Automated Imaging-to-Flow for Patient-Specific Fluid-Structure Interactions
PRESENTING AUTHORS FULL NAME: Sarah Vigmostad
INSTITUTION: The University of Iowa
ADDITIONAL AUTHORS NAMES, AS TO BE PUBLISHED: Aaron Goddard
BACKGROUND/PURPOSE: Computational fluid mechanics has long been employed to study cardiovascular hemodynamics in the context of disease mechanisms, progression, and treatment, but its translational potential has yet to be fully realized. Further advances in computational tools are necessary in order to make these tools practical for clinical use, so that patient-specific parameters can be coupled with biomechanical indices to guide diagnosis and treatment decisions. Currently, the modeling of patient-specific hemodynamics continues to require extensive user-input, which hinders high-throughput models or models that include wall motion. Since patient geometry such as the left ventricle can be captured at various points during the cardiac cycle, information like wall motion can be obtained and prescribed a priori to reduce real-time computations without compromising on patient-specific knowledge.
METHOD: With these goals in mind, the current work describes an Eulerian approach to represent cardiovascular geometry, displacement, and motion. This is achieved using levelsets, which are used throughout the modeling workflow, from segmenting the medical images to representing the geometry within a Cartesian grid flow solver. To integrate with the specific needs of the current flow solver, a morphing technique has been developed that is compatible with adaptive mesh refinement, parallel domain decomposition, and the need to supply local interface velocities to the flow solver which describe both normal and tangential motion. Through skeletonization, warp and blend morphing ensure that tangential wall motion is accurately captured, a limitation of previous Eulerian-based approaches.
RESULTS: This framework has been validated with several moving boundary problems and shows sub-grid accuracy. Figure 1 shows streamtraces of a 3D left ventricle simulation throughout the cardiac cycle. The simulation was performed without any meshing, by morphing levelset-based interface positions that were initially obtained over fifteen time points within the cardiac cycle, to fully describe wall motion at a CFD-compatible temporal resolution and drive flow throughout the cardiac cycle.
CONCLUSION: By employing an Eulerian representation throughout the modeling workflow, the current approach circumvents the need for Lagrangian-based surface or volume meshing, to support the long-term goal of integrated imaging-to-flow without user intervention, so that computational modeling can be employed to aid clinical decision-making, and high-throughput computations can be performed on large patient cohorts.