In Silico, Patient-Specific Assessment of Local Hemodynamic Predictors and Neointimal Hyperplasia Localisation in an Arteriovenous Graft

Abstract

Purpose: Most computational fluid dynamics (CFD) studies on arteriovenous grafts (AVGs) adopt idealised geometries and simplified boundary conditions (BCs), potentially resulting in misleading conclusions when attempting to predict neointimal hyperplasia (NIH) development. Moreover, they often analyse a limited range of hemodynamic indices, lack verification, and fail to link the graft-altered hemodynamics with follow-up data. This study develops a novel patient-specific CFD workflow for AVGs using pathophysiological BCs. It verifies the CFD results with patient medical data and assesses the co-localisation between CFD results and NIH regions at follow-up.

Methods: Contrast-enhanced computed tomography angiography images were used to segment the patient's AVG geometry. A uniform Doppler ultrasound (DUS)-derived velocity profile was imposed at the inlet, and three-element Windkessel models were applied at the arterial outlets of the domain. Transient, rigid-wall simulations were performed using the k-ω SST turbulence model. The CFD-derived flow waveform was compared with the patient's DUS image to ensure verification. Turbulent kinetic energy (TKE), helicity and near-wall hemodynamic descriptors were calculated and linked with regions presenting NIH from a 4-month follow-up fistulogram.

Results: In the analysed patient, areas presenting high TKE and balanced helical flow structures at baseline exhibit NIH growth at follow-up. Transverse wall shear stress index is a stronger predictor of NIH than other commonly analysed near-wall hemodynamic indices, since luminal areas subjected to high values greatly co-localise with observed areas of remodelling.

Conclusion: This patient-specific computational workflow for AVGs could be applied to a larger cohort to unravel the link between altered hemodynamics and NIH progression in vascular access.

Keywords: Arteriovenous graft; Computational fluid dynamics; Hemodynamics; Neointimal hyperplasia; Vascular access.

Fig. 1.

3D reconstructed patient-specific computational domain. The arrows indicate blood flow direction and distinguish between arterial (red) and venous (blue) pathways. The manually extracted patient’s DUS velocity waveform was applied as inlet boundary condition at the brachial artery. Three-element Windkessel models were implemented at the arterial outlets of the domain (AO1, AO2 and AO3), and their parameters (R_p, R_d and C) were calibrated through a lumped-parameter 0D model of the computational domain. Q_AO1, Q_AO2 and Q_AO3 refer to the flow rates through the respective outlets determined by their respective cross-sectional areas. A static pressure of 8 mmHg was imposed at the venous outlets (VO1 and VO2)

Fig. 2.

Flow waveforms derived from the DUS image (DUS-derived Q) and obtained from the CFD simulation (CFD-derived Q) at the proximal edge of the arterial stent

Fig. 3.

Velocity streamlines at peak systole in the whole domain and ROIs

Fig. 4.

Normalised velocity fluctuations of the streamwise component (u’), expressed as a percentage of the cycle-averaged inlet velocity (U_ref), at peak systole in the whole domain and ROIs

Fig. 5.

Volume rendering of TKE at peak systole for the whole domain and ROIs

Fig. 6.

Cycle-averaged LNH in the whole computational domain. Blue and red colours indicate left-handed and right-handed helical flow rotation, respectively

Fig. 7.

Identified critical areas—more prone to occlusion—below/above patient-specific thresholds (Table 2)

Fig. 8.

A 4-month follow-up fistulogram image of the cephalic vein region, and identified critical areas at the same location by the different near-wall hemodynamic indices. The black circle and arrows on the fistulogram show areas of occlusion at follow-up, characterised by a reduced vessel diameter compared to the surrounding vessel sections. These regions are also highlighted on the reconstructed cephalic vein region at baseline. The transWSS index best co-localised qualitatively with the observed occlusion