Classification of occlusal caries severity using spectrophotometric CIELAB measurements and machine learning...

Area of Science:

• Dental diagnostics
• Biophotonics
• Machine learning in healthcare

Background:

• Dental caries assessment relies on visual inspection, which can be subjective.
• Standardized, objective methods are needed to improve caries diagnosis accuracy.
• Spectrophotometry offers a quantitative approach to tooth color analysis.

Purpose of the Study:

• To evaluate spectrophotometric CIELAB color measurements for machine-learning (ML) classification of occlusal caries severity.
• To compare the performance of different ML models in classifying caries based on color data.
• To assess the potential of color-based methods as adjunctive tools for caries assessment.

Main Methods:

• Three hundred extracted teeth were visually classified using the International Caries Detection and Assessment System (ICDAS).
• Spectrophotometric CIELAB color measurements were taken at five occlusal sites per tooth.
• CIELAB values were transformed into features, and five ML models (Deep Sets, MLP, RF, XGBoost, CatBoost) were trained for caries classification.

Main Results:

• The Deep Sets architecture achieved the highest multiclass performance (Balanced Accuracy = 0.89).
• MLP showed high sensitivity (100%) for operative lesion detection (ICDAS 3-4).
• Neural network models demonstrated greater benefit from increased training data compared to tree-based models.

Conclusions:

• Spectrophotometric CIELAB measurements can support ML classification of occlusal caries severity under controlled in-vitro conditions.
• The Deep Sets model shows promise for objective, image-independent caries assessment.
• Color-based methods may serve as standardized, reproducible adjuncts to visual caries evaluation.