Machine Learning for Micro-Tensile Bond Strength Classification of Dental Adhesives

Objectives: Application of machine learning to micro-tensile bond strength classification of commercial dental adhesives.
Methods: 176 Micro-tensile bond strength values (µTBS) of 84 commercial dental adhesives were collected from the literature based on the following criteria: 1) bonding substrate was sound dentin of permanent human teeth, 2) the manufacturer’s instructions were followed during the bonding process, and 3) the specimens were stored in water at 37°C for 24-h before the measurement. Meanwhile, chemical composition and pH-values were collected from the manufacturers. The original dataset was constructed using nine input features, including eight chemical ingredients (HEMA, Bis-GMA, UDMA, MDP, PENTA, filler, fluoride, and organic solvent) and the effective etching pH-value (pH=0.7 for total-etching), and one output feature (µTBS). Next, the µTBS feature of each adhesive (different etching modes of universal were counted differently) was labeled as either 0 (if <35 MPa) or 1 (if ≥35 MPa) based on their average collected values, resulting in 39 of class 0 and 45 of class 1. Then ten different machine learning (ML) classifiers were used to analyze the binary-µTBS dataset. Grid search cross-validation was used for hyperparameter tuning and model optimization. The optimal classification accuracies were compared among the ten classifiers. The most important features were selected through feature analysis. A reduced dataset was reconstructed using the selected features and classification accuracy was re-calculated for the ten classifiers.
Results: The classification accuracy ranged from 76.4% to 86.9% for the ten classifiers. Feature analyses selected four important features: pH, HEMA, MDP and filler. Feature-reduced dataset showed similar accuracy for all classifiers.
Conclusions: Machine learning is an effective tool for classifying dental adhesives with low and high µTBS. ML can also identify important chemical ingredients that affect the µTBS of dental adhesives. ML has great potential to play an important role in dental biomaterial development and characterization.