Can Deep Learning Methods Differentiate Temporomandibular Joint Disorders From Healthy Joints? A 3D Artificial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral and Maxillofacial Surgery

Background:

• Interpreting bone-related temporomandibular joint disorders (TMD) on CBCT scans is challenging.
• Advanced AI models are crucial for accurate TMD diagnosis and subcategorization.

Purpose of the Study:

• Develop deep learning models for comprehensive temporomandibular joint evaluation on CBCT scans.
• Automate segmentation, classification, and grading of TMD on CBCT images.

Main Methods:

• nnU-Net v2 for mandibular condyle segmentation.
• 3D CNN for classifying healthy vs. pathological condyles and differentiating five TMD types.
• Exploratory nnU-Net v2 for grading erosion, osteophyte formation, and sclerosis severity.

Main Results:

• Segmentation model achieved 0.87 DSC and 0.77 IoU.
• Classification F1-scores ranged from 0.65 (healthy vs. TMD) to 0.88 (flattening, sclerosis).
• Highest performance in grading was observed for Grade 1 erosion, Grade 2 osteophyte formation, and Grade 3 sclerosis.

Conclusions:

• Deep learning shows clinical relevance in condyle segmentation and TMD differentiation.
• Larger, balanced datasets are required to enhance TMD grading accuracy.