A machine learning based approach to standardizing tooth color and shade recommendations

Qijing Li; Du Chen; Hang Wang; Jiefei Shen

doi:10.1016/j.prosdent.2024.09.010

A machine learning based approach to standardizing tooth color and shade recommendations

J Prosthet Dent. 2024 Sep 30:S0022-3913(24)00639-5. doi: 10.1016/j.prosdent.2024.09.010. Online ahead of print.

Authors

Qijing Li¹, Du Chen², Hang Wang³, Jiefei Shen⁴

Affiliations

¹ Master's student, State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Prosthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, PR China.
² Doctoral student, State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Prosthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, PR China.
³ Professor, State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Prosthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, PR China.
⁴ Professor, State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Prosthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, PR China. Electronic address: shenjiefei@scu.edu.cn.

PMID: 39353755
DOI: 10.1016/j.prosdent.2024.09.010

Abstract

Statement of problem: Achieving the precise reproduction of tooth color is pivotal in esthetic restorations. However, existing visual and instrumental shade matching methods, each with inherent limitations, have often resulted in inconsistent color communication and subsequently suboptimal esthetic results in clinical practice.

Purpose: The purpose of this in vitro study was to evaluate a 2-tier color correction strategy to accurately acquire standardized tooth color and to provide shade recommendations for esthetic restorations.

Material and methods: Photographs of a standard color card (ColorChecker Classic; X-rite) and a commercially available shade guide (VITA 3D-Master; Vita Zahnfabrik) were captured under standard lighting conditions. Machine learning (ML) models for color correction, including polynomial regression (PR), backpropagation neural network (BPNN), and extreme learning machine (ELM), were trained using color values extracted from these standardized photographs. Subsequently, photographs made under clinical lighting conditions and featuring both the standard color card and the shade guide underwent the first color correction using the trained ML models. The secondary color correction was then executed based on the custom color space of VITA 3D-Master, yielding corrected color values for shade recommendations. The prediction accuracy of the ML models and the precision of color correction were evaluated using the root mean square error (RMSE), coefficient of determination (R²), and color difference (α=.05 for all statistical analyses).

Results: Compared with the PR and BPNN models, the ELM model provided more precise and reliable predictions with the lowest RMSE (2.2) and the highest R² (0.996). After 2 rounds of color correction, the color difference was reduced from 7.6 to 1.0, which was lower than the 50:50% acceptability threshold (1.8) and closer to the 50:50% perceptibility threshold (0.8). Furthermore, the matching results between the secondary values and the ground truth of the shade guides achieved an accuracy of 73.1% for shade recommendations.

Conclusions: The 2-tier color correction strategy based on the ML models and the color space specified by the VITA 3D-Master system effectively standardized the color of dental photographs and provided a more accurate and stable method of communicating color between dentists and dental laboratory technicians.