Artificial Intelligence for Cone Beam Evaluation of Lower Third Molars and Possible Postoperative Nerve Injury: A Systematic Review

AIM: The risk of injuring the inferior alveolar nerve (IAN) is always present during the surgical removal of mandibular third molars. Cone beam computed tomography (CBCT) helps visualize their anatomical relationship, but interpretations vary between operators. Computational models with deep learning can provide constant evaluation of third molar position and canal proximity, to help in improving preoperative assessment, and thus reduce this risk. METHODS: This systematic review followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines to evaluate the effectiveness of computational models in predicting IAN injury using CBCT of lower third molars. A PICOS (P, Population; I, Intervention; C, Comparison; O, Outcome; S, Study design) framework guided study selection. Several databases, like PubMed, Scopus, etc., were searched for studies published between October 2020 and February 2025. From the 983 records screened, 7 studies met eligibility. Three individual reviewers extracted the data and examined the study quality using the Newcastle-Ottawa Scale to determine bias and methodological rigor. RESULTS: Seven high-quality studies examined computational models for mandibular canal (MC) segmentation using CBCT and panoramic imaging. Three-dimensional U-shaped Convolutional Neural Network (U-Net) models reached up to 99% accuracy with Dice similarity coefficients (DSC) between 0.76 and 0.782, outperforming two-dimensional models. Classification networks such as DenseNet201 and MobileNet achieved over 98% accuracy and F1-scores up to 0.94. Segmentation was more than 100 times faster than manual tracing. Metrics included DSC values up to 0.9730 and Hausdorff Distance as low as 0.705 mm. Quality of studies scores ranged from 9 to 10 and risk of bias was low in maximum domains. CONCLUSIONS: Computational models show significant potential to improve preoperative assessment and predict IAN injury during mandibular third molar surgeries by enhancing MC segmentation and diagnostic accuracy.