Turbulence is an important feature of blood flow near the heart. A discretization of moderate resolution must therefore model the interactions between "coarse" resolvable flow behaviors and "fine" subgrid-scale features. We compare several choices of turbulence model in terms of their impact on clinically interesting flow statistics. We also present an investigation of a novel method for augmenting classical turbulence models with machine learning algorithms trained on coarse-fine simulation pairs.
