Comparison of numerical methods for cerebrospinal fluid representation and fluid-structure interaction during...

Spinal cord impacts can have devastating consequences. Computational models can investigate such impacts but require biofidelic numerical representations of the neural tissues and fluid-structure interaction with cerebrospinal fluid. Achieving this biofidelity is challenging, particularly for efficient implementation of the cerebrospinal fluid in full computational human body models. The goal of this study was to assess the biofidelity and computational efficiency of fluid-structure interaction methods representing the cerebrospinal fluid interacting with the spinal cord, dura, and pia mater using experimental pellet impact test data from bovine spinal cords. Building on an existing finite element model of the spinal cord and pia mater, an orthotropic hyperelastic constitutive model was proposed for the dura mater and fit to literature data. The dura mater and cerebrospinal fluid were integrated with the existing finite element model to assess four fluid-structure interaction methods under transverse impact: Lagrange, pressurized volume, smoothed particle hydrodynamics, and arbitrary Lagrangian-Eulerian. The Lagrange method resulted in an overly stiff mechanical response, whereas the pressurized volume method over-predicted compression of the neural tissues. Both the smoothed particle hydrodynamics and arbitrary Lagrangian-Eulerian methods were able to effectively model the impact response of the pellet on the dura mater, outflow of the cerebrospinal fluid, and compression of the spinal cord; however, the arbitrary Lagrangian-Eulerian compute time was approximately five times higher than smoothed particle hydrodynamics. Crucial to implementation in human body models, the smoothed particle hydrodynamics method provided a computationally efficient and representative approach to model spinal cord fluid-structure interaction during transverse impact.