Use of artificial intelligence for detection of MB2 canals in maxillary first molars on CBCT: a systematic review and meta-analysis.
Authors
Affiliations (9)
Affiliations (9)
- Dentofacial Deformities Research Centre, Research Institute of Dental Sciences, Shahid Beheshti University of Medical Sciences, Tehran, Iran. [email protected].
- UCLA School of Dentistry, Los Angeles, CA, USA.
- Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Azad Tehran University of Medical Science, Tehran, Iran.
- School of Dentistry, Faculty of Medicine and Dentistry, University of Alberta, Edmonton, T6G 1C9, AB, Canada.
- School of Dentistry, Başkent University, Ankara, Türkiye.
- Department of Oral and Maxillofacial Radiology, School of Dentistry, Shahid Sadoughi University of Medical Science, Yazd, Iran.
- Department of Prosthodontics and Dental Implantology, College of Dentistry, King Faisal University, Al-Ahsa, 31982, Saudi Arabia.
- Centre for Artificial Intelligence and Innovation (CAII), Faculty of Dentistry, Chulalongkorn University, 34 Henri-Dunant Road, Wang-Mai, Pathumwan, Bangkok, 10330, Thailand.
- Centre for Artificial Intelligence and Innovation (CAII), Faculty of Dentistry, Chulalongkorn University, 34 Henri-Dunant Road, Wang-Mai, Pathumwan, Bangkok, 10330, Thailand. [email protected].
Abstract
Detecting the second mesiobuccal (MB2) canal in maxillary first molars is challenging, even with cone-beam computed tomography (CBCT). Artificial intelligence (AI), especially deep learning, has been explored as a tool to aid detection. This systematic review and meta-analysis evaluated the diagnostic accuracy of AI in identifying MB2 canals on CBCT. Following PRISMA guidelines, a comprehensive electronic search across five databases (PubMed, Scopus, Web of Science, Embase, and Scopus Secondary) retrieved 52 articles. After removing duplicates and screening titles/abstracts, 7 full texts were assessed, of which 4 met the inclusion criteria. Studies were eligible if they applied AI algorithms for MB2 detection in CBCT images and reported diagnostic performance outcomes. Data extraction included study design, dataset size, AI model architecture, and diagnostic metrics. Pooled estimates of sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were calculated using a random-effects model. Heterogeneity was assessed with the I² statistic, and publication bias was evaluated with Egger's test. Four studies were included. AI models achieved a pooled sensitivity of 0.82 and specificity of 0.74. Deep learning models outperformed traditional machine learning, with higher sensitivity (0.87 vs. 0.80), specificity (0.90 vs. 0.68), and accuracy (0.84 vs. 0.75). Considerable heterogeneity and small sample sizes limited generalizability. AI, particularly deep learning, shows promise in detecting MB2 canals on CBCT. While current evidence is preliminary, standardised AI training and reporting protocols, together with larger multicenter studies, are needed to validate these tools. Clinically, AI could serve as a supplementary aid to improve diagnostic consistency and reduce missed canals during endodontic treatment.