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

OBJECTIVES: The aim of this study was to evaluate the accuracy of machine learning regression models in predicting the final color of leucite-reinforced glass CAD/CAM ceramic veneer restorations based on substrate shade, ceramic shade, thickness and translucency.

METHODS: Leucite-reinforced glass ceramics in four different shades were sectioned in thicknesses of 0.3, 0.5, 0.7, and 1.2 mm. The CIELab coordinates of each specimen were obtained over four different backgrounds (black, white, A1, and A3) interposed with an experimental translucent resin cement using a calibrated spectrophotometer. The color change (CIEDE2000) values, as well as all the CIELab values for each one of the experimental groups, were submitted to 28 different regression models. Each regression model was adjusted according to the weights of each dependent variable to achieve the best-fitting model.

RESULTS: Different substrates, ceramic shades, and thicknesses influenced the L, a, and b of the final restoration. Of all variables, the substrate influenced the final ceramic shade most, followed by the ceramic thickness and the L, a, and b of the ceramic. The decision tree regression model had the lowest mean absolute error and highest accuracy to predict the shade of the ceramic restoration according to the substrate shade, ceramic shade and thickness.

CLINICAL SIGNIFICANCE: The machine learning regression model developed in the study can help clinicians predict the final color of the ceramic veneers made with leucite-reinforced glass CAD/CAM ceramic HT and LT when cemented with translucent cements, based on the color of the substrate and ceramic thicknesses.