Bi-Graph Reasoning for Masticatory Muscle Segmentation From Cone-Beam Computed Tomography.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Automated segmentation of masticatory muscles is difficult due to soft tissue ambiguity and artifacts in low-radiation cone-beam computed tomography (CBCT) images.
• Accurate segmentation is crucial for diagnosing and treating various conditions affecting the jaw and chewing function.

Purpose of the Study:

• To propose a novel bi-graph reasoning model (BGR) for simultaneous detection and segmentation of multi-category masticatory muscles from CBCT images.
• To address challenges posed by ambiguous soft tissue attachments and image artifacts in automated muscle segmentation.

Main Methods:

• Developed a bi-graph reasoning model (BGR) integrating category and region graphs.
• The category graph encodes learnable muscle prior knowledge, handling high-level dependencies and enhancing feature representation.
• The region graph models local and global dependencies of candidate muscle regions, improving robustness to artifacts.

Main Results:

• The BGR model effectively segments masticatory muscles from clinically acquired CBCTs.
• Achieved state-of-the-art accuracy in masticatory muscle segmentation.
• Demonstrated improved feature representation in the presence of entangled soft tissue and image artifacts.

Conclusions:

• The proposed bi-graph reasoning model (BGR) offers a robust and accurate solution for automated masticatory muscle segmentation in CBCT images.
• BGR effectively handles complex anatomical variations and image quality issues, paving the way for improved clinical diagnostics.