Preinterventional Third-Molar Assessment Using Robust Machine Learning.

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Machine learning (ML) models, particularly deep neural networks, are increasingly utilized in medical image analysis and clinical decision support.
• Accurate assessment of mandibular third molar (M3M) removal risks, including proximity to the inferior alveolar nerve (IAN) and root development, is crucial for safe surgical procedures.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) system for dental decision-making in mandibular third molar (M3M) removal.
• To assess the risk of M3M removal by analyzing 2D orthopantomograms for M3M-IAN superimposition and M3M root development.

Main Methods:

• Utilized a dataset of 4,516 panoramic radiographic images for training a spatially dependent U-Net and a ResNet-101 model.
• Employed deep neural networks for classifying M3M-IAN superimposition and M3M root development.
• Evaluated model performance on a control set of 120 images using 10-fold cross-validation and Matthew's correlation coefficient.

Main Results:

• Achieved high accuracy values (0.94 and 0.93) for M3M-IAN superimposition and root development tasks during cross-validation.
• Demonstrated robust generalization on a control dataset with accuracies of 0.9 and 0.87, and Matthew's correlation coefficients of 0.82 and 0.75.
• Developed a diagnostic table to guide decisions on the need for additional 3D imaging based on AI predictions.

Conclusions:

• The developed AI system effectively supports clinical decision-making for mandibular third molar removal.
• The AI tool enhances efficiency and reduces risks, particularly for less experienced practitioners, by providing reliable risk assessments.
• Computer-aided decision-making tools show significant potential to improve dental surgical planning and patient outcomes.