Spatiotemporal Cardiac Statistical Shape Modeling: A Data-Driven Approach.

Area of Science:

• Medical imaging
• Biomedical engineering
• Computational anatomy

Background:

• Spatiotemporal statistical shape modeling (SSM) is crucial for understanding anatomical changes in clinical studies.
• Current SSM methods often rely on cross-sectional data or predefined atlases, limiting their ability to capture longitudinal shape dynamics.
• Particle-based shape modeling (PSM) offers a data-driven approach but is typically limited to cross-sectional designs.

Purpose of the Study:

• To develop a novel, data-driven approach for population-level spatiotemporal SSM that learns directly from shape data.
• To create an SSM optimization scheme for establishing correspondences across subjects and time-series.
• To validate the method's efficacy in analyzing dynamic anatomical changes, specifically the left atrium in atrial fibrillation patients.

Main Methods:

• A novel SSM optimization scheme inspired by PSM was developed to learn spatiotemporal shape variations.
• The method establishes corresponding landmarks across subjects (inter-subject) and time-series (intra-subject).
• The approach was applied to 4D cardiac data from atrial fibrillation patients.

Main Results:

• The proposed method effectively represents the dynamic changes of the left atrium in atrial fibrillation patients.
• It outperforms a conventional image-based spatiotemporal SSM approach.
• Linear Dynamical System (LDS) models utilizing the proposed SSM showed improved generalization and specificity.

Conclusions:

• The developed data-driven spatiotemporal SSM method accurately captures population-level anatomical changes over time.
• This approach offers a powerful tool for characterizing organ dynamics and disease progression.
• The method provides a more robust and accurate analysis of time-dependent shape variations compared to existing techniques.