ASME Journal of Biomechanical Engineering
Abdominal aortic aneurysm endovascular repair: profiling post-implantation morphometry and hemodynamics with image-based computational fluid dynamics Paola Tasso Department of Mechanical and Aerospace Engineering Politecnico di Torino, Torino 10129, Italy
[email protected] Anastasios Raptis Laboratory for Vascular Simulations Institute of Vascular Diseases, Ioannina 45500, Greece
[email protected] Mitiadis Matsagkas Department of Vascular Surgery Faculty of Medicine, University of Thessaly, Larissa 41334, Greece
[email protected] Maurizio Lodi Rizzini Department of Mechanical and Aerospace Engineering Politecnico di Torino, Torino 10129, Italy
[email protected] Diego Gallo Department of Mechanical and Aerospace Engineering Politecnico di Torino, Torino 10129, Italy
[email protected] Michalis Xenos Department of Mathematics University of Ioannina, Ioannina 45500, Greece
[email protected] Umberto Morbiducci Department of Mechanical and Aerospace Engineering Politecnico di Torino, Torino 10129, Italy
[email protected]
1
ASME Journal of Biomechanical Engineering
ABSTRACT Endovascular aneurysm repair (EVAR) has disseminated rapidly as an alternative to open surgical repair for the treatment of abdominal aortic aneurysms (AAAs), because of its reduced invasiveness, low mortality and morbidity rate. The effectiveness of the endovascular devices used in EVAR is always at question as postoperative adverse events can lead to re-intervention or to a possible fatal scenario for the circulatory system. Motivated by the assessment of the risks related to thrombus formation, here the impact of two different commercial endovascular grafts on local hemodynamics is explored through 20 image-based computational hemodynamic models of EVAR-treated patients (N=10 per each endograft model). Hemodynamic features, susceptible to promote thrombus formation, such as flow separation and recirculation, are quantitatively assessed and compared with the local hemodynamics established in image-based infrarenal abdominal aortic models of healthy subjects (N=10). Moreover, the durability of endovascular devices is investigated analyzing the displacement forces acting on them. The hemodynamic analysis is complemented by a geometrical characterization of the EVAR-induced reshaping of the infrarenal abdominal aortic vascular region. The findings of this study indicate that: (1) the clinically observed propensity to thrombus formation in devices used in EVAR strategies can be explained in terms of local hemodynamics by means of image-based computational hemodynamics approach; (2) reportedly pro-thrombotic hemodynamic structures are strongly associated with the geometry of the aortoiliac tract postoperatively; (3) displacement forces are associated with cross-sectional area of the aortoiliac tract postoperatively. In perspective, our study suggests that future clinical follow up studies could include a geometric analysis of the region of the implant, monitoring shape variations that can lead to hemodynamic disturbances of clinical significance.
2
ASME Journal of Biomechanical Engineering
1
INTRODUCTION
2
Abdominal aortic aneurysm (AAA) is a vascular disease characterized by an enlargement of the
3
abdominal aorta lumen due to the loss of collagen and elastin in the wall [1]. As very recently
4
reported by the guidelines of the Society for Vascular Surgery, several issues need to be
5
considered carefully in the clinical management of AAA patients, such as (1) varying risks of
6
aneurysm rupture, (2) patient-specific factors influencing life expectancy, (3) need for
7
intervention, and (4) related operative risks [2]. In case of intervention, the guidelines suggest
8
careful attention to the choice of AAA operative strategy, along with appropriate post-intervention
9
surveillance, in order to minimize subsequent aneurysm-related death or morbidity [2].
10
Focusing on AAA treatment, endovascular aneurysm repair (EVAR), adopted for the first time in
11
1991, has disseminated rapidly as an alternative to open surgical repair of AAA [3], because of its
12
reduced invasiveness, low mortality and low morbidity rate [3-9]. There is clear evidence that, in
13
the United States, EVAR is being used with increasing frequency, with a decrease in associated
14
mortality [10].
