Automatic multi-class classification of 3D relationships between the mandibular third molar and canal on cone-beam...

Area of Science:

• Oral and Maxillofacial Surgery
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate assessment of the spatial relationship between the mandibular third molar (M3) and the mandibular canal (MC) is crucial for preventing inferior alveolar nerve injury during M3 extraction.
• Cone-beam computed tomography (CBCT) is commonly used for 3D evaluation, but manual analysis can be time-consuming and prone to errors.

Purpose of the Study:

• To develop and validate a geometry-aware deep learning framework for automated classification of M3-MC spatial relationships.
• To categorize these relationships into five distinct types based on anatomical contact, involvement, and spatial proximity using CBCT images.

Main Methods:

• A two-stage deep learning framework was employed: modified mAttUNet for M3 and MC segmentation, and DenseAttNet for multi-class classification.
• The framework incorporated attention mechanisms and signed distance map (SDM) inputs to capture geometric and anatomical features for improved classification accuracy.

Main Results:

• The mAttUNet achieved high segmentation performance (0.96 for MC, 0.84 for M3).
• The DenseAttNet demonstrated superior classification performance with an overall AUC of 0.97, accurately distinguishing all five spatial relationship types.

Conclusions:

• The automated classification framework provides accurate and reliable assessment of M3-MC spatial relationships.
• This technology offers significant clinical utility for enhanced risk evaluation and optimized surgical planning in M3 extraction, potentially reducing nerve injury complications.