Transient flow prediction in an idealized aneurysm geometry using data assimilation

Abstract

Hemodynamic simulations are restricted by modeling assumptions and uncertain initial and boundary conditions, whereas Phase-Contrast Magnetic Resonance Imaging (PC-MRI) data is affected by measurement noise and artifacts. To overcome the limitations of both techniques, the current study uses a Localization Ensemble Transform Kalman Filter (LETKF) to fully incorporate noisy, low-resolution Phase-Contrast MRI data into an ensemble of high-resolution numerical simulations. The analysis output provides an improved state estimate of the three-dimensional blood flow field in an intracranial aneurysm model. Benchmark measurements are carried out in a silicone phantom model of an idealized aneurysm under pulsatile inflow conditions. Validation is ensured with high-resolution Particle Imaging Velocimetry (PIV) obtained in the symmetry plane of the same geometry. Two data assimilation approaches are introduced, which differ in their way to propagate the ensemble members in time. In both cases the velocity noise is significantly reduced over the whole cardiac cycle. Quantitative and qualitative results indicate an improvement of the flow field prediction in comparison to the raw measurement data. Although biased measurement data reveal a systematic deviation from the truth, the LETKF is able to account for stochastically distributed errors. Through the implementation of the data assimilation step, physical constraints are introduced into the raw measurement data. The resulting, realistic high-resolution flow field can be readily used to assess further patient-specific parameters in addition to the velocity distribution, such as wall shear stress or pressure.

Keywords: 4D flow; Data assimilation; Ensemble Kalman Filter; Intracranial aneurysm; PC-MRI.