Automated Tuning of Cardiovascular Boundary Conditions via Differentiable Surrogate Modeling

Area of Science:

• Computational fluid dynamics
• Biomedical engineering
• Cardiovascular modeling

Background:

• Cardiovascular simulations are crucial for understanding blood flow dynamics.
• Current methods face computational bottlenecks limiting clinical application.
• Need for efficient and accurate simulation tools.

Purpose of the Study:

• To develop an end-to-end differentiable framework for cardiovascular simulations.
• To enable hemodynamic surrogate modeling, automated model calibration, and stochastic parameter tuning.
• To address limitations in the clinical adoption of cardiovascular simulations.

Main Methods:

• Introduction of a hybrid mechanistic and data-driven reduced-order model (ROM).
• Nonlinear parametrization of lumped parameter networks for vascular domains.
• Exploitation of native differentiability for parameter calibration against 3D CFD simulations.

Main Results:

• The framework achieves efficient surrogate modeling capabilities.
• Automated calibration of ROM parameters against high-fidelity CFD data.
• Optimized ROM enables gradient-based deterministic and stochastic boundary condition calibration.

Conclusions:

• The developed framework offers computational efficiency and high fidelity.
• Directly addresses critical bottlenecks in clinical cardiovascular simulations.
• Facilitates wider adoption of patient-specific hemodynamic modeling.