Interpretable AI Explores Effective Components of CAD/CAM Resin Composites.

Area of Science:

• Materials Science
• Biomaterials Engineering
• Data Science

Background:

• High flexural strength is crucial for computer-aided manufacturing resin composite blocks (CAD/CAM RCBs) in clinical applications.
• Traditional trial-and-error methods for optimizing material composition are inefficient for determining flexural strength contributors.
• Machine learning (ML) offers a powerful approach to predict material properties and identify key components.

Purpose of the Study:

• To develop machine learning models for predicting the flexural strength of CAD/CAM RCBs.
• To identify the specific material components that significantly influence the flexural strength of these blocks.

Main Methods:

• Collected composition and flexural strength data from 12 commercial CAD/CAM RCBs.
• Augmented dataset size to 120 samples using a fluctuation range of 0.1 for multidimensional vectors.
• Developed five ML regression models (Random Forest, Extra Trees, Gradient Boosting, LightGBM, XGBoost) to predict flexural strength.
• Utilized exhaustive search and feature importance analysis to identify critical components.

Main Results:

• ML models achieved high predictive accuracy with R-squared values ranging from 0.927 to 0.998.
• All models demonstrated relative errors within 15%.
• Urethane dimethacrylate was consistently present in high flexural strength compositions.
• Filler content and triethylene glycol dimethacrylate were identified as the top two influential features.
• ZrSiO4 was the third most important feature for most models.

Conclusions:

• Established ML models effectively predict CAD/CAM RCB flexural strength.
• Successfully identified key components, including urethane dimethacrylate, filler content, triethylene glycol dimethacrylate, and ZrSiO4, that impact flexural strength.
• This ML-driven approach provides an efficient alternative to traditional methods for material optimization.