Twenty patients with persistent AF from Jiangsu Province Hospital were included in this study. The patients were divided into two groups based on their stroke history: non-stroke group (n = 10) and stroke group (n = 10). The inclusion criteria were patients with persistent atrial fibrillation who underwent cardiac CT imaging before AF ablation. The exclusion criteria were: (1) valvular heart disease; (2) non cardioembolic stroke; (3) incomplete left atrium models due to insufficient cardiac CT imaging qualities. All the patients were on the blood thinner and the cardiac CT examination was taken timely for the stroke group after their stroke. The studies involving human participants were reviewed and approved by the Southeast University Human Research Ethics Committee with all participants providing written informed consent.

Images were acquired by using CT systems (Somatom Force and Definition; Siemens) with an ECG-gated dual-source single-energy protocol over the entire heart. The acquisition parameters were as follows: retrospective ECG-gated protocol; automatic tube voltage adjustment (Care kV; Siemens) from 80 to 120 kVp; automatic tube current modulation ranging between 150 and 400 mAs. Beta-blockers and nitroglycerine were used if necessary. Images was obtained following the administration of contrast medium (370 mg iodine/mL, Ultravist, Bayer Schering Pharma, Berlin, Germany) via a 20-G intravenous catheter, followed by 40 mL saline flush using a dual-head power injector. The injection duration was determined by scan time plus 8 s and the flow rate ranging from 4.0 to 6.0 mL/s was tailored to the body weight, heart rate (HR) and tube voltage.

The LA geometry was segmented from the cardiac CT images using Mimics Medical 21.0 (Materialise, Inc., Leuven, Belgium) (see Figure 1), and then reconstructed with Geomagic studio (Geomagic, Inc., Morrisville, the United States). The morphological parameters were measured manually using Siemens NX (Siemens Digital Industries Software, Inc., Plano, the United States).

In the measurement, the volume and surface area of the LA and LAA were recorded (see Figure 2A) and so were the cross section of the LAA orifice and its long and short axes (see Figure 2B). Meanwhile, to evaluate the tortuosity of the LAA, the direct length (the distance between the LAA tip and the orifice center) and the actual depth (the distance along the centerline from orifice center to the LAA tip) were measured (see Figure 2C). The tortuosity was calculated based on the LAA direct length and actual depth:

For each patient, based on patient-specific 3D geometrical model, the pulmonary veins (PVs) were set as the velocity-inlet, and the mitral valve (MV) was the pressure-outlet. For the inlets, the velocity profile was derived from the measurement of pulmonary venous profile for a persistent-AF patient, and detailed in our former study (24) as:

where V is the flow velocity for the inlet, and t is the time in the cardiac time.

The open and close conditions of MV were set by the high- and low-pressure, which were expressed as:

The diastole is from 0 to 0.26 s, and the systole is from 0.26 to 0.7 s.

Tetrahedral-element meshes were generated by Ansys Meshing using a maximum element size of 0.8 mm and a minimum size of 0.2 mm, which were confirmed to be sufficient by performing a denser mesh.

The CFD analysis were performed on the node (2*Intel Xeon E5-2680 v3) in the High-Performance Computing Center of Collaborative Innovation Centre of Advanced Microstructures using the commercial software Fluent 2019 R3 (Ansys, Inc., Canonsburg, the United States).

In the simulation, the blood was set as incompressible Newtonian fluid with density ρ = 1,060 kg/m³ and viscosity μ = 0.0035 Pa⋅s. The flow was simulated by the realizable k-ε turbulence model with enhanced wall treatment, and the walls were assumed as rigid and with no-slip conditions to represent the worst AF symptom. The coupled second order upwind method was used for the simulations, and the time step was set as 0.001 s.

Meanwhile, to evaluate the hemodynamics in a more intuitive way, the species-transport model was employed to study the mechanisms of thrombus formation and the DPM analysis was performed to stratify the thrombus dynamics.

For the species-transport model, the mass fraction of each species was estimated through the convective-diffusion equation:

where Y_i is the mass fraction of the ith species, and J_i is the diffusive flux of the ith species, which is calculated as follows in turbulence:

where SC_t is the turbulent Schmidt number and D_{i, m} is the diffusion coefficient of the ith species.

For the DPM, the trajectory of discrete-phase particles is solved by integrating the particle force differential equation in the Laplace coordinate system. The force balance equation of particles in the Cartesian coordinate system is in calculated as follows:

where F_D(u−u_p) is the drag force per unit mass of particles:

where u is fluid velocity, u_p is particle velocity, μ is fluid dynamic viscosity, ρ is fluid density, ρ_p is particle density, d_p is particle diameter, Re is relative Reynolds number, and C_D is drag coefficient.

For the species-transport model, the initial blood in the LA was defined as one phase, which would be replaced by the fresh blood (defined as the other phase) flowing from the PVs. The residual rate of the initial blood was calculated, and the higher blood residual ratio corresponds to the more probability of thrombus formation in the LAA. In order to study the mechanism of thrombus flowing-out from the LAA, particles were set on the LAA walls at the beginning of the third cardiac cycle, the stable flow condition. The risk of thrombus flowing-out was represented by the residual ratio of particles in the LAA, and a higher particle residual ratio corresponds to a higher ratio of particles flowing out of the LAA, and thus a higher risk of stroke.

Statistical analysis was performed the open-source software Jamovi. The data were tested for normal distribution using Shapiro-Wilk test and the results were described as mean ± SD or median (IQR). Independent samples t-test or Mann-Whitney U test was used to compare the geometrical and hemodynamic parameters between the non-stroke group and stroke group. It was considered statistically significant when p-value < 0.05.

Patient characteristics are shown in Table 1. No significant (p > 0.05) differences existed in age, hypertension, diabetes and left ventricular ejection fraction (LVEF) between the two groups.

Morphological parameters of the LA and LAA are presented in Table 2. Both the actual depth and direct length of LAA were significantly larger in the non-stroke group than in the stroke group (p < 0.05). The surface area and tortuosity of LAA showed certain differences between the two groups but they were not statistically significant. There were no significant differences in the LA volume, LA surface area, LAA volume, LAA orifice long radius, LAA orifice short radius and LAA orifice ovality (the ratio of long radius to short radius).

The residual ratios of initial blood at the end of each cardiac cycle in the LA and LAA were recorded, which represent how efficient the blood clears out from LAA at each cardia cycle. An example of the simulation results of the blood residual ratio is shown in Figure 3A. The ratios were compared between the non-stroke and stroke groups in Table 3.

As shown in Table 3, from the third cycle, the residual ratios for the LA were significantly smaller in the stroke group than in the non-stroke group (p < 0.05). While for the LAA, the residual ratios were significantly smaller in the stroke group than in the non-stroke group (p < 0.05) from the first cycle.

Meanwhile, we also calculated the time when the residual ratios of the initial blood reduced to 20, 15, and 10%. No significant differences existed in the time parameters in the LA (p > 0.05) between the two groups, but the time parameters for LAA were significantly larger in the stroke group than in the non-stroke group (Table 3).

DPM analysis was used to quantify the efficiency of thrombus particles within LAA flowing out of LAA. Figure 3B shows an example of the particle distributions within LAA in a patient case. The residual ratio of the particles in LAA at each cardiac cycle was recorded, which are shown in Table 4. The particles were released at the third cardiac cycle, so the results were recorded from the third cycle. No significant (p > 0.05) differences existed in the residual ratios of particles in LAA between the two groups at the beginning three cycles, but the particle residual ratios became significantly smaller in the stroke groups than in the non-stroke group (p < 0.05) from the fourth cardiac cycle.