Deep Learning-Based Automatic Muscle Segmentation of the Thigh Using Lower Extremity CT Images.

Area of Science:

• Musculoskeletal research
• Metabolic research
• Medical imaging analysis

Background:

• Sarcopenia and muscle composition are key indicators in musculoskeletal and metabolic research.
• Accurate quantitative muscle assessment is crucial for understanding these conditions.

Purpose of the Study:

• To develop and validate an automated deep learning method for thigh muscle segmentation using non-contrast CT images.
• To quantitatively evaluate thigh muscles, including lean and fat volumes.

Main Methods:

• Three deep learning models (Dense U-Net, MANet, SegFormer) were implemented and compared for thigh muscle segmentation.
• Models were trained on 136 CT scans and validated on 40 external scans.
• Segmentation performance was assessed using Dice similarity coefficient (DSC), sensitivity, specificity, and accuracy.

Main Results:

• All models achieved high segmentation accuracy (mean DSC >96%).
• MANet excelled in internal validation; SegFormer showed superior volumetric agreement in external validation (ICC ≥0.995, p<0.01).
• Automated quantitative indices closely matched manual measurements.

Conclusions:

• A deep learning framework enables accurate, reproducible segmentation of functional thigh muscle groups from CT scans.
• Comparative analysis supports the feasibility of CT-based quantitative muscle assessment for sarcopenia and musculoskeletal research.