Patient-specific prediction of abdominal aortic aneurysm expansion using efficient physics-based machine learning approaches

Computational vascular Growth and Remodeling (G&R;) models have been developed to capture key physiological and morphological features during the arterial disease progression and have shown promise for aiding clinical diagnosis, prognosis prediction, and staging classification. However, the translation of computational G&R; models into their applications has yet to wait for clinical practice. Partly, due to the high complexity of the arterial adaptation mechanism, high-fidelity arterial G&R; simulations typically require hours or even days, which hinders its time-consuming applications such as patient-specific parameter estimation, disease prediction, verification, validation, and sensitivity analysis. Furthermore, the typical Finite Element Method (FEM) based computational G&R; model should be extended to provide the uncertainty quantification associated with simulation and prediction results. Therefore, to enhance practicality of the G&R; modeling, we develop a novel and computationally efficient simulation framework that comprehensively combines physics-based G&R; simulations and data-driven machine learning methods using a Multi-Fidelity Surrogate (MFS) approach. This greatly enhances the computational efficiency of arterial G&R; simulations, enabling more time-consuming applications such as personalized parameter estimation. The proposed framework is then tested for a specific disease, Abdominal Aortic Aneurysms (AAAs), by estimating G&R; model parameters from follow-up CT images in 21 patients.