Automatic Masseter Muscle Accurate Segmentation from CBCT Using Deep Learning-Based Model.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Oral and maxillofacial surgery

Background:

• Accurate masseter muscle (MM) segmentation on cone-beam computed tomography (CBCT) is difficult due to poor soft-tissue contrast.
• Manual segmentation is time-consuming and labor-intensive, limiting clinical efficiency.

Purpose of the Study:

• To develop and validate a deep learning-based automatic approach for precise MM segmentation on CBCT.
• To leverage high-quality paired computed tomography (CT) for refining CBCT segmentation accuracy.

Main Methods:

• A 3D U-shape network was designed for MM segmentation.
• Manual annotations on CT served as the ground truth.
• Paired CBCT and CT datasets (n=42) and independent CBCT datasets (n=50) were utilized.
• Auto-segmentation results were clinically evaluated by an expert.

Main Results:

• The deep learning model achieved accurate MM segmentation on CBCT, comparable to CT results.
• The automatic approach significantly improved similarity to the ground truth compared to manual CBCT segmentation.
• The automated method was over 332 times faster than manual segmentation, requiring only 0.52% manual revision.

Conclusions:

• The proposed automatic model accurately segments MM on both CBCT and CT simultaneously.
• This approach enhances clinical efficiency and efficacy in oral and maxillofacial applications.
• Provides critical data for personalized treatment planning and long-term patient follow-up.