Tooth color prediction in intraoral images under different clinical lights using ML algorithms and CLAHE technique: an In-Vivo study.

Abstract

Tooth color selection is a crucial step in prosthetic dental treatments. However, the process often suffers from subjectivity, environmental light variability, and the high cost or lack of standardization in instrumental methods. This study aims to develop a consistent and reliable tooth color prediction model using machine learning (ML) techniques, even under different clinical lights. In-vivo intraoral images of anterior teeth were collected from volunteer patients under five different clinical light sources. The teeth were annotated, segmented, and matched with corresponding color labels. LAB color space was used, and the CLAHE (Contrast Limited Adaptive Histogram Equalization) technique was applied to the L-channel to reduce light-induced glare. To address class imbalance, data augmentation was performed, resulting in a balanced dataset of 16,640 images. The dataset was split into 80% training and 20% testing subsets. Five ML algorithms-Random Forest (RF), K-Nearest Neighbors (KNN), Logistic Regression, Decision Tree, and XGBoost-were evaluated. As a result of experimental studies, RF and XGBoost obtained the highest performance, both achieving an accuracy of 97% in predicting tooth color across different lighting scenarios. These results demonstrate the robustness of the approach under variable lighting conditions. This study demonstrates that ML algorithms combined with image enhancement techniques such as CLAHE can provide accurate and light-independent tooth color predictions. The proposed method offers a practical and low-cost tool to support clinical decision-making in prosthetic dentistry, potentially enhancing standardization and efficiency in color matching processes.