Multi-objective optimization of pulsatile ventricular assist device hemocompatibility based on neural networks and a genetic algorithm.

Given the benefit of pulsatile blood flow for perfusion of coronary arteries and end organs, pulsatile ventricular assist devices (VADs) are still widely used as paracorporeal mechanical circulatory support devices in clinical applications. However, poor hemocompatibility limits the service period of the VADs. Most previous improvements on VAD hemocompatibility were conducted by trial-and-error CFD analysis, which does not easily arrive at the best solution. In this paper, a multi-objective optimization method integrating neural networks and NSGA-II (Non-dominated Sorted Genetic Algorithm-II) based on FSI simulation was developed and applied to a pulsatile VAD to optimize its hemocompatibility. First, the VAD blood chamber was parameterized with the principal geometrical parameters. Three hemocompatibility indices including hemolysis, platelet activation, and platelet deposition were chosen as goal functions. The neural networks were built to fit the nonlinear relationship between goal functions and geometrical parameters. Next, a multi-objective optimization algorithm (NSGA-II) was used to search out the Pareto optimal solutions in the built neural networks. Finally, the best compromise solution was selected from the Pareto optimal solutions by a fuzzy membership approach and validated by FSI simulation. The best compromise solution simultaneously possesses an acceptable hemolysis index, platelet activation index, and platelet deposition index, and the corresponding relative errors between the indices predicted by optimization algorithm and the one calculated by FSI simulations are all less than 5%. The results suggest that the proposed multi-objective optimization method has the potential for application in optimizing pulsatile VAD hemocompatibility, and may also be applied to other blood-wetted devices.