Multimodal deep learning with anatomically constrained attention for screening MRI-detectable TMJ abnormalities from...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthodontics

Background:

• Early diagnosis of temporomandibular disorders (TMDs) is difficult.
• Intra-articular temporomandibular joint (TMJ) abnormalities require magnetic resonance imaging (MRI) for confirmation.

Purpose of the Study:

• To develop a comprehensive screening method for MRI-detectable TMJ pathologies.
• To create an interpretable deep learning framework for TMJ abnormality detection.

Main Methods:

• Developed a deep learning framework using paired open- and closed-mouth TMJ panoramic radiographs and clinical metadata.
• Integrated anatomically guided attention, multimodal clinical features, and ensemble learning.
• Utilized Gradient-weighted Class Activation Mapping for visualization and interpretability.

Main Results:

• The best-performing ensemble framework achieved an area under the curve of 0.86.
• Demonstrated balanced classification of MRI-negative and -positive cases.
• Confirmed focus on condylar regions and showed added value of clinical metadata and spatial attention.

Conclusions:

• The prototype workflow can effectively triage patients for MRI referral.
• Supports early detection of TMJ abnormalities.
• Facilitates timely interventions for temporomandibular disorders.