15
Notwithstanding the marked evolution in both the technology and the assessment of long-term
16
outcomes, debate still exists about EVAR effectiveness [2, 11]. In fact, EVAR is associated with
17
postoperative complications that arise with variable frequency among the commercial stent-graft
18
systems, suggesting that a device-specific analysis could be more insightful [2, 12, 13]. Clinical
19
studies
20
occlusion/thrombosis as the most usual adverse events with inherent biomechanical triggers, such
21
as ill-directional displacement forces acting on the device structure and sub-optimal postoperative
22
hemodynamic environment [14-17]. In particular, the altered hemodynamics establishing inside
23
the endograft are suspected to contribute to thrombus formation immediately after the
highlight
the
stent-graft
migration,
endoleaks
(mainly
type
I)
and
limb
3
ASME Journal of Biomechanical Engineering
24
implantation or after months, potentially increasing the risk of an ischemic episode [18]. Very
25
recently, the Society for Vascular Surgery practice guidelines on the care of AAA patients indicated
26
the study of thrombosis in endograft as an area in need of further research [2].
27
Here, the impact of two different commercial endovascular devices on local hemodynamics is
28
explored through image-based computational simulations of blood flow in EVAR-treated patients
29
one month after implantation. Several hemodynamic features are identified, quantitatively
30
assessed and compared between postoperative and healthy infrarenal lumen structures derived
31
from medical imaging data reconstructions. The study is completed with the analysis of the
32
displacement forces acting on the implanted endografts, which are considered a primary factor
33
responsible for implants migration [19-23].In fact, as observed elsewhere, “displacement forces
34
are arguably the most important factor in determining the risk of device migration and future
35
complications” [20].
36
As geometry shapes the flow, the hemodynamic analysis is complemented by the geometrical
37
characterization of the EVAR-induced reshaping of the infrarenal abdominal aortic vascular region,
38
to mark out the impact of different endograft design features on local arterial morphology. The
39
overall objective was the identification of unique post-implantation morphological and
40
hemodynamic features marking up the endografts under examination, based on a sample of AAA
41
patients and heathy subjects.
42
MATERIALS AND METHODS
43
The schematic of the workflow applied in this study is provided in Fig. 1 and its parts are detailed
44
in the following subsections.
45
Study Subjects and Image Acquisition
4
ASME Journal of Biomechanical Engineering
46
Clinical data for this study were acquired at the University Hospital of Ioannina (Ioannina, Greece).
47
Twenty subjects affected by infrarenal fusiform AAA without extension of the disease to the
48
common iliac arteries were selected for this study. Subjects affected by AAA were divided into two
49
subgroups: ten patients (males, age = 73.9±8.41 years) were treated with the Endurant
50
(Medtronic Vascular, Santa Rosa, CA) stent graft, and the other ten (males, age = 68.6±8.62 years)
51
with the Excluder (W.L. Gore & Associates, Flagstaff, AZ) stent graft; both devices are fixed just
52
below the renal bifurcation and below the iliac bifurcation. As detailed elsewhere [15] the two
53
patient subgroups were statistically homogeneous in terms of (1) age, and (2) preoperative AAA
54
morphological features (in detail: neck length, neck diameter, suprarenal angle, infrarenal angle,
55
aortic length, right iliac length, left iliac length, right iliac diameter, left iliac diameter, neck
56
diameter to right iliac diameter, neck diameter to left iliac diameter). Physiologic ranges of the
57
considered hemodynamic and geometric variables are defined by considering ten healthy subjects
58
(males, age = 57.3±15.7 years).
59
Clinical images of healthy and repaired abdominal aortas and common iliac arteries were
60
extracted from computer tomography angiography (CTA) scans (Fig. 1), as detailed elsewhere [15,
61
24]. In detail, CTA scans were obtained with a 16-slices CT system (Brilliance CT 16 slice, Philips
62
Healthcare, Best, The Netherlands) with intravenous injection of contrast agent. Imaging
63
parameters included: 0.75-2.00 mm slice thickness, 0.78125/0.78125 pixel spacing, 120 KVp, 366
64
mA. On the twenty subjects affected by AAA, images were acquired one month after stent-graft
65
implantation. Healthy subjects had no sign of AAA and underwent CTA for other reasons. The
66
protocol was approved by the Institutional Review Board of the University of Ioannina, Ioannina,
67
Greece, and all the subjects gave informed consent for the use of their screening data.
