Predicting enamel depth distribution of maxillary teeth based on intraoral scanning: A machine learning study

Area of Science:

• Biomedical Engineering
• Dental Materials Science
• Artificial Intelligence in Healthcare

Background:

• Accurate enamel depth distribution (EDD) is crucial for dental procedures like preoperative design, restorative previews, and monitoring enamel wear.
• Current methods for obtaining EDD lack non-invasive and efficient capabilities.
• Intraoral scanning (IOS) offers a promising non-invasive imaging modality for dental diagnostics.

Purpose of the Study:

• To develop a machine learning (ML) framework for non-invasive and radiation-free prediction of enamel depth distribution (EDD).
• To utilize intraoral scanning (IOS) images as input for the ML model to estimate EDD.
• To provide a novel method for preoperative design, aesthetic preview, and enamel wear monitoring.

Main Methods:

• A machine learning framework was developed using cone-beam computed tomography (CBCT) and IOS images from 200 volunteers.
• Five-dimensional features were extracted from labial enamel surfaces and used to train an eXtreme gradient boosting (XGB) model.
• The model's accuracy was evaluated using R-squared and mean absolute error (MAE), with predictions compared against CBCT-derived ground truths.

Main Results:

• The XGB model achieved high training accuracy with an average R-squared of 0.926 and MAE of 0.080.
• Independent validation demonstrated robust EDD prediction capabilities with no significant deviation from ground truths.
• Low prediction errors (Frobenius norm: 12.566-18.312) were observed, indicating reliable performance despite minor IOS noise.

Conclusions:

• Preliminary validation confirms the efficacy of an IOS-based ML model for high-quality EDD prediction.
• This non-invasive, radiation-free approach shows potential for improving dental diagnostics and treatment planning.
• Further research can explore broader clinical applications of this ML framework in dentistry.