Using parametric model order reduction for inverse analysis of large nonlinear cardiac simulations.

Area of Science:

• Computational mechanics
• Biomedical engineering
• Cardiovascular modeling

Background:

• High-fidelity finite element simulations of cardiac mechanics are computationally demanding.
• Coupled cardiac and hemodynamic models require extensive computational resources, limiting clinical application and model calibration.
• Solving linear systems constitutes 90% of computation time in nonlinear cardiac simulations.

Purpose of the Study:

• To reduce the computational cost of patient-specific cardiac simulations.
• To accelerate model calibration and enable clinical use of complex cardiac models.
• To integrate model order reduction into optimization frameworks for inverse analysis.

Main Methods:

• Projection-based model order reduction applied to a coupled structure-Windkessel system.
• Proper orthogonal decomposition (POD) used for generating a reduced-dimensional subspace from displacement snapshots.
• Integration of the reduced-order model into gradient-based optimization for multivariate inverse analysis.

Main Results:

• Achieved considerable speedups in simulations using very few reduced degrees of freedom.
• Obtained good approximations of displacement fields and key cardiac outputs.
• Demonstrated successful application in a real-world multivariate inverse analysis scenario.

Conclusions:

• Projection-based model order reduction significantly speeds up cardiac model personalization.
• The method facilitates many-query tasks essential for clinical settings.
• This approach enhances the clinical applicability of predictive cardiac simulations.