Grid convergence errors in hemodynamic solution of patient-specific cerebral aneurysms

Mind the Gap: Impact of Computational Fluid Dynamics Solution Strategy on Prediction of Intracranial Aneurysm Hemodynamics and Rupture Status Indicators

“…The impact on hemodynamic indices was not addressed in that study, and for our HR simulations, the average element side length in the aneurysm sac was 0.06 mm, well below the 0.08 mm reported for the converged solutions of Hodis et al 8 The pronounced impact of the CFD solution strategy on the magnitude of derived hemodynamic indices implies that the solution strategy must be acknowledged as an additional source of variability, especially if and when the findings of different groups, potentially using a wide range of strategies, are compared or analyzed together. In striving for more patient-specific simulations, most studies have focused on the impact of other assumptions or uncertainties such as inflow/outflow conditions, 31 non-Newtonian rheology, 32 compliance, 33 and choice of imaging technique.…”

“…These "routine" convergence studies appear to fall significantly short of what is needed, as may be deduced from the aneurysm CFD convergence study reported by Hodis et al, 8 in which it was concluded that "the grid convergence errors showed oscillatory behavior; therefore, each patient-specific model required its own grid convergence study to establish the accuracy of the analysis." The impact on hemodynamic indices was not addressed in that study, and for our HR simulations, the average element side length in the aneurysm sac was 0.06 mm, well below the 0.08 mm reported for the converged solutions of Hodis et al 8 The pronounced impact of the CFD solution strategy on the magnitude of derived hemodynamic indices implies that the solution strategy must be acknowledged as an additional source of variability, especially if and when the findings of different groups, potentially using a wide range of strategies, are compared or analyzed together.…”

“…In principle, these are chosen in anticipation of the expected hemodynamics and then must be demonstrated to converge to within some desired error tolerance via methodical refinement studies. 8 In practice however, discretizations are usu-ally constrained by available computational resources and/or desired solution times, and rarely are proper convergence studies performed or reported. Moreover, commercial CFD solvers tend to use low-order stabilization terms as the default 9,10 to ensure a result, even for otherwise-inadequate discretizations; this tendency amounts to adding artificial viscosity or dissipation to the solution.…”

Artery Length Sensitivity in Patient-Specific Cerebral Aneurysm Simulations

“…For each aneurysm model, we generated 3 model truncations (T): T 0 had the longest possible inlet artery obtained from the scan imaging data, T 1 was truncated orthogonal to the artery axis after the first major arterial turn at the cavernous segment, and T 2 was truncated after the second major arterial turn, at the clinoid segment. For each patient model, the arterial outlets lengths were kept constant at Ͼ5 times the artery diameter 7,10,11 for each of the 3 arterial inlet truncations. When the outlet was near a bifurcation, we truncated even further to allow a more remote outlet boundary condition.…”

“…In addition, the inlet boundary conditions used in the literature, usually velocity profiles, are taken as fully developed and axisymmetric. 1,4,[6][7][8] These assumptions are valid only for flow in straight tubes, which is very different from flow in patient arteries, where the flow is not fully developed even in the common carotid artery, the location where measurements are more readily available. Therefore, using the fully developed axisymmetric velocity profile on a short artery inlet may result in unrealistic flow estimates in the aneurysm.…”