One-dimensional modeling of fractional flow reserve in coronary artery disease: Uncertainty quantification and Bayesian optimization

Non-invasive estimation of fractional flow reserve (FFR) values, the key index in the diagnosis of obstructive coronary artery disease, is a promising alternative to traditional way of performing invasive coronary angiography. With the advances in computational fluid dynamics (CFD), one can estimate FFR based on the solution obtained in a reconstructed coronary geometry from coronary computed tomography (CT) angiography. However, the computational cost to perform three-dimensional (3D) simulations has limited the use of CFD in most clinical settings. This could become more restrictive if one aims to quantify the uncertainty associated with FFR calculations due to the uncertainty in anatomic and physiologic properties as a significant number of 3D simulations is required to sample a relatively large parametric space. We have developed a predictive probabilistic model of FFR, which quantifies the uncertainty of the predicted values with significantly lower computational costs. Based on global sensitivity analysis, we first identify the important physiologic and anatomic parameters that impact the predictions of FFR. Our approach is to employ one-dimensional blood flow simulations of coronary trees that offer fast FFR predictions with uncertainty quantification in computing blood pressure and flow distributions within the coronaries. This is complemented with a multifidelity algorithm that is used to infer their optimal values using available patient-specific clinical measurements. (C) 2019 Elsevier B.V. All rights reserved.