Artificial Intelligence for Assessing the Relationship Between Mandibular Third Molars and the Inferior Alveolar Nerve on Panoramic Radiographs: A Systematic Review
Related Experiment Video
Updated: Oct 31, 2026
Reliability of Artificial Intelligence-Based Cone Beam Computed Tomography Integration with Digital Dental Images
Published on: February 23, 2024
F-K Elwy1, J Parra-Moreno, B González-Navarro
1Sonia Egido Moreno Faculty of Medicine and Health Sciences (Dentistry), University of Barcelona, Bellvitge University Campus. Government Pavilion. Office 2-29. C/Feixa Llarga, s/n 08907-L'Hospitalet de Llobregat (Barcelona) - Spain soniaegido@ub.edu.
Background:
Extracting mandibular third molars is a common procedure that carries a risk of injury to the inferior alveolar nerve (IAN). Panoramic radiographs (PR) are commonly used for preoperative assessment of the mandibular third molars, but they are not very reliable for evaluating the relationship of the IAN. This systematic review aims to evaluate the accuracy of artificial intelligence (AI) models in assessing the relationship between mandibular third molars and the IAN.
Materials And Methods:
We conducted systematic searches in MEDLINE (via PubMed), the Cochrane Library, Embase, and Scopus for studies published between January 2020 and April 2026. Eligible studies evaluated artificial intelligence systems that analyze panoramic radiographs to assess the relationship between mandibular third molars and the IAN, using cone-beam computed tomography (CBCT) or expert assessments as references. The studies had to report key diagnostic performance metrics. Two independent reviewers carried out study selection, data extraction, and risk-of-bias assessments using the QUADAS-2 tool.
Results:
We included eleven studies with more than 9,700 patients. The AI models reviewed, including architectures such as ResNet, VGG, and YOLO, demonstrated promising diagnostic performance. Their accuracy ranged from 0.630 to 0.896, with sensitivity and specificity reaching up to 0.92 and 0.95, respectively. No single architecture was consistently superior, but AI models outperformed less experienced human observers in the studies that included this comparison. Validation against CBCT usually yielded more reliable performance metrics than validation based solely on expert opinion.
Conclusions:
AI models show promise as decision-support tools in the radiographic assessment of the relationship between mandibular third molars and the inferior alveolar nerve on panoramic radiographs. However, variations in datasets, reference standards, and validation methods across studies mean that no single architecture can yet be considered superior. Larger, multicenter, prospective studies with external validation are necessary to confirm these systems' performance before they can be widely used in clinical practice.
