High-speed high-fidelity computational modeling approaches for cardiac applications

This dissertation presents a high-speed, high-fidelity computational modeling approach for cardiac simulations, with a focus on replacement heart valves and left ventricular function. High-fidelity, structurally-based computational models were developed to simulate replacement heart valves. Through a structural constitutive model for electrospun biomaterials, the study discovered the presence of novel fiber-fiber interactions and the absence of fiber-gel interactions in electrospun biomaterials. Moreover, the model parameters were related to parameters controllable in the manufacturing process, presenting a way to narrow down optimal biomaterials in-silico. A comprehensive simulation pipeline was also developed to simulate BHV response under cycling loading upto 50 million cycles. The model predicted lasting changes to BHV leaflet shape and underlying structure within the first 20 million cycles, and found that these changes stabilize thereafter, a crucial insight that must be incorporated during BHV design. While these models are capable of providing excellent insights, their practical clinical applications using traditional simulation methods are impeded due to their prohibitively slow speed. To overcome these computational challenges, a novel neural network finite element approach was developed for high-speed high-fidelity cardiac simulations. This approach learns a family of solutions to a parametric PDE describing cardiac mechanics. The novelty of our model lies in 1) learning the underlying physics directly from the weak form of the PDE, either through the potential energy or the virtual work formulation, 2) no reliance on any additional experimental or simulation generated data for accurate predictions, 3) simultaneous training over the complete physiological loading range, 4) prediction accuracy of 0.1\% relative to the conventional finite element method along with a prediction time of a few seconds and 5) NURBS mapping integrated within the method to accurately capture complex geometry. Previous research has repeatedly demonstrated the importance of a full heart model for accurate heart valve simulations, hence this study also includes the development of a NNFE model for a heart valve leaflet and a left ventricle, serving as intermediate steps towards more sophisticated full cardiac models. Finally, an NNFE model was developed to simulate effects of myocardial infarction in the left ventricle. While this work is a proof-of-concept on idealized geometries, there is a clear potential to extend it to realistic shapes and simulate additional variables including variable geometry and material properties. This approach overcomes the computational challenges in practical applications of high-fidelity cardiac computational models and opens pathways for patient-specific clinical diagnosis and treatment planning.

Name thesis:degree_name

Doctor of Philosophy

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

The University of Texas at Austin

Year dc:date.issued

2024

Author dc:creator

• Motiwale, Shruti

Advisor dc:contributor.advisor

• Sacks, Michael S.

Committee members dc:contributor.committeemember

• Kenneth Diller
• Nicholas P Fey
• Yen-Hsi Richard Tsai

• Replacement heart valves
• Neural networks
• Left ventricle
• Scientific machine learning
• Myocardial infarction

Language dc:language.iso

English

Identifier URI

https://doi.org/10.26153/tsw/59588

OAI identifier oai:identifier

oai:repositories.lib.utexas.edu:2152/132244