CFD-Based Kriging Optimization Using OpenFOAM for Determining Empirical Coefficients in Blood Damage Models for Medical Devices

As blood flows through a medical device, flow-induced damage to red blood cells (RBCs) can occur causing them to release hemoglobin, which is potentially toxic. This form of blood damage is referred to as mechanical hemolysis. Most blood-contacting medical devices (e.g., ventricular assist devices, artificial heart valves) must be shown to be safe by demonstrating that they do not cause excessive hemolysis levels. Historically, in vitro experiments and animal studies have been performed to evaluate the hemolytic potential of a device. More recently, CFD has become widely utilized in the design and analysis of medical devices, and there is much interest in using CFD to predict flow-induced hemolysis as a complement to, or potentially in lieu of, some physical testing. A major challenge of hemolysis modeling is the multi-scale nature of the problem, which ranges from the flow physics at the device scale to the biophysics of RBC membrane pore formation and membrane rupture at the sub-micron. As a result, simplistic continuum-scale hemolysis models are often used that attempt to correlate the amount of hemoglobin released from RBCs as a function of the flow-induced stress and exposure time. Such models include several empirical coefficients that are often obtained from experimental measurements in simplified blood shearing devices with uniform shear conditions and well-defined exposure times. These idealized empirical coefficients are then used in CFD simulations to predict the hemolytic potential of a medical device with complex fluid dynamics, often yielding poor predictions that can vary by several orders of magnitude compared to experiments. A straight-forward approach for potentially improving the predictive accuracy of existing hemolysis models is to use empirical coefficients determined from experiments with more realistic flow conditions. To date, however, improved empirical model coefficients have yet to be determined in a complicated device due to the difficulty of accurately defining a characteristic flow-induced stress and exposure time. In this work, we develop a new approach for determining device-specific empirical coefficients by combining CFD and Kriging surrogate modeling using OpenFOAM and Python. Using this method, we determine empirical coefficients for an existing continuum-scale hemolysis model in a device with non-uniform shear conditions. We demonstrate that the resultant coefficients are much different than the traditional idealized empirical coefficients. We then assess the predictive accuracy of using improved empirical coefficients by prospectively predicting hemolysis levels in a hypodermic needle model and comparing our results with complementary experiments. This approach represents a first step towards improved CFD modeling of hemolysis in blood-contacting medical devices. Ongoing research aims to develop an improved physics-based hemolysis model, which will undoubtedly have some degree of empiricism that can be tuned using the present CFD-based Kriging optimization approach.