Accuracy and Time Efficiency of Automated Tooth Segmentation in Dental Imaging-A Systematic Review and Meta-Analysis.

Abstract

This systematic review examined the accuracy and efficiency of AI-based automated tooth segmentation methods compared to manual or ground truth techniques. A comprehensive search was conducted in MEDLINE (via PubMed), the Cochrane Central Register of Controlled Trials, ScienceDirect, SciELO, LILACS, the German National Library of Medicine, IEEE Xplore, Web of Science and grey literature sources (OpenGrey) up to 1 January 2024, without restrictions. The Quality Assessment Tool for Diagnostic Accuracy Studies-2 (QUADAS-2) was used to evaluate the risk of bias in the included studies. Forty-two studies were included in the systematic review, of which 37 were included in the meta-analysis. Sensitivity and specificity values ranged from 0.75 to 1 and 0.85 to 1, respectively, indicating good segmentation accuracy. Comparisons of the dice segmentation coefficient between AI and manual methods showed no significant difference (SMD = 0.05, p = 0.9), whereas ground truth AI algorithms outperformed proposed AI algorithms (SMD = 2.42, p < 0.00001). The Hausdorff distance revealed no significant difference between AI and manual methods, but proposed AI algorithms demonstrated superiority over ground truth AI algorithms (SMD = -5.76, p < 0.01). AI algorithms were also significantly faster than manual methods. Current evidence suggests that AI algorithms for tooth segmentation perform comparably to manual segmentation. Moreover, recent automated algorithms have shown superior performance compared to ground truth algorithms. However, these findings should be interpreted cautiously due to the very low certainty of evidence, largely attributed to bias and high heterogeneity. Further well-designed and rigorously reported studies are necessary.