An automated statistical shape model developmental pipeline: application to the human scapula and humerus

Area of Science:

• Medical image analysis
• Computational anatomy
• Biomedical engineering

Background:

• Statistical shape models (SSMs) are crucial for analyzing anatomical variations.
• Existing methods often require manual intervention (landmarks) or are biased by reference selection.
• Improving the robustness and automation of SSM creation is essential for clinical applications.

Purpose of the Study:

• To develop an automated pipeline for creating robust statistical shape models.
• To enhance the domain of shape variability using probabilistic principal component analysis (PPCA).
• To compare the proposed method against existing algorithms using established metrics.

Main Methods:

• Utilized a pipeline combining robust rigid-groupwise registration and pointset nonrigid registration.
• Implemented an automated approach, eliminating the need for manual landmarks or regionalization.
• Employed the probabilistic principal component analysis (PPCA) framework for shape variability analysis.

Main Results:

• The proposed method demonstrated comparable results to existing algorithms in terms of compactness.
• Achieved superior generality and specificity curves compared to expectation maximization-iterative closest point (EM-ICP) and another groupwise method.
• Successfully applied the method to complex anatomical structures: the human scapula and humerus.

Conclusions:

• The developed automated pipeline offers an unbiased and robust approach to statistical shape modeling.
• The method enhances the representation of shape variability, outperforming existing techniques in key metrics.
• This technique holds promise for the analysis of complex anatomical structures in medical imaging.