68
Image Segmentation and 3D reconstruction
5
ASME Journal of Biomechanical Engineering
69
The two-dimensional DICOM images from CT scans were segmented using a semi-automatic
70
approach minimizing user interactions, and converted into three-dimensional models using the
71
software Mimics (Materialise, Leuven, Belgium) [15, 24]. Segmentation was performed on vessel
72
lumen for healthy arteries, and on the stent internal graft surface for EVAR treated arteries,
73
neglecting vessels and devices wall thickness. An overview of the reconstructed three-dimensional
74
geometries, divided in the three groups (i.e., healthy, Endurant- and Excluder-treated subjects), is
75
presented in Fig. 2.
76
The region of interest (ROI) was selected based on the position of the aortic segment where the
77
endograft is located. In detail, the radiopaque markers located at the opposite ends of the
78
endograft were used as guidelines. For healthy subjects, the rationale behind the ROI selection
79
was to consider the same aortic region as in treated patients. In detail, the ROI was determined
80
based on the indications from the clinical practice for the endograft landing zone, i.e. below the
81
renal bifurcation and, at least, 1.5 cm below the iliac bifurcation [25, 26]. This results in different
82
branches length for healthy or treated subjects, due to the more proximal position of the device
83
bifurcation with respect to the native iliac bifurcation.
84
Morphometric analysis
85
On the 30 reconstructed geometries, a centerline-based morphometric analysis was carried out, as
86
proposed elsewhere [27-29]. Technically, centerlines were extracted as the geometrical locus of
87
the centers of the maximum inscribed spheres in the model, as given by the Vascular Modelling
88
Toolkit software (VMTK, Orobix, Bergamo, Italy). Free-knots regression splines were then adopted
89
[30] as a basis of representation for a vessel’s centerline to provide a continuous, noise free
90
analytical formulation C with continuous derivatives. By differentiation of the free-knots
91
regression spline centerline representation C, curvature and torsion were calculated [28, 31]. In 6
ASME Journal of Biomechanical Engineering
92
detail, local curvature κ is defined as the reciprocal of the radius of the circle lying on the
93
osculating plane (identified by the normal and tangent vectors to the curve at that point), to
94
measure the rate of change in the tangent vector orientation along the curve. Local torsion τ
95
measures the deviation of curve C from the osculating plane. Technically, the curvature κ and
96
torsion τ of a curve C along the curvilinear abscissa s can be defined as: = =
| × | | |
[′ × ′′] ∙ ′′′ |′ × ′′|
97
where C’(s), C’’(s) and C’’’(s) are, respectively, first, second and third derivative of curve C with the
98
respect to the curvilinear abscissa s. Both vessel’s curvature and torsion of are known to have an
99
influence on arterial hemodynamics [32].
100
Local cross-sectional area A(s) of the vessel and its variation rate r(s), defined as the derivative of
101
A(s) with respect to s along the centerline C, were also computed using Matlab (Mathworks,
102
Natick, MA) [27, 28].
103
The centerline-based morphometry analysis of the infrarenal abdominal aortic bifurcation region
104
was divided into three segments (Fig. 3): body (identified by the centerline segment upstream of
105
the bifurcation point), left branch, and right branch (identified by the centerline segments
106
downstream the bifurcation point). Over the whole infrarenal abdominal aortic bifurcation region
107
and over each segment, average values (τmean, κmean, Amean, rmean), peak values (τpeak, κpeak,
108
Apeak, rpeak) and peak-to-peak (intended as the difference between maximum and minimum
109
values) amplitudes (τpp, κpp, App, rpp) of each geometric descriptor were computed [28].
