An Automated Framework for Large-Scale Simulation of Patient-Specific Diseased Coronary Arteries

Atherosclerosis is an inflammatory vascular disease characterized by the development of lipid-rich plaques within the arterial wall. In coronary arteries, their growth leads to coronary artery disease (CAD) as a consequence of arterial lumen occlusion and eventually to myocardial infarction (MI) as most severe clinical event. Therefore, there is an urgent need of developing methods for CAD diagnosis and management. In clinical practice, the assessment of coronary lesion severity typically relies either on an anatomical evaluation, based on parameters associated with the local luminal narrowing, or on functional evaluation, based on fractional flow reserve (FFR), representing the flow reduction in the vessel due to the presence of a coronary lesion. In addition, hemodynamics, and particularly wall shear stress (WSS) acting directly on the endothelium, have been recognized as key factors in the onset and progression of atherosclerotic plaques and ultimately leading to MI. In this context, Computational Fluid Dynamics (CFD) has become an essential tool in cardiovascular research, enabling the quantitative characterization of local hemodynamic features in patient-specific coronary arteries reconstructed from clinical imaging. Despite its clinical relevance, the quantitative assessment of the prognostic role of WSS calculated through CFD in patient-specific coronary geometries remains challenging, mainly due to the limited availability of large datasets and the associated computational costs. In recent years, artificial intelligence (AI) techniques have emerged as promising surrogates for CFD-based analyses, offering the potential to accelerate WSS estimation. However, these approaches also depend on extensive datasets. Altogether, these limitations underscore the need for automated and scalable CFD simulations. Therefore, this work aims at the automatization of a pipeline for steady-state CFD simulations designed to reproduce patient-specific coronary hemodynamics under both rest and hyperaemic conditions. A dataset comprising 187 coronary arteries from 80 patients, reconstructed from three-dimensional quantitative coronary angiography (3D-QCA), was investigated. After a first mesh independence grid analysis, CFD simulations were carried out with the developed automatized pipeline and finally a quantitative analysis of both surface-based and volume-based hemodynamic parameters was conducted. In detail, for the simulations at rest, the focus of the analysis was put on WSS, while the pressure field and pressure-based derived quantities (i.e., vFFR and pressure drop across the lesion segment) were analysed for the simulations in hyperaemia. While surface-averaged WSS over the lesion segment was characterized by a distribution with median equal to 2.44 Pa (interquartile range IQR 1.63 – 2.95 Pa), vFFR and the translesional pressure drop were instead characterized by distributions with median equal to 0.93 (IQR = 0.91 – 0.96) and 5.10 mmHg (IQR = 2.16 – 6.11 mmHg) respectively. In conclusion, the present thesis enabled the development of an automated framework to perform CFD simulations with minimal user-intervention. The proposed pipeline can facilitate large-scale in silico investigations and support data-driven or AI–based methodologies relying on large amount of data.