2026 Theses Doctoral

Development and Application of Personalized Lumped Parameter (0D) and Multiscale (3D-0D) Computational Models of Blood Flow for Cardiovascular Disease Applications

Chen, Yurui

Cardiac diseases such as borderline left ventricle (BLV), coronary artery disease (CAD), and aortic root aneurysm represent some of the most critical cardiovascular disorders, each requiring precise, patient-specific assessment to guide surgical and interventional decisions. This dissertation focuses on developing and applying multiscale, personalized computational modeling frameworks to improve diagnosis, predict surgical outcomes, and deepen our understanding of underlying hemodynamic mechanisms across these conditions.

For borderline left ventricle (BLV) patients—where surgeons face a crucial choice between biventricular repair (BiVR) and stage-1 palliation (S1P)—we developed a personalized lumped parameter network (LPN) model that captures neonatal cardiovascular physiology in detail. The model was adapted and expanded from existing literature to include a four-chamber heart, patent ductus arteriosus, and patient-specific pulmonary/systemic resistances.

By integrating retrospective clinical data, the model reproduced preoperative hemodynamics and enabled virtual surgery simulations of both BiVR and S1P procedures. The predicted postoperative outcomes successfully reflected observed clinical trends. To address variability in patient data and modeling assumptions, I implemented Bayesian parameter estimation using the DREAM MCMC algorithm, quantifying both aleatory (measurement noise) and epistemic (model structure) uncertainties. This probabilistic framework improved confidence in model-based surgical decision-making.

In coronary artery disease (CAD), I developed a multiscale 0D–3D coupled model that explicitly incorporated myocardial motion extracted from patient-specific CT images using deformable image registration. The moving-wall blood flow was simulated using an arbitrary Lagrangian–Eulerian (ALE) formulation implemented in our in-house solver svMultiphysics.

Results revealed that ignoring cardiac motion leads to underestimation of hemodynamic indices such as pressure drop and instantaneous wave-free ratio (iFR), particularly in the right coronary artery where motion is most pronounced. Across an 11-patient cohort, motion inclusion consistently produced lower iFR values near the clinical intervention threshold, emphasizing the need to consider wall kinematics in computational diagnostics for CAD.

For aortic root aneurysm (ARA), I constructed a fully coupled fluid–structure interaction (FSI) model incorporating personalized aortic root geometry, valve dynamics, and coronary flow coupling through an LPN boundary model. The framework was used to simulate progressive aneurysm dilation under different loading conditions, providing insights into how flow–structure interactions alter wall shear stress distributions and oscillatory shear index (OSI). The results suggested that changes in local hemodynamic environments—rather than diameter alone—may better predict aneurysm progression and rupture risk.

Collectively, these studies establish a personalized, multiscale computational ecosystem spanning from 0D lumped-parameter to full 3D FSI models. By integrating uncertainty quantification, patient-specific imaging, and virtual surgical simulations, this dissertation demonstrates how computational modeling can evolve from a purely research tool into a clinically informative system that supports quantitatively informed cardiovascular decision-making. Future work will focus on prospective validation and AI-driven model acceleration to achieve near–real-time clinical integration.