Fast and robust parameter estimation with uncertainty quantification for the cardiac function.

Area of Science:

• Computational cardiology
• Biomedical engineering
• Digital twins

Background:

• Parameter estimation and uncertainty quantification are vital for creating patient-specific digital twins in computational cardiology.
• Fitting complex cardiac electromechanics and hemodynamics models requires robust methods using limited, noisy, non-invasive data.
• Clinical translation necessitates computationally efficient methods with low resource demands.

Purpose of the Study:

• To develop and validate a computationally efficient approach for parameter estimation and uncertainty quantification in cardiac modeling.
• To enable the creation of accurate digital twins for individual patients.
• To meet the demands of clinical application and Green Computing.

Main Methods:

• Bayesian inference combining Maximum a Posteriori estimation and Hamiltonian Monte Carlo.
• Utilizing an Artificial Neural Network surrogate model for fast, geometry-specific cardiac function simulation.
• Employing matrix-free methods, automatic differentiation, and automatic vectorization for computational efficiency.
• Accounting for surrogate model and measurement errors.

Main Results:

• Accurate estimation of all model parameters across diverse in silico test cases (ventricular function, cardiocirculatory system).
• Achieved highly accurate parameter estimations in short computational times using a single CPU on a standard laptop.
• Generated posterior distributions that successfully contain the true parameter values within 90% credibility regions.

Conclusions:

• The developed approach enables fast and robust identification of cardiovascular system model parameters with minimal hardware.
• The method is effective with limited non-invasive data and high signal-to-noise ratios.
• The approach aligns with clinical exploitation requirements and Green Computing principles.