Learning spatiotemporal statistical shape models for non-linear dynamic anatomies.

Area of Science:

• Medical imaging analysis
• Computational anatomy
• Biomedical engineering

Background:

• Understanding anatomical shape changes over time is crucial for clinical research, including disease progression and organ dynamics.
• Existing spatiotemporal statistical shape modeling (SSM) methods often fail with sequential data due to assumptions of sample independence or linearity.
• Previous adaptations for dynamic SSM struggle with within-subject correlations and non-linear shape dynamics.

Purpose of the Study:

• To develop a novel spatiotemporal SSM approach that accurately models population-level dynamic shape variations over time.
• To relax restrictive assumptions of previous methods, specifically sample independence and linearity of shape dynamics.
• To improve the quantification of temporal dependencies in anatomical shape data.

Main Methods:

• Proposed a principled approach to spatiotemporal SSM incorporating time dependency into correspondence optimization.
• Utilized regularized principal component polynomial regression to model non-linear temporal dynamics.
• Validated the method on synthetic data and cardiac MRI-derived left atrium shapes.

Main Results:

• The new method effectively captures population modes of shape variation over time.
• Demonstrated a statistically significant ability to model time dependency in shape data.
• Outperformed existing methods in analyzing dynamic shape changes, particularly for sequential observations.

Conclusions:

• The developed spatiotemporal SSM approach offers a flexible and robust framework for analyzing dynamic anatomical changes.
• This method accurately models non-linear temporal dynamics and within-subject correlations, advancing longitudinal shape analysis.
• The findings have significant implications for clinical studies involving disease progression and organ function over time.