A Fully Automated 3D CT U-Net Framework for Segmentation and Measurement of the Masseter Muscle, Innovatively...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomedical Engineering

Background:

• Accurate masseter muscle segmentation is crucial for radiological evaluation.
• Manual segmentation is time-consuming and inefficient.
• Deep learning offers potential for automated segmentation.

Purpose of the Study:

• To develop and evaluate a U-Net-based framework for automated masseter muscle segmentation and volume measurement.
• To provide baseline data on masseter muscle characteristics in a large East Asian cohort.
• To introduce a self-supervised algorithm to minimize deep learning sample size requirements.

Main Methods:

• Utilized a database of 840 individuals with head CT scans.
• Developed a U-Net-based coarse-to-fine learning framework for automated segmentation.
• Compared automated segmentation with manual delineation for volume, morphology, and runtime.
• Evaluated masseter asymmetry and correlated volumetric measurements with clinical parameters.

Main Results:

• The U-Net algorithm achieved volume accuracy equivalent to manual segmentation (P > 0.05).
• Automated segmentation runtime was significantly faster (<1 second) compared to manual segmentation (937.3 ± 95.9 seconds).
• Mean masseter asymmetry was 4.6% ± 4.6% in the cohort.

Conclusions:

• The U-Net-based algorithm provides reliable and efficient masseter muscle segmentation and volume measurement.
• This automated tool demonstrates high concordance with manual segmentation for CT-based assessments.
• The study establishes baseline masseter muscle data for healthy East Asian populations.