A Deep Learning Approach for Masseter Muscle Segmentation on Ultrasonography

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Biomedical engineering

Background:

• Deep learning algorithms are increasingly utilized in medical image processing.
• Accurate segmentation of anatomical structures is crucial for diagnosis and treatment planning.
• Masseter muscle ultrasonography is valuable for assessing various conditions.

Observation:

• A dataset of 388 adult masseter muscle ultrasound images was curated.
• Images were manually segmented using the CranioCatch labeling program and verified by experts.
• A PyTorch U-Net deep learning model was developed for masseter muscle segmentation.

Findings:

• The U-Net artificial intelligence model achieved perfect detection and segmentation of masseter muscles in all test images.
• The model demonstrated F1, sensitivity, and precision scores of 1.0 for image estimation.
• The AI model successfully automated the segmentation process.

Implications:

• Artificial intelligence offers a promising tool for automated masseter muscle segmentation on ultrasound.
• This technology can significantly reduce diagnostic time for surgeons and radiologists.
• The findings support the integration of AI into routine clinical workflows for improved efficiency.