Deep Learning for Predicting the Difficulty Level of Removing the Impacted Mandibular Third Molar

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Preoperative assessment of impacted mandibular third molars (LM3) is crucial for surgical planning.
• Panoramic radiography is a standard imaging technique for evaluating LM3.
• Predicting surgical difficulty aids in optimizing patient care and resource allocation.

Purpose of the Study:

• To develop and evaluate a computer-aided deep learning (DL) system for predicting the surgical removal difficulty of impacted LM3.
• To utilize panoramic radiographs as the primary input for the DL system.
• To enhance preoperative surgical planning for impacted LM3 extraction.

Main Methods:

• A retrospective study of 1367 LM3 images from 784 patients (2021-2023).
• Development of a 3-phase DL system integrating ResNet101V2, RetinaNet, and Vision Transformer models.
• ResNet101V2 for impaction classification, RetinaNet for precise localization, and Vision Transformer for difficulty assessment.

Main Results:

• ResNet101V2 achieved 0.8671 classification accuracy for identifying impacted LM3.
• RetinaNet demonstrated a 0.9928 mean average precision for LM3 detection.
• Vision Transformer attained 0.7899 average accuracy in predicting surgical difficulty levels.

Conclusions:

• A 3-phase computer-aided DL system shows strong performance in predicting impacted LM3 surgical removal difficulty.
• The developed system effectively uses panoramic radiographs for preoperative assessment.
• This AI-driven approach offers a promising tool for improving surgical planning in dentistry.