Related Experiment Video
Updated: Sep 11, 2025
Author Spotlight: Advancing CBCT and Digital Dental Image Integration with AI-Assisted Digitization
Published on: February 23, 2024
Validity, Reliability and Reproducibility of Artificial Intelligence-Based Methods in Tooth Widths, Bolton Ratios and
Rumeysa Bilici Geçer1, Aslı Pelin Kaya Bursa2, Nilüfer Altay2
1Department of Orthodontics, Hamidiye Faculty of Dental Medicine, University of Health Sciences, Istanbul, Turkey.
Objectives:
To evaluate the validity, reliability and reproducibility of manual, fully automated and semi-automated artificial intelligence (AI)-based methods for measuring tooth widths, calculating Bolton ratios and performing space analysis.
Materials And Methods:
A total of 102 plaster models and 102 corresponding occlusal photographs were analysed using three methods: manual, fully automated AI and semi-automated AI. For each case, mesiodistal tooth widths were measured, and the anterior and overall Bolton ratios were calculated, along with the required and available space and the space discrepancy in both arches. Validity was assessed using repeated measures ANOVA, reliability using Pearson's correlation coefficients and reproducibility using intraclass correlation coefficients.
Results:
Statistically significant differences were observed in mean tooth widths among the three methods (p < 0.05), with the semi-automated AI method producing the highest values. No significant differences were found in Bolton ratios (p > 0.05). Significant differences were noted between T1 and T2 measurements for all methods (p < 0.05). The fully automated AI method showed the highest intra-examiner reliability in tooth width, Bolton ratios and space analysis (r = 0.873-0.996), followed by the manual (r = 0.449-0.961) and semi-automated methods (r = 0.248-0.739). All methods showed excellent reproducibility in tooth width (ICC: 0.966-0.983). However, the manual method had the highest reproducibility in Bolton ratios (ICC: ABR = 0.964; OBR = 0.960), while AI-based methods showed lower reliability, especially in derived measurements such as Bolton ratios and upper arch discrepancies.
Conclusion:
The semi-automated AI method produced higher tooth width and space analysis values than the manual method, while the fully automated method exhibited greater variability. Although Bolton ratios were statistically similar, variability in individual measurements was more pronounced with AI methods. While all methods demonstrated high reproducibility for tooth width measurements, reduced consistency was observed for AI-based methods in derived outcomes such as Bolton ratios and space discrepancies.
