Artificial intelligence-based segmentation of mandibular canal on cone-beam computed tomography: a systematic review and meta-analysis.
Authors
Affiliations (9)
Affiliations (9)
- Periodontology and Oral Diagnosis, Faculty of Dentistry, King Salman International University, South Sinai, El Tur, 11371, Egypt. [email protected].
- Undergraduate student, Faculty of Dentistry, King Salman International University, El-Tor, South Sinai, Egypt.
- Faculty of medicine, Merowe university of technology, Northern, Sudan.
- Faculty of Medicine, University College of Medicine and Dentistry, Lahore, Pakistan.
- Department of Medicine, Universitas Padjadjaran, Bandung, West Java, Indonesia.
- Faculty Of Dentistry, Al Azhar University, Cairo, Egypt.
- Faculty of Dentistry, Galala University, Suez, Egypt.
- College of Dentistry, King Saud University, Riyadh, Saudi Arabia.
- Removable prosthodontic division, Faculty of Dentistry, King Salman International University, South Sinai, El Tur, Egypt.
Abstract
To systematically evaluate the accuracy and generalizability of artificial intelligence (AI)-based segmentation of mandibular canal and related anatomical structures on cone-beam computed tomography (CBCT) and to examine methodological factors influencing reported performance. A systematic search of PubMed/MEDLINE, Scopus, Web of Science, Cochrane Library, and Wiley Online Library was conducted through April 10, 2026. Studies evaluating AI-based segmentation of the mandibular canal or related anatomical structures on CBCT were included. Risk of bias was assessed using QUADAS-2 tool. Random-effects meta-analysis using restricted maximum likelihood estimation was performed to pool segmentation performance, with the Dice Similarity Coefficient (DSC) as the primary outcome. Subgroup analyses, meta-regression, publication bias assessment, and GRADE certainty evaluation were conducted. Fifty-nine studies were included qualitatively, and 46 studies were eligible for quantitative synthesis. Across 40 effect sizes, the pooled DSC was 0.806 (95% CI: 0.773-0.839), indicating high average segmentation performance, although heterogeneity was substantial (I² = 99.68%). Pooled secondary outcomes were 0.757 for Intersection-over-Union, 2.085 mm for HD95, and 0.498 mm for mean distance error. No significant differences were observed according to validation strategy or AI architecture. Publication year was the only signific. AI-based segmentation of mandibular canals on CBCT demonstrates promising performance; however, substantial heterogeneity, limited external validation, and low certainty of evidence restrict confidence in its generalizability. AI systems should be considered adjunctive tools, and further high-quality, externally validated studies with standardized methodologies are required to support reliable clinical implementation. AI-assisted mandibular canal segmentation may improve workflow efficiency and support treatment planning in implant dentistry and oral surgery. Nevertheless, clinician oversight remains essential because current evidence does not support fully autonomous clinical implementation.