Kristian Valen-Sendstad
Computational fluid dynamics (CFD) is increasingly relied upon for elucidating blood flow dynamics in cerebral aneurysms and their possible role in determining rupture risk. Compared with a decade ago, when “patient-specific” aneurysm CFD studies were confined to a few specialized labs, often using their own in-house solvers, today the use of CFD in aneurysm research is widespread, facilitated by more user-friendly commercial solvers as well as the now-routine availability of 3D angiography. With this popularity, however, has come increased scrutiny by clinicians.1 While the many underlying physical assumptions and approximations behind CFD models have been roundly questioned, much less attention has been paid to reliability of the CFD solutions themselves.
Our study, “Mind the Gap,”2 arose from our earlier observation of turbulent-like flow instabilities in a single MCA aneurysm case, when ultra-high-resolution CFD techniques were brought to bear.3 Subsequently, flow instabilities were observed in 4/8 MCA bifurcation aneurysm cases subjected to high-resolution (HR) CFD, albeit under steady inflow conditions.4 In light of these findings, we were surprised to find few, if any, reports of flow instabilities in the extensive aneurysm CFD literature. On the other hand, in vitro and in vivo studies from the 1970s and 1980s reported high-frequency velocity fluctuations in aneurysms, which had been postulated to explain the prevalence of aneurysm bruits. Based on our reading of the aneurysm CFD literature, we hypothesized that the solver numerics and/or model resolutions, informed by the assumption of smooth laminar flow, might be to blame.
To test this, in “Mind the Gap” we performed a controlled study of the same MCA aneurysm cases from our earlier work, now under more physiologically realistic pulsatile flow conditions, and, crucially, using both our HR solution strategy and a “normal resolution” (NR) strategy typical of the aneurysm CFD literature. Our HR solutions identified high-frequency flow instabilities in the same 4/8 cases from our earlier study; however, they were absent from the corresponding NR solutions, which in some cases predicted very different flow and wall shear stress dynamics. The good news was that at least some of the commonly-used reduced hemodynamic indices (derived by time and spatial averaging of the full CFD data) were relatively impervious to the solution strategy.