110
Computational hemodynamics 7
ASME Journal of Biomechanical Engineering
111
In this study, blood was assumed to be an incompressible, isothermal, homogeneous, Newtonian
112
fluid (ρ = 1050 kg/m3; μ = 0.0035 Pa∙s) [24]. The governing equations of fluid motion, in the form:
!"
+ " ∙ $" + $ ∙ %μ$" + $"' ( + ∇* = + , - ∇∙"=0
113
where is the density, . the dynamic viscosity, " the velocity, + the body force and * the
114
pressure, were numerically solved by applying the finite volume method. To do that, the general
115
purpose CFD code Fluent (ANSYS Inc., Canonsburg, PA) was used on fluid domains discretized by
116
using tetrahedrons (with near-wall refinement obtained implementing a radial meshing approach
117
with elements density higher near the wall). A preliminary analysis of the sensitivity of the
118
numerical solution to the cardinality of the mesh was carried out. Steady state simulations were
119
performed on one model with 13 different mesh densities (ranging from 1 to 14 million
120
tetrahedrons). The satisfactory compromise between solution accuracy and computational costs
121
resulted in a mesh grid with elements size between 0.58 and 0.63 mm in the bulk and between
122
0.19 and 0.22 mm for near-wall elements. Therefore, mesh cardinality in the 30 models ranges
123
from 3.5 to 12.5 million.
124
As for the applied numerical scheme, second order accuracy was prescribed to solve both the
125
momentum equation and pressure with the SIMPLE pressure-velocity coupling scheme. The
126
backward Euler implicit scheme was adopted for time integration, with a fixed 0.0035 s time
127
increment. Convergence was achieved when the maximum mass and momentum residuals fell
128
below 10-4.
129
The strategy applied to prescribe the conditions at inflow and outflow boundaries is detailed
130
elsewhere [33]. Briefly, a Neumann condition was prescribed at the inflow boundary in terms of
131
time-dependent pressure waveform available from the literature [33]. At the outflow sections, 8
ASME Journal of Biomechanical Engineering
132
flow rate waveforms available from the literature and scaled according to the outflow sections
133
area were prescribed in terms of fully developed velocity profiles [15, 33]. To minimize the
134
influence from boundary conditions, flow extensions were added to the inlet and outlet faces of all
135
CFD models. Arterial walls were assumed to be rigid, with no-slip condition.
136
Near-wall and intravascular hemodynamic descriptors
137
Near-wall hemodynamics was explored by using the Time Averaged Wall Shear Stress (TAWSS),
138
i.e., the average value of the magnitude of the wall shear stress (WSS) vector τw along the cardiac
139
cycle: 1 ' TAWSS2 = 5 |67 2, 8|98 ; 4 :
140
where T is the duration of cardiac cycle and s the generic location at the vessel wall [27, 34]. Here
141
TAWSS was applied to identify low WSS regions, which in general correspond to low velocity
142
regions of flow separation, more prone to thrombus formation. In this study, we also considered
143
the TAWSS value averaged over the luminal surface S of a healthy or treated model (AWSS):
AWSS =
1 5 TAWSS2 9< > < =
144
Hemodynamic-related thrombus formation risks was additionally assessed by considering
145
recirculating flow, which has been demonstrated to correlate with thrombotic markers. To
146
quantify it, a modified version of the descriptor proposed by Martorell et al. [35]. was adopted to
147
calculate the volume of recirculating flow. Technically, the volume of recirculating flow was
148
computed by projecting the velocity vector field along the local vessel centerlines (i.e., axial
149
direction), and then integrating all finite volumes containing a negative axial velocity component.
150
In detail, the axial component ("AC ?@AB ) of the cycle-average velocity vector ("?@AB ) was calculated 9
ASME Journal of Biomechanical Engineering
151
according to the scheme proposed elsewhere [29]. Finite volume cells with negative "AC ?@AB value,
152
indicating average reverse flow, are markers of local recirculation. The Volume of Recirculation
153
(VolRec) was defined as the amount of fluid volume characterized a negative value of "AC ?@AB : L
DEFGHI = J δ KMN
1 if δk = 0 if
154
D O K K
(v ) (v ) ax
mean k ax
mean k
≤0 >0
155
Where Vk is the k-th finite volume cell, and N is the total number of finite volume cells of the fluid
156
domain. Here, the percentage VolRec value (%VolRec) for each model was calculated as follows: %DEFGHI =
DEFGHI ∙ 100 T 4E8QF DEFRSH
157
Eq. 7, allowed to use %VolRec for comparing the amount of blood recirculation among different
158
subjects, accounting for volumetric dimension intervariability of vascular segments.
159
Descriptors of the intravascular fluid structures such as helical flow were also applied, as it has
160
been observed that helical flow (1) has a physiological significance in arteries [36, 37], and (2)
161
smooths out WSS extremes [38-40]. The visualization of intravascular helical flow structures (for
162
the purposes of comparison) was performed by applying the Local Normalized Helicity (LNH): UVW =
"2, 8 ∙ X2, 8 = cos \ ] |"2, 8 ∙ X2, 8|
163
where ϕ is the angle between velocity and vorticity vectors [41]. By definition, the absolute value
164
of LNH ranges between one, when the flow is purely helical, and zero (in general) in presence of
165
reflectional symmetry in the flow. As reported elsewhere (see, e.g., [41]), LNH sign is an indicator 10
ASME Journal of Biomechanical Engineering
166
of the local right/left-handed direction of rotation of helical blood flow structures in the vessel.
167
Helical flow structures were characterized quantitatively by helicity-based hemodynamic
168
descriptors defined elsewhere [38, 39, 42], quantifying the (normalized to volume) average helicity
169
(h1), helicity intensity (h2), and helical flow topology (h3 and h4):
ℎN =
ℎ =
1 5 5 "2, 8 ∙ X2, 8 9D 98 a 4D ' `
1 5 5 |"2, 8 ∙ X2, 8| 9D 98 b 4D ' ` ℎ = ℎc =
ℎN
ℎ
|ℎN | ℎ
170
where T is the cardiac cycle, V is the vascular volume of interest, " the velocity vector, X the
171
vorticity vector and s the position in the fluid domain [38]. Briefly, eq. 9 and eq. 10 are non-
172
dimensional quantities and quantify the relative presence of counter-rotating helical structures. In
173
detail, h3 (-1 ≤ h3 ≤ 1) has by construction positive (negative) value when right-handed (left-
174
handed) helical structures are predominant in the fluid domain, while h4, defined as the absolute
175
value of h3, measures the strength of relative rotations of helical fluid structures in the fluid
176
domain, neglecting what is the major direction of rotation [38].
177
Displacement forces
178
The implanted endograft experiences a displacement force exerted by blood flow on its walls, and
179
its local value is given by the contribution of the pressure force and of the friction force. At each
180
time point along the cardiac cycle, the total displacement force (DF) acting over the entire wall of
11
ASME Journal of Biomechanical Engineering
181
the implanted device can be calculated by taking an area integral of the net pressure and of the
182
WSS vector [20]:
de8 = 5 * f 9< + 5 67 9< - =
=
183
where n is the unit vector locally normal to the wall and S is the surface area of the implanted
184
device. Here the quantity DFpeak, i.e., the magnitude of DF given by Eq. (13) calculated at peak
185
pressure time point at the inflow boundary, when the most elevated pressure and WSS values are
186
expected, was considered for the analysis.
187
Statistical analysis
188
To test for differences among healthy, Endurant or Excluder groups in quantities describing
189
geometric and hemodynamic features, the Mann-Whitney test was applied [43]. To identify
190
relationships among hemodynamic and geometric quantities, the Shapiro-Wilk test of normality
191
was preliminarily applied, due to the small sample investigated [44]. Pearson correlation
192
coefficients were then used if both variables were normally distributed, or Spearman correlation
193
coefficients in case one or both variables were not normally distributed. Regressions are reported
194
as the individual standardized correlation coefficient (β). For all analyses, significance was
195
assumed for p