MitrAccel

MitrAccel: accelerated simulation of mitral heart valve biomechanics 

PI: Dominik Obrist (University of Bern)

Co-PIs: Michael Sacks 

July 1, 2021 – June 30, 2024

The aim of this project is to develop a comprehensive, high-speed software base for modelling the structural and hemodynamic function of the diseased mitral heart valve (MV). Contrary to the aortic heart valve, the MV is an intimately coupled, fully functional part of the LV. In situations where the MV fails to fully close during systole, the resulting blood regurgitation into the left atrium typically causes pulmonary congestion, leading to heart failure and/or stroke.   

A major cause of mitral regurgitation (MR) is induced by ischemic cardiomyopathy and is termed ischemic mitral regurgitation (IMR). IMR is present in up to 40% of patients and more than doubles the probability of cardiovascular morbidity after 3.5 years. There is now agreement that adjunctive procedures are required to treat IMR caused by leaflet tethering. However, there is no consensus regarding the best procedure for a particular patient. Given the number of proposed procedures and the complexity and duration of such studies, it is highly unlikely that IMR procedure optimization will be achieved by prospective clinical trials. Moreover, the selected method of repair is typically based only on previous clinical experience and does not take advantage of the considerable information present in modern clinical imaging. There is also a lack of predictive data which could support interventional planning, such that there exists a high need for validated computational tools to support the decision process of the clinician with quantitative data. Novel computational approaches directed toward optimized surgical repair procedures can substantially reduce the need for trial-and-error approaches.  

Detailed, patient-specific numerical prediction of post-surgical mechanical stress fields within the valvular tissue and of complex blood flow that drive the MV can (a) provide new insight into the pathogenesis of MR pre- and post-intervention, (b) support the development of implantable devices and (c) enable novel diagnostic methods for mitral valve disfunction using machine learning techniques. In the project MitrAccel, we will focus on edge-to-edge repair of the MV which can be performed with minimally invasive technology using clips. We will develop a software base and a workflow to simulate this repair technique on patient-specific MV models. These models will be embedded in a high-performance fluidstructure interaction (FSI) solver which simulates stress fields within the valve structure, in the blood flow and at the interface between fluid and structure. Novel material laws for the leaflet tissue will enable the prediction of post-repair remodeling of the leaflet tissue which may involve effects such as MV plasticity. This predictive capacity of MitrAccel will be a major step toward the support of clinical decision making with high-performance computing tools.   

The project MitrAccel will profit from a rich software framework that has been developed in the context of two earlier PASC projects (AV-FLOW, 2014-17; HPC-PREDICT, 2017-21). These projects addressed similar objectives in the context of the aortic valve and the ascending aorta. Several software components developed in these projects (e.g. FSI solver, IO library, data securtity and encryption) can be reused. New elements (MV finite element model, segmentation tools, neural network material laws) are based on work by the renowned Sacks Lab and will be adapted to MitrAccel and added to the software base.   

A significant portion of the project efforts will go into the continued development and acceleration of the numerical solvers to run on a GPU-based HPC infrastructure. We will integrate GPU-accelerated kernels a novel Poisson solver into the FSI solver. This Poisson solver (originally developed for our Navier-Stokes solver IMPACT) has been shown to speed up the time-to-solution for complex flow problems on multi-GPU systems by two orders of magnitude. This work on GPU-acceleration of the software base will be complemented by development work to make the solvers performance-portable (e.g. by using  Kokkos|Trilinos). Finally, we will define a simulation workflow for MitrAccel to facilitate the adoption of this software by clinicians. This task will profit from our experience gained in the project HPC-PREDICT where a complex clinical imaging workflow was designed to improve clinical diagnostics.   

The project comprises three work packages:  

• Mitral valve modelling: patient-specific image segmentation, definition of FE-model and material law, integration of MV model into FSI solver 
• Acceleration of FSI solver: integration of novel Poisson solver into FSI solver, performanceportable solvers, GPU-acceleration of kernels 
• Workflow integration: definition of clinical workflow, collection and container-packaging of tools, testing and deployment of software base  

Two PhD students and one post-doctoral researcher shall be (co-)funded through this project.  The PhD students will focus on mitral valve modeling and FSI solver acceleration, respectively, whereas the postdoctoral researcher will have the task of integrating the mitral valve FE-model into the FSI framework and to guide the design of the MitrAccel workflow. To enhance the collaboration between the involved labs, the post-doctoral researcher work as a shared research fellow between the Obrist lab at the University of Bern and the Sacks lab at the University of Texas at Austin.   

In addition to the scientific core team, the MitrAccel project will install a Software Engineering Service with a part-time software engineer to provide access and support for various development tools. This includes the management of a CI/CD environment, the GIT version control system and the programming environment with numerical libraries. Further, this service provides technical support for web-based user interfaces, object store and data encryption. The software engineer maintains a close relation with the PASC core team of CSCS.