CDS&E: Uncertainty Quantification and Bayesian Updating in Data-Driven Cardiovascular Modeling

Stanford University, Stanford CA

Investigators

Abstract

CBET - 1508794 Marsden, Alison L. Cardiovascular disease is one of the major problems facing US and the world. While simulations of cardiovascular hemodynamics are now being used to study fundamental processes, trust in personalized simulation results before making clinical decisions for a patient is absent due to several uncertainties. This is exactly what this proposal is about, investigating these uncertainties and developing techniques to allow informed medical decisions. The co-PIs propose to disseminate their computational tools as open source programs. Though it is well known that cardiovascular simulations require numerous assumptions and assimilation of uncertain clinical data, these uncertainties currently get swept under the rug, asking end-users to accept deterministic simulation predictions as "truth" with no associated statistics. As a result, researchers and clinicians are left to wonder "How reliable are simulation predictions in light of myriad uncertainties?" and "How do the statistics on output predictions change with differing methodologies and assumptions?". These questions lead to justified skepticism in the research and clinical community, and are a roadblock to adoption. Development of transformative technology to assess uncertainty, currently lacking in the field, is of paramount importance for safe and routine adoption of simulations for personalized medicine and biomechanics research. This is the area that this proposal comes to cover, as it aspires to develop techniques that can lead to the incorporation of data-driven cardiovascular models to inform decisions surrounding choices of drug therapy, device placement, surgical methods and interventions for individual patients. The proposal has two goals: 1) Develop fast automated methods for parameter estimation and assimilation of uncertain data into multiscale models, 2) Develop an efficient framework to propagate uncertainties from clinical and imaging data to simulation predictions. It is proposed to demonstrate the uncertainty quantification (UQ) framework through application to multiscale simulations of coronary artery disease (CAD), though the framework will apply to a wide range of other cardiovascular and respiratory diseases. Simulations will be run in a high performance computing (HPC) environment using a multi-level parallel algorithm structure. The ultimate goal is to address currently unanswered questions about reliability and robustness in cardiovascular simulation. Results from this work, if successful, would enable acceptance of computational models and establish reliability metrics to guide model improvement and data collection. Cardiovascular simulations have potential to personalize treatments for individual patients and to characterize the in vivo mechanical environment, providing key biomechanical data that cannot be readily obtained from medical imaging. The proposed computational framework could be applicable to a range of problems in biomedical computing, biological modeling, and engineering applications using computational fluid dynamics. Dissemination will be achieved through contributions to the SimVascular open source project, for which Dr. Marsden is the PI. Activities that integrate research and teaching by introducing statistics concepts in graduate level courses and through outreach to middle and high school students are proposed.