Machine learning regression models for color prediction of CAD-CAM materials against different tooth-colored backgrounds

OBJECTIVES: To develop and assess the accuracy of a color prediction model for CAD-CAM materials with three thicknesses against tooth-colored backgrounds. MATERIALS AND METHODS: Specimens of Lava Ultimate, Grandio Blocs, VITA Enamic, and Vita Mark II with thicknesses of 0.5, 1.0, and 1.5 mm (n = 3) were studied. A spectroradiometer was used to obtain CIE-L*a*b* coordinates against white, black, and nine colored (ND1-ND9) backgrounds. A leave-one-out cross-validation (LOOCV) strategy was used to develop and evaluate the predictive models. Partial Least Squares (PLS) regression models were employed to predict L*, a*, and b* based on the input variables data: material, thickness, and background. Root Mean Square Error (RMSE) was used as performance assessment. CIEDE2000 color differences (ΔE00) between measured and predicted data were calculated, and accuracy was assessed by comparing ΔE00 values with 50:50 % acceptability and perceptibility thresholds. RESULTS: The integration of the predictive methods resulted in acceptable (ΔE00<1.81) color difference values for all models and imperceptible ΔE00 (ΔE00<0.80) when data were separated by material and thickness. The best model fit was consistently observed for the a* coordinate, showing the RMSE values closer to zero, with the separation of input data on material and thickness demonstrating the highest predictive adjustment due to the lowest RMSE=0.14. Predictive models performed better for greater thicknesses. CONCLUSIONS: The developed integration of LOOCV and PLS regression produced a stable, reproducible, and clinically applicable predictive model for CAD-CAM restoratives with varying thicknesses and across different tooth-colored backgrounds, demonstrating acceptable color differences. CLINICAL SIGNIFICANCE: Integrating the machine learning regression models into clinical workflows may be a promising tool for achieving more predictable, efficient, and high-quality aesthetic indirect restorations, even when dealing with varying abutment tooth shades.