Virtual Physiological Blood: an HPC framework for blood flow simulations in vasculature and in medical devices
PI: Petros Koumoutsakos (ETH Zurich)
Co-PIs: Bastien Chopard, Mauro Pezzè
July 1, 2017 - June 30, 2020
Blood flow is involved in most of the fundamental functions of living organisms in health and disease. It is essential for the transport of oxygen and nutrients, as well as of infectious parasites and metastasizing tumor cells, to tissues and organs. Blood flow has been studied for thousands of years. Observations and experiments have evolved from qualitative descriptions to precise measurements of blood flow rates in vivo. Despite remarkable advances, experiments have limitations on the type and detail of information they can provide for blood flow. The quantification of blood rheology, in particular in complex vascular geometries and in disease, is an open challenge. Even more, the prediction of important quantities such as shear stresses, margination and drug transport associated with blood flow in capillaries and medical devices are still today obtained by decades old empirical formulas with non-quantified uncertainties.
In the past twenty years simulations have advanced to complement experiments and have become an essential tool for investigations of blood flow in animal research and patient care. Simulations have provided insight and detailed quantitative information on the functioning of blood in arteries and capillaries and their effects on the surrounding blood vessels and tissues. Success stories include the elucidation of the inception of aneurysms and the devising of mechanisms for their repair. More recently simulations have been used to design microfluidic devices that aim to diagnose the transport of circulating tumor cells, a potent marker for cancer metastasis.
Despite such advances, we believe that there is significant room for improvement in terms of fidelity and clinically relevant scales for such simulations. For example, most large scale blood flow simulations to date have discarded the particulate nature of blood and its rheology and they have utilised ad-hoc, a-priori specified, Newtonian or non-Newtonian viscosity coefficients. Even particle based simulations of blood flow, with techniques such as Dissipative Particle Dynamics (DPD), have been found to predict quantities of interest that differ drastically from those obtained via experiments or from high fidelity simulations of canonical problems using boundary integral methods. Such discrepancies are to be expected as both experimental and numerical observations of Red Blood Cells (RBCs) behavior in flow have revealed very complex dynamics for individuals and populations of cells. We expect that approaches that allow micro-scale descriptions of blood at spatiotemporal scales afforded by the present and proposed work on DPD and Lattice Boltzmann (LB) could remedy this situation. Furthermore, we expect that fast and validated simulations of blood flow are also essential to design, optimize and understand micro-fluidic devices aimed at performing tasks such as blood separation, detection of Circulating Tumor Cells (CTC) and bacteria, and molecular recognition with high sensitivity and specificity. Simulations, such as those proposed here will help to advance the technology and have impacts ranging from clinical diagnostics to regenerative medicine, proteomics and organs on a chip.
The goal of this project is to provide computational tools for a virtual physiological blood flow in complex geometries pertinent for simulations in vasculatures and medical devices. The project combines the expertise of three research teams in Switzerland (University of Geneva, ETHZ, USI) to provide a portable and performing simulation tool for blood flow, with fully resolved RBCs and platelets. We will deliver integrative and scalable HPC framework, capable of performing validated simulations at the micro- and macro-scale and overcoming several of the limitations of existing software in terms of time to solution, portability and ease of use by non-computing experts. We will base our developments on two state of the art codes: udeviceX that uses DPD allowing for micro-scale descriptions of individual RBCs and Palabos that employs LB methods for meso- and macro-scale continuum flows. We will validate systematically DPD and LB simulations tools using relevant experiments and high fidelity simulations involving boundary integral methods. Our validation and uncertainty quantification studies will recognise and tackle the heterogeneity of experimental and computationally available data. We will couple DPD and LB methods in order to provide a multi-scale description of the flow. Our goal is software that can be easily used and to conform with the decision of the user to tackle blood flow in the particulate, the continuum or the hybrid level. We expect to provide a framework of validated computational models that can assist scientists and clinicians in understanding blood flow and at the same time for designing devices, such high throughput micro-fluidics used to diagnose blood borne diseases.
Currently there are several teams (including Biros, Hoekstra, Karniadakis, Kaxiras, Melchiona, Sotiropoulos and Succi) that develop and use software for large scale continuum or particle based simulations of blood flows. We mention the group of Karniadakis (Brown University) who has pioneered the use of DPD for blood flow simulations and the group of Alfons Hoekstra (University of Amsterdam) who has performed extensive simulations using LB methods. These groups are among our collaborators and we plan to continue our interactions with them. The group of Biros has performed state of the art simulations of RBCs using Boundary Integral Methods (BIM). We have established a contact in order to use such high fidelity simulation to assess the capabilities of DPD methods in resolving canonical problems involving the interaction of RBCs. We believe that this project will distinguish itself by combining validated DPD and LB approaches with quantified uncertainties along with a proper software engineering framework that we feel is lacking from most engineering driven approaches.
We note that DPD and LB are techniques that are used across many disciplines (from fluid dynamics, to traffic simulations and materials science) and we will develop software that is capable of allowing such trans-disciplinary use of our codes. Furthermore, we expect to make such portability evident and hope to establish these sate of the art developments as working tools for blood flow simulations in the supercomputing platforms at CSCS. We plan to continue our tradition of making software open source (both Palabos and uDeviceX are openly available at github) and to provide the necessary documentation such that the code is readily usable and accessible on multiple platforms and by users with varying expertise in supercomputing. Our team has expertise in this field through the design and implementation of integration middleware for grid (LAMMPS) and particle based simulation engines (MRAG) developed within the PASC project “Angiogenesis in Health and Disease: in-vivo and in-silico”.
On the application side, our team has established collaborations with clinicians and experimentalists that can guide and use our developments. There is an existing tight collaboration between the Chopard group and Prof. Karim Zouaoui, biologist at ULB and CHU Charleroi, who is performing flow chamber experiments with whole blood. This collaboration started in the FP7 THROMBUS project. Another strong link with the medical community is given by the H2020 CompBioMed project in which B. Chopard is partner. CompBioMed is a center of excellence for HPC biomedical simulations whose goal is to provide clinicians and medical companies with access to HPC codes helping them to optimize devices, to choose the best treatments for a disease, and of course to bring a better understanding of many physiological processes. The group of Petros Koumoutsakos has tight collaborations with the group of Mehmet Toner at Harvard Medical School, on the design of high throughput micro-fluidic devices for the capturing of Circulating Tumor Cells. Further collaborations include the group of Mauro Ferrari at Southern Methodist University on simulations of RBCs in artificial capillaries and with the group of Luisa Iruella-Arispe at UCLA on blood flow in the retina. Chopard and Koumoutsakos are the Swiss coordinators of the OpenMultiMed COST action that can be a platform for sharing and discussing the results of this project.