Multi-fidelity uncertainty-aware coronary hemodynamics personalized by CT myocardial perfusion imaging

This work presents a novel framework for personalized and uncertainty-aware coronary artery bypass graft (CABG) surgical planning that is informed by CT myocardial perfusion imaging. Current computational models of coronary hemodynamics distribute flow amongst coronary arteries based on purely empirical scaling between flow and the size of each artery. However, this does not account for anatomical inaccuracy, patient variability, disease, etc. Patient-specific models also do not incorporate uncertainty stemming from the clinical data they are based on. We demonstrate a framework for improved model personalization by estimating patient-specific and vessel-specific coronary flows from non-invasive myocardial perfusion imaging. We show significantly improved CABG predictions using these personalized models. We also develop an uncertainty-aware personalization framework that accounts for noisy clinical data. We use probabilistic Bayesian estimation of model parameters based on clinical data. We propagate clinical uncertainty into predicted quantities of interest using a novel multi-fidelity uncertainty quantification technique relying on non-linear dimensionality reduction. We demonstrate clinical data-informed confidence intervals with vastly improved precision in the estimated quantities of interest.