Personalized and uncertainty-aware coronary hemodynamics simulations: From Bayesian estimation to improved...

Area of Science:

• Cardiovascular Physiology
• Computational Fluid Dynamics
• Medical Imaging Analysis

Background:

• Non-invasive coronary hemodynamics simulations enhance risk stratification for coronary artery disease (CAD).
• Current simulation methods often use empirical flow distribution, neglecting patient-specific factors and data uncertainty.
• Accurate modeling requires incorporating individual variability, disease states, and clinical data uncertainties.

Purpose of the Study:

• To develop an end-to-end pipeline for personalized coronary flow simulations.
• To integrate vessel-specific flows and cardiac function, accounting for clinical data uncertainty.
• To enhance the precision of predicting clinical and biomechanical outcomes.

Main Methods:

• Assimilated patient-specific myocardial blood flow from CT perfusion imaging to estimate branch-specific coronary artery flows.
• Employed adaptive Markov Chain Monte Carlo sampling to estimate model parameters under simulated measurement noise.
• Utilized multi-fidelity Monte Carlo estimation with non-linear dimensionality reduction for posterior predictive distributions.

Main Results:

• The framework accurately reproduced cardiac function and branch-specific flows, accounting for measurement uncertainty.
• Observed significant reductions in confidence intervals compared to single- and multi-fidelity Monte Carlo methods.
• Achieved reduced computational cost for multi-fidelity Monte Carlo estimators while maintaining specified confidence levels.

Conclusions:

• The developed pipeline enables personalized, uncertainty-aware predictions of coronary hemodynamics using routine clinical data.
• Leverages advanced CT myocardial perfusion imaging techniques for improved accuracy.
• Offers substantial improvements in predictive precision and computational efficiency for clinical applications.