• AF
• artificial intelligence.
• atrial fibrillation
• CFD
• computational fluid dynamics
• hemodynamics
• LAA
• left atrial appendage
• neural network
• u-net

The left atrial appendage (LAA) is a finger-like extension of the main body of the left atrium (LA) with a highly variable shape. In patients with atrial fibrillation (AF), the LAA is the region most prone to thrombus formation because of its unique morphology, which is characterized by high inter-patient variability, and the low flow velocities that occur within it, which promote blood stasis and thus thrombus formation. In this context, numerical simulation, in particular Computational Fluid Dynamics (CFD), represents a useful investigative tool for the assessment of LA and LAA hemodynamics and related thrombogenic risk on a strictly patient-specific level.
Despite the growing interest in the development of these models, there remain limitations related to (i) the time required for the segmentation procedure and thus the definition of patient-specific geometry, and (ii) the definition of boundary conditions and mechanical properties that realistically represent the flows and behavior of the atrial wall of the patient under study.
In fact, most CFD studies of the LA/LAA performed fixed-wall simulations, assuming the absence of contraction typical of AF. In contrast, other studies have implemented fluid-structure interaction (FSI), but approximate the mechanical behavior of the atrial wall with data from tests performed on ex-vivo animal specimens.
This thesis proposes an approach to overcome the limitations of both fixed-wall simulation and FSI by imposing patient-specific motion of the LA walls as measured from CT images, througoh performing CFD moving wall simulations.
Thus, this thesis aims to develop a semi-automated framework to obtain patient-specific CFD moving walls models from patient CT images for the study of LAA hemodynamics and its correlation with thrombo-embolic risk.