communication in everyday endodontic care.
Emshal et al. (Sat,) studied this question.