The Generalized Log-Ratio Transformation: Learning Shape and Adjacency Priors for Simultaneous Thigh Muscle...

Area of Science:

• Medical Imaging
• Computer Vision
• Anatomical Modeling

Background:

• Thigh muscle segmentation is crucial for understanding chronic obstructive pulmonary disease (COPD)-related muscle weakness.
• Existing methods struggle with indistinct muscle features and boundaries.
• Accurate segmentation aids in assessing disease impact on lower limb skeletal muscles.

Purpose of the Study:

• Introduce a novel probabilistic shape representation, the generalized log-ratio (GLR) representation.
• Develop an automated technique for accurate thigh muscle segmentation.
• Improve understanding of COPD effects through precise anatomical analysis.

Main Methods:

• Utilized the generalized log-ratio (GLR) representation for shape modeling.
• Employed a presegmentation-based alignment to remove pose variability.
• Developed a rotationally invariant random forest boundary detector for inter-muscular interfaces.
• Integrated shape model and boundary detector for a globally optimal, automatic segmentation technique.

Main Results:

• The GLR representation significantly improved thigh muscle segmentation accuracy.
• Achieved an average Dice similarity coefficient of 0.808 ±0.074.
• Segmentation produced probabilistic outputs, enabling uncertainty quantification for further analysis.

Conclusions:

• The GLR representation is effective for complex anatomical segmentation tasks.
• The proposed method offers a robust and accurate solution for thigh muscle segmentation.
• This technique facilitates better assessment of skeletal muscle conditions like COPD.