Numerical blood flow simulation in surgical corrections: what do we need for an accurate analysis?

Background: Computational fluid dynamics has been increasingly used in congenital heart surgery to simulate pathophysiological blood flow, investigate surgical options, or design medical devices. Several commercial and research computational or numerical codes have been developed. They present different approaches to numerically solve the blood flow equations, raising the question whether these numerical codes are equally reliable to achieve accurate simulation results. Accordingly, we sought to examine the influence of numerical code selection in several complex congenital cardiac operations. Material and methods: The main steps of blood flow simulations are detailed (geometrical mesh, boundary conditions, and solver numerical methods) for congenital cardiac operations of increasing complexity. The first case tests different numerical solutions against an analytical, or exact, solution. In the second case, the three-dimensional domain is a patient-specific superior cavopulmonary anastomosis. As an analytical solution does not exist in such a complex geometry, different numerical solutions are compared. Finally, a realistic case of a systemic-to-pulmonary shunt is presented with both geometrically and physiologically challenging conditions. For all, solutions from a commercially available code and an open-source research code are compared. Results: In the first case, as the mesh or solver numerical method is refined, the simulation results for both codes converged to the analytical solution. In the second example, velocity differences between the two codes are greater when the resolution of the mesh were lower and less refined. The third case with realistic anatomy reveals that the pulsatile complex flow is very similar for both codes. Conclusions: The precise setup of the numerical cases has more influence on the results than the choice of numerical codes. The need for detailed construction of the numerical model that requires high computational cost depends on the precision needed to answer the biomedical question at hand and should be assessed for each problem on a combination of clinically relevant patient-specific geometry and physiological conditions.