Description
- Abstract:
- Aortic dissection is responsible for significant morbidity and mortality in children, young and older adults. One possible outcome for an artery undergoing dissection is that the crack turns inward and forms a false lumen, while the other one can be the crack turning outward and causing lethal internal bleeding. Although most of the studies focused on the stage with a developed false lumen, there has been modest attention on the initiation and propagation of aortic dissection. One hypothesis states that aggregates of glycosaminoglycans (GAGs) play an important mechanical role in facilitating aortic dissection. However, the underlying biomechanical factors (such as elastin alignment and intralamellar collagen) are neglected, which are imperative in predicting the dissection outcomes. This significant knowledge gap causes difficulty in understanding the mechanism of the initiation and propagation of aortic dissection. In this thesis, we propose to investigate GAGs-initiated aortic dissection and its relevant biomechanical systems by developing imaging-driven multimodality and multifidelity computational and artificial intelligence models. Specifically, synthetic and ex vivo data, such as subject-specific geometries, regional wall displacement, and collagen/elastin alignment, will be incorporated in machine learning and particle-based models for investigating the initiation and propagation of aortic dissection under the influence of subject-specific biomechanical factors. In addition, we will inform relevant ex vivo/synthetic data to novel physics-informed models for inferring underlying biomaterial and biomechanical properties. Thus, we propose the following specific aims: Aim 1: Use novel ex-vivo data collected from mouse models in conjunction with simulated data to develop computational models for investigating intramural delamination processes and their relevant biomechanical systems. Toward this end, we will develop neural operators (DeepONet) informed by continuum-level finite element models and imaging-driven particle models to investigate vascular biomechanics and the early stage of aortic delamination. Aim 2: Develop machine learning models to quantify uncertainty of subject-specific biomechanical properties of the aorta based on measurable data. Toward this end, I will develop physics-informed machine learning to non-invasively infer the regional biomaterial properties of a thrombus in an artery based on synthetic imaging data. Then, we will develop a probabilistic framework to learn the functional knowledge of the constitutive model of the aorta (described by the four-fiber family model) using generative adversarial network (GAN) and DeepONet. The trained models will be combined with a Bayesian inference framework to enable efficient quantification of the posterior biomechanical stresses given sparse measurements. Under the umbrella of vascular mechanics and diseases, this thesis developed a series of robust computational models and machine learning models with uncertainty quantification to address the pressing challenges in understanding aortic dissection, e.g., nonlinearity and complexity of the system, unknown mechanism, expensive data acquisition and in silico investigations, etc. Completing the proposed aims will contribute towards building an improved patient-specific probabilistic modeling of chronic dissection and interventional planning in the future.
- Notes:
- Thesis (Ph. D.)--Brown University, 2022
Citation
Yin, Minglang,
"Hybrid Computational-Machine Learning Models with Uncertainty Quantification for Aortic Dissection"
(2022).
Engineering Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:g8xtm552/
