Using artificial intelligence to predict the final color of leucite-reinforced ceramic restorations

Predicting the final color of dental restorations remains a clinical challenge. Clinicians rely on visual estimation and empirical methods, which lack precision. Leucite-reinforced ceramics are widely used for veneers due to their esthetic properties. However, the interaction between substrate shade, ceramic thickness, and cement translucency is complex. Prior research has shown that these variables influence color perception. No prior work had resolved how machine learning could model these interactions systematically. This gap motivated the development of a predictive model using regression techniques. The study aimed to address the uncertainty in color prediction by integrating multiple variables.

The study aimed to assess machine learning models' ability to predict ceramic restoration color. It focused on leucite-reinforced ceramics used for veneers with translucent cement. The goal was to determine how substrate shade, ceramic shade, thickness, and translucency interact. The researchers sought to identify the most accurate predictive model for clinical use. They wanted to quantify the influence of each variable on final color outcomes. This approach could improve esthetic outcomes in dental restorations. The study also aimed to validate the model's accuracy using spectrophotometric data. By doing so, clinicians could better anticipate final restoration color before placement.

Leucite-reinforced ceramics were cut into four thicknesses: 0.3, 0.5, 0.7, and 1.2 mm. Specimens were placed on four background substrates: black, white, A1, and A3. A translucent resin cement was used to simulate clinical bonding conditions. CIELab color coordinates were measured using a calibrated spectrophotometer. Each specimen was evaluated under standardized lighting and viewing conditions. Color change values (CIEDE2000) were calculated for each experimental group. Twenty-eight regression models were tested to find the best fit for predicting final color. Each model was optimized by adjusting variable weights to minimize prediction error.

The decision tree regression model showed the lowest mean absolute error in predictions. Substrate shade had the strongest influence on final restoration color (L*, a*, b* values). Ceramic thickness ranked second in impact, followed by ceramic shade and translucency. The model predicted color changes with an accuracy of less than 1.0 ΔE00 in most cases. CIELab values varied significantly across different substrate and ceramic combinations. Translucency of the resin cement contributed to color shifts in all tested groups. The model's accuracy was highest when ceramic thickness exceeded 0.7 mm. These findings suggest that machine learning can enhance color prediction in clinical settings.

The study found that machine learning models can predict ceramic restoration color with high accuracy. The decision tree model outperformed other regression techniques in this context. Substrate shade was the most influential variable in determining final color outcomes. Ceramic thickness and shade also played significant roles in color prediction. Translucency of the cement contributed to color variation in all tested conditions. The model's accuracy was validated using spectrophotometric data from multiple experimental groups. These findings support the use of machine learning in dental color prediction for clinical applications. The authors propose that such models can improve esthetic outcomes in ceramic veneer restorations.