Annals of Biomedical Engineering, Vol. 40, No. 10, October 2012 ( 2012) pp.

2228–2242
DOI: 10.1007/s10439-012-0579-3

Patient-Specific Multiscale Modeling of Blood Flow for Coronary Artery

Bypass Graft Surgery
SETHURAMAN SANKARAN,1 MAHDI ESMAILY MOGHADAM,1 ANDREW M. KAHN,2 ELAINE E. TSENG,3
JULIUS M. GUCCIONE,3 and ALISON L. MARSDEN1
1
Department of MAE, UCSD, 9500 Gilman Drive, La Jolla, CA 92093-0411, USA; 2Department of Medicine, UCSD, La Jolla,
CA, USA; and 3Department of Surgery, UCSF, San Francisco, CA, USA
(Received 24 January 2012; accepted 14 April 2012; published online 27 April 2012)

Associate Editor John H. Linehan oversaw the review of this article.

Abstract—We present a computational framework for mul- INTRODUCTION

tiscale modeling and simulation of blood flow in coronary
artery bypass graft (CABG) patients. Using this framework, Bypass graft (BG) surgeries require surgical con-
only CT and non-invasive clinical measurements are required struction of a conduit or graft over a blocked blood
without the need to assume pressure and/or flow waveforms
vessel. Coronary artery bypass graft (CABG) surgery
in the coronaries and we can capture global circulatory
dynamics. We demonstrate this methodology in a case study is usually performed when one or more coronary
of a patient with multiple CABGs. A patient-specific model arteries, that supply blood to the heart muscle, are
of the blood vessels is constructed from CT image data to fully or partially blocked due to the build up of ath-
include the aorta, aortic branch vessels (brachiocephalic erosclerotic plaque. Approximately 500,000 CABG
artery and carotids), the coronary arteries and multiple
surgeries are performed annually in the U.S. Coronary
bypass grafts. The rest of the circulatory system is modeled
using a lumped parameter network (LPN) 0 dimensional grafts are typically harvested from native tissue such as
(0D) system comprised of resistances, capacitors (compli- saphenous vein from the leg, left/right internal thoracic
ance), inductors (inertance), elastance and diodes (valves) artery or radial artery from the forearm, or harvested
that are tuned to match patient-specific clinical data. A finite from