Artificial intelligence models may match the accuracy of manual segmentation in CBCT measurements

• Published in: The journal of evidence-based dental practice p. 102195
• Main Authors: Iqbal, Kisa, Chen, Michael, Wong, Shannon, Chogle, Sami
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.08.2025
• Subjects: Artificial intelligence; CBCT; Convolutional Neural Networks; Dentistry; Meta-analysis; Tooth segmentation; Convolutional Neural Networks; Tooth segmentation; Artificial intelligence; CBCT; Meta-analysis
• ISSN: 1532-3382
• DOI: 10.1016/j.jebdp.2025.102195

The electronic databases search was conducted via Web of Science, Scopus, PubMed, Embase, and IEEE Explore until May 19, 2023. The studies selected in this systematic review and meta-analysis were ones that used AI algorithms to segment teeth in CBCT imaging of human subjects. All studies had to be written in English and include the training set sample sizes and dice similarity coefficient (DSC) score for tooth segmentation. All the studies had to either develop or validate a tooth segmentation model from CBCT imaging using either a neural network or deep learning algorithms. These techniques were compared to conventional manual tooth segmentation. The main outcome measured was focused around the accuracy and efficiency of AI-based tooth segmentation models relative to conventional techniques. The accuracy was measured by the DSC, which measures the overlap between the predicted segmentation and ground-truth segmentation. A higher DSC indicated a more accurate segmentation result. Efficiency was measured in the time it took for segmentation. From 5642 search entries, 35 studies were selected for the systematic review and 18 studies were included for meta-analysis. AI- based techniques for the automation of teeth segmentation from CBCT imaging displayed favorable accuracy. The pooled effect estimate for DSC score for tooth segmentation in 18 studies was 0.96 with a 95% confidence interval of 0.94-0.96. There was a 10 fold reduction in time required to perform tooth segmentation. Egger’s publication bias test resulted in a p-value of 0.772 (>0.05), indicating lack of publication bias. There is a potential for the use of AI algorithms and programming to perform tooth segmentation and furthermore, provide precision in tooth measurements in respect to endodontic, orthodontic and implant treatment planning. These programs can assist in the delineation of teeth and have the potential of improving outcomes regarding the oral care of patients.