Deep Learning Enhances Failure Mode Analysis in Dental Adhesion Studies.

Area of Science:

• Biomaterials Science
• Dental Materials
• Artificial Intelligence in Dentistry

Background:

• Evaluating dental adhesive performance relies on shear bond strength testing.
• Failure mode analysis of tested samples is crucial but traditionally subjective using optical microscopy.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) for automated failure mode classification.
• To improve objectivity and reduce operator variability in analyzing dental adhesive failures.

Main Methods:

• A CNN was trained on optical microscopy images of 434 fractured tooth-adhesive interfaces.
• Ground truth for training was established using focus variation profilometry, validated by scanning electron microscopy.
• CNN performance was compared against expert human observers using the weighted F1 score.

Main Results:

• The CNN achieved a weighted F1 score of 0.893 ± 0.032, comparable to expert observers (0.824–0.911).
• The automated system demonstrated expert-level performance in classifying failure modes.
• Misclassifications primarily occurred at the threshold between mixed and predominantly adhesive failures.

Conclusions:

• A CNN trained on profilometry-derived data can objectively classify failure modes from optical images.
• This automated method offers a scalable and reliable alternative to subjective visual inspection.
• The study advances standardization in failure mode analysis for dental adhesion research.