Patient-specific mesoscopic blood flow simulation: From image data to flow analysis

With the advancement of tomographic imaging techniques and high performance computing facilities, more realistic blood flow simulations based on patient-specific geometries have become feasible. One important application of such simulations is the haemodynamical assessment of carotid artery stenosis in order to assist the physicians in the planning of endoscopic or surgical interventions. However, despite the considerable capabilities of the various tools available, the clinical use of CFD methods is still limited due to the following reasons: • Even the simplest possible flow models are still computationally expensive. In many cases, a large amount of the computation time is spent on the complex mesh generation. In addition, more advanced model specifications such as pulsatile flow, non-Newtonian flow models, and fluid-structure interaction, will drastically increase the computational costs. • The accuracy of the simulation results, depends on the choices of inflow and outflow conditions. These conditions involve the imposition or extrapolation of unknown velocities or pressure values which often do not correspond with the ones in reality. • The workflow for constructing the models is complex and user-dependent, and currently can only be done by CFD experts. In fact, apart from very few research tools, no specialized software is available for performing computational haemodynamics. We present a software framework for image-based blood flow analysis in an effort to meet the above challenges. Our software solution spans a broad range of techniques, from MRI image processing to parallel flow simulation and finally, to a flow analysis geared towards specific clinical problems. For our flow simulation, we use the Lattice-Boltzmann (LB) method. This method provides advantages compared to traditional flow solvers due to its simplicity and parallel performance. In- and outflow conditions are based on 4D flow-sensitive MRI measurements (Phase-contrast MRI), which are increasingly used in clinical establishments and provide an ideal basis for parameterization of simulation models. The simulation geometry (segmented vessel) is constructed from the MRI anatomic data. Furthermore, the initial and boundary conditions are set with a high degree of automation using flow-sensitive MRI measurements. A surface model is then generated and voxelized to a desired grid resolution. The flow solver is implemented with the help of the open-source Lattice-Boltzmann library OpenLB using BGK dynamics and Skordos boundary conditions. For analysis of simulation results and quantification of clinical parameters, a research software dedicated to cardiovascular flow analysis is used. Quantification methods include the evaluation of flow rates and velocity profiles on user-defined vessel cross-sections, centerline pressure analysis and probabilistic flow tracking. In our first evaluation study, the workflow has been tested for steady flow in stenosed vessel phantoms. At moderate Reynolds numbers (Re=625 at stenosis), velocity and pressure results are accurate compared to established FEM flow solvers. For our second evaluation, results for steady systolic flow in a healthy human carotid artery bifurcation were compared to 4D flow-sensitive MRI image data. In our current study, we investigate the effects of virtual treatment of carotid artery stenosis. For this purpose, steady systolic flow is simulated in a patient with plaque in the internal carotid artery (ICA), based on MRI anatomy and flow measurements. The stenosis is then virtually removed by local dilation of the surface mesh. The efficiency of treatment is evaluated by quantifying the pressure drop along the centerline of the ICA.