Modelling Aortic Stenosis:Towards clinical decision support and in silico trials

Aortic stenosis (AS) is a narrowing of the aortic valve that can lead to symptoms such as chest pain, fainting and heart failure. AS severity is typically assessed at rest using Doppler Echocardiography, by measuring peak velocity and estimating aortic valve area and mean transvalvular pressure drop. However, these metrics depend on the flow generated by the patient's left ventricle, making it difficult to distinguish true from pseudo-severe AS, the latter occurring when the left ventricle cannot generate sufficient flow to fully open the valve. Moreover, assessments in rest may underestimate severity under higher flow conditions, such as exercise, during which symptoms usually arise. These limitations pose a significant clinical challenge in decision-making on valve replacement. Even after the clinical decision has been made to proceed with valve replacement, success is not guaranteed. Especially, transcatheter aortic valve implantation (TAVI), which is a minimally invasive alternative for surgical valve replacement, is associated with several complications such as paravalvular leakage (PVL) and conduction problems. To mitigate these issues, manufacturers are still innovating and improving TAVI devices, but development is hindered by a long, costly validation process for regulatory approval, involving bench tests, animal studies and human trials. This process faces limitations such as limited repeatability, ethical concerns, and small sample sizes, posing a huge industrial challenge regarding regulatory approval for cardiovascular implantable devices. This thesis aims to address both clinical and industrial challenges using in silico modelling techniques. The clinical challenge can be addressed with patient-specific fluid-structure interaction (FSI) models that can simulate valve dynamics and haemodynamics, while having full control over boundary conditions and thus being independent on the patient's cardiac function. In FSI The industrial challenge can be addressed by integrating in silico trials in the validation pipeline, enabling device testing on large virtual patient cohorts, represented by computational models. Hereby, it is beneficial to make use of virtual cohorts based on synthetic patient data to facilitate large-scale in silico trials. Therefore, this thesis has two objectives regarding the clinical and industrial challenges:1.To develop and verify a patient-specific FSI model to support clinical decision-making on valve replacement;2.To develop and validate a virtual cohort generator for creating synthetic aortic stenosis patients to enable in silico TAVI trials. Chapter 2 addresses the first objective by verifying a patient-specific FSI model against in vitro experiments. The in vitro experiments involved a mock-loop circulation with patient specific rubber valve models. The FSI model replicated the experimental setup by using identical valve geometries and boundary conditions derived from patient data. A high level of agreement was observed between the FSI simulations and the in vitro experiment results in terms of valve dynamics and haemodynamics. This similarity supports the reliability of FSI models as a non-invasive tool for supporting clinical decision-making on valve replacement. Chapter 3-5 focus on the second objective. In Chapter 3, a virtual cohort generator (VCG) is developed to create synthetic aortic valve stenosis geometries. The VCG was based on a dataset of 97 patient-specific stenotic aortic valve geometries, that were parametrised with 33 shape parameters derived through non-parametric statistical shape modelling. Synthetic geometries were generated by sampling from a copula distribution fitted to these shape parameters. The resulting virtual cohorts preserved the distributions of, and correlations between, shape features observed in the real patient data, demonstrating the ability to generate realistic and anatomically plausible synthetic AS geometries, while maintaining population-level statistics. In Chapter 4 the VCG is extended by incorporating synthetic boundary conditions corresponding to the synthetic geometries. Patient-specific pressure signals, obtained from the same patient dataset as described in Chapter 3, were parametrised by calibrating a lumped parameter model, yielding 12 model parameters per patient. Missing model parameters were estimated through data imputation. The 12 model parameters were combined with the 33 shape parameters in a joint copula distribution. Sampling from this distribution yielded virtual cohorts with anatomically and physiologically plausible AS geometries and haemodynamic signals, suitable for use in computational models for in silico trials. Chapter 5 evaluates the feasibility to use the virtual cohorts from Chapter 3 in combination with a fast-to-evaluate model for in silico TAVI trials. The model consisted of a computationally cheap model, simulating TAVI deployment, followed by PVL estimation with a lumped parameter model. Using this model, an in silico trial was conducted on a virtual cohort of 500 patients, with around 50 different model parameter combinations evaluated per patient. The observed relationships between PVL and anatomical and procedural features aligned with clinical findings, demonstrating the feasibility of using virtual cohorts with the fast-to-evaluate model for in silico TAVI trials. To conclude, this thesis presents several in silico modelling techniques to address clinical and industrial challenges in AS. We established a basis for integrating patient-specific FSI models into clinical workflows, offering a non-invasive method to improve AS severity assessment and guide clinical decision-making on valve replacement. Additionally, we proposed methods to advance the use of in silico trials in regulatory approval, paving the way for faster, safer and more efficient validation pipeline for cardiovascular implantable devices.