Deep learning-based approach to third molar impaction analysis with clinical classifications.

Area of Science:

• Dentistry and Artificial Intelligence
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Manual classification of impacted third molars is complex and time-consuming.
• Standardized classification systems like Pell and Gregory, Winter's, and Pederson Difficulty Index are crucial for treatment planning.
• Automating this process can improve efficiency and reduce diagnostic variability.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated detection and classification of impacted third molars.
• To utilize established classification systems (Pell and Gregory, Winter's, Pederson) within the model.
• To assess the model's performance against manual classifications.

Main Methods:

• A dataset of 2,300 panoramic radiographs was used for training the YOLOv11 model, with additional validation and testing sets.
• Impacted teeth were manually annotated using bounding boxes via CVAT software.
• The YOLOv11 model was trained with optimized hyperparameters and data augmentation techniques.

Main Results:

• The deep learning model achieved high performance metrics: precision (0.980), recall (0.948), F1 score (0.974), mAP@50 (0.990), and mAP@50:95 (0.974).
• The model demonstrated high accuracy in detecting and classifying impacted third molars.
• Some specific classifications showed lower F1 scores, indicating areas for potential improvement.

Conclusions:

• The developed deep learning model offers a reliable and efficient automated solution for impacted third molar classification.
• This AI tool can serve as a valuable decision support system in clinical dentistry, streamlining workflows.
• Further improvements can be achieved by enhancing dataset diversity and refining the model's handling of challenging classifications.