Cascaded statistical shape model based segmentation of the full lower limb in CT

Area of Science:

• Orthopedic research
• Biomechanical analysis
• Medical imaging

Background:

• Image segmentation is crucial for orthopedic and biomechanical research but is often time-consuming.
• Manual segmentation of complex anatomical structures like the lower limb is laborious and prone to variability.

Purpose of the Study:

• To develop a fully automatic, model-based segmentation pipeline for the entire lower limb in computed tomography (CT) images.
• To enable the generation of accurate surface meshes for anatomical and mechanical simulations.

Main Methods:

• A novel pipeline combining prior shape model fitting with gradient-defined free-form deformation.
• Continuous model refinement using an iterative approach with newly segmented training samples.
• Validation through comparison of automatic segmentations against manual segmentations.

Main Results:

• The pipeline achieved convergence in 250 out of 253 cases, with average distance errors between 0.53–0.76 mm.
• Segmentation accuracy improved with more training data, plateauing with 20 samples for free-form deformation optimization.
• Maximum segmentation error stabilized after 100 training samples.

Conclusions:

• The proposed automatic segmentation pipeline offers an efficient and accurate solution for lower limb CT image analysis.
• The method facilitates the creation of surface meshes essential for advanced biomechanical and anatomical simulations.
• Continuous model updating enhances segmentation accuracy and robustness over time.