Global Sensitivity Analysis for Patient-Specific Aortic Simulations: The Role of Geometry, Boundary Condition and Large Eddy Simulation Modeling Parameters

Abstract

Uncertainties affect the reliability of the numerical simulation of hemodynamics in patient-specific settings and rigorous Uncertainty Quantification (UQ) is in order. This work presents a UQ study on the aorta flow, for assessing the sensitivity of the clinical relevant quantities to the morphology and imprecise knowledge of the inflow boundary condition using the Polynomial Chaos Expansion based Sobol' indices. The geometrical uncertainty is modeled based on a set of longitudinal imaging data of a patient with the abdominal aortic aneurysm. The images of the patient's aorta at different stages of the disease are used to create a map that drives the realistic variation of the reconstructed morphology. The aorta is a peculiar site for hemodynamics, since the flow is highly disturbed due to the high Reynolds number during systole, and the modeling of turbulence helps to avoid the high computational costs. The deconvolution-based Leray model was considered in the past for these simulations. The LES model features problem-dependent numerical parameters to tune. We borrow the same UQ tools used for physical uncertain quantities to assess the sensitivity of the simulations to one of these numerical parameters, the filter radius. The sensitivity of the total kinetic energy, the time average wall shear stress, and the oscillatory shear index are analyzed. The results show that the geometry has the most dominant contribution to the uncertainty of all the quantities of interest. The sensitivity analysis provides confidence intervals for the simulations that quantify the reliability of the numerical predictions.

Aneurysmal Growth in Type-B Aortic Dissection: Assessing the Impact of Patient-Specific Inlet Conditions on Key Haemodynamic Indices

“…TL pressure dominates FL pressure up to the first re-entry tear, where FL pressure begins to dominate. This trend is observed in other computational studies 43 . FL pressurisation is not directly correlated with FL growth in this patient, with growing regions all showing different TMP characteristics; TMP is strongly positive near α , near-zero at β and negative from γ onward.…”

Medical Image-Based Computational Fluid Dynamics and Fluid-Structure Interaction Analysis in Vascular Diseases

“…It uses a stochastic approach to obtain continuous response surfaces of the hemodynamic parameters starting from a few deterministic simulations and is computationally more efficient than a Monte Carlo approach. Furthermore, using the method of polynomial chaos expansion for a global sensitivity analysis, it has been found that the sensitivity to geometry may be different during different instants of the heartbeat and in different vascular regions ( Xu et al, 2021 ).…”

Turbulence Descriptors in Arterial Flows : Patient-Specific Computational Hemodynamics

“…Considering the computational expensive CFD strategies used for computing phaseaveraged solutions within this thesis (including the large amount of manual labor in both preand post-processing steps), the future focus should be on identifying cost reductions towards the most effective or "bang for the buck" solution strategies. In a newly published study [258], a variance-based sensitivity method was utilized to attribute the uncertainty of the outputs to specific modeling inputs, which required significantly less number of samples than standard UQ methods. This UQ framework used LES on patient-specific aortic aneurysm models and an idealized aortic arc benchmark to assess the output sensitivity to uncertainties in geometry and inflow variations, and modeled subgrid dissipation strength.…”