Among Artificial Intelligence/Machine Learning Methods, Automated Gradient-Boosting Models Accurately Score Intraoral...

Area of Science:

• Dentistry
• Machine Learning
• Image Analysis

Background:

• Previous automated dental plaque detection models struggled with non-standardized images.
• Accurate plaque scoring is crucial for preschooler dental prevention trials.

Purpose of the Study:

• Develop and validate automated methods for image selection and intraoral plaque scoring.
• Establish a reliable primary outcome measure for preschooler dental prevention trials.

Main Methods:

• Utilized 1650 plaque-disclosed primary teeth images from clinical trials.
• Employed machine learning (ML) algorithms, including Support Vector Machine-Gaussian and Gradient-Boosting, for classification and regression.
• Preprocessed images using Laplacian filters and extracted features like hue, saturation, and brightness.

Main Results:

• Achieved high performance with ML models: Support Vector Machine-Gaussian for image selection (AUC-ROC 0.99) and Gradient-Boosting for plaque scoring (AUC-ROC 0.99, R² 0.72).
• Demonstrated efficient training times, with image selection models training in under 2 seconds.

Conclusions:

• Accurate automated plaque scoring is achievable without expensive deep learning models.
• The developed automated system requires minimal user intervention, enhancing practicality.