Machine Learning-Based Tooth Color Assessment Using Color Moments to Prevent Metamerism in Different Clinical Lights

Abstract

Choosing the right shade in prosthodontic treatment is of great importance in terms of achieving a natural aesthetic appearance and increasing the patient's satisfaction with the treatment. However, this process is affected by many technical and environmental factors. In particular, variable light sources in clinical and laboratory environments cause the problem of metamerism, which leads to misleading results in color perception. This study proposes a method that reduces the effect of metamerism by detecting color under different light conditions, eliminates the subjectivity of traditional color matching methods and offers an alternative to costly measurement devices. The 29 color samples from the Vita 3D Master shade guide were imaged five times each in four different clinical light conditions. Feature extraction was performed using color moments in RGB, LAB and HSV color spaces. Experimental studies were carried out with different machine learning algorithms on the datasets created with these data. As a result, 100% accuracy was obtained for the classification of four clinical light conditions, 85% for the light-independent classification of 29 Vita colors, 100% under white light, 97% under natural light, 92% under flash light and 94% under yellow light. These findings demonstrated that the limitations of traditional or costly color selection processes can be overcome and metamerism can be reduced by machine learning techniques.