Automated segmentation for patella from lateral knee X-ray images

Area of Science:

• Medical Image Analysis
• Biomedical Engineering
• Radiology

Background:

• X-ray image segmentation, particularly for organs like the patella, is challenging due to uneven X-ray absorption, noise, and complex anatomical structures.
• Accurate segmentation is crucial for clinical evaluation and biomechanical studies of the knee.

Purpose of the Study:

• To develop a novel, automated method for segmenting the patella from lateral knee X-ray images.
• To improve the accuracy and robustness of patella segmentation in medical imaging.

Main Methods:

• A patella shape model was constructed using principal component analysis (PCA) on training data.
• An edge-tracing strategy was employed for initial shape model placement near the patella boundary.
• A dual-optimization approach, combining a genetic algorithm (GA) for global transform and active shape model (ASM) for iterative deformation, was used for fitting the model.

Main Results:

• The proposed method successfully segmented the patella in tested lateral knee X-ray images.
• Promising results were obtained on a dataset of 20 images, demonstrating the method's effectiveness.
• The automated procedure showed capability in handling variations in knee X-ray images.

Conclusions:

• The developed active shape model-based method provides an effective and automated solution for patella segmentation in lateral knee X-rays.
• This technique shows potential utility for clinical evaluation and biomechanical research involving the knee.
• The combination of PCA, edge tracing, GA, and ASM offers a robust approach to medical image segmentation.