Prediction of tooth color at different preparation depths using artificial intelligence

Objective: The aim of this study is to use artificial intelligence algorithms to predict the color changes that will occur at different preparation depths in anterior teeth before any preparation is performed. Method: A total of 106 extracted anterior teeth without caries, restorations, or endodontic treatment were included. The vestibular surfaces were prepared at depths of 0.3, 0.5, 0.7, 1.0, and 1.5 mm. Color measurements were performed with the Vita Easyshade V spectrophotometer, and L*, a*, b* values were recorded at each stage. ΔE00 values were calculated relative to the baseline measurements. Linear regression, polynomial regression, decision tree regression, random forest regression, and support vector regression models were applied to predict L*, a*, b* based on preparation depth. In addition, statistical analyses were conducted to compare color changes across different depths. Results: A significant increase in ΔL and ΔE values was observed as the preparation depth increased. Δb values were significantly higher in maxillary central incisors compared to other tooth groups. All machine learning models produced low R² values, indicating that depth alone was insufficient for predicting L*, a*, and b* values. Conclusion: As preparation depth increased, the influence of dentin became more pronounced, leading to significant increases particularly in ΔL and ΔE. While ΔL changes remained limited at minimal preparation levels (0.3–0.5 mm), a marked increase was observed from 0.7 to 1.5 mm. Δa and Δb values showed no consistent depth-related pattern overall. Tooth-type differences were evident: Δb increase was more pronounced in maxillary central incisors, whereas canine teeth demonstrated higher ΔL and ΔE changes. Even at 0.3 mm preparation, most samples exhibited ΔE00 > 0.8, indicating that minimal preparation can still cause clinically perceptible color changes. The low R² values obtained from all machine learning models suggest that depth alone is insufficient to accurately predict natural tooth color.