Anatomically Parameterized Statistical Shape Model: Explaining Morphometry Through Statistical Learning.

Area of Science:

• Medical imaging and biomechanics
• Computational anatomy
• Statistical modeling

Background:

• Statistical Shape Models (SSMs) are vital for anatomical analysis in clinical settings.
• Current SSMs lack direct links between shape coefficients and clinically relevant anatomical measures, limiting their interpretability.
• This gap hinders the subjective evaluation and application of SSMs in practice.

Purpose of the Study:

• To develop a novel Statistical Shape Model (SSM) controlled by anatomical parameters derived from morphometric analysis.
• To establish a clear, one-to-one relationship between shape coefficients and clinically relevant anatomical measures.
• To enhance the clinical interpretability and utility of SSMs.

Main Methods:

• A linear mapping was learned between shape coefficients (latent space) and selected anatomical parameters (anatomical space) using a synthetic population.
• The anatomically parameterized SSM (ANAT-SSM) was constructed by determining the pseudo-inverse of the learned mapping.
• Orthogonality constraints were applied (OC-ANAT-SSM) to ensure independent shape variation patterns, evaluated on femoral and scapular bone datasets.

Main Results:

• Generated synthetic shapes demonstrated realistic anatomical statistics.
• The learned mapping matrices accurately reflected the statistical relationships between shape and anatomical parameters.
• The developed SSMs achieved moderate to excellent accuracy in predicting anatomical parameters for novel shapes.

Conclusions:

• The proposed method successfully creates anatomically parameterized SSMs, overcoming the interpretability limitations of standard SSMs.
• This approach enables objective analysis of bone morphometry and supports clinical applications like pre-surgery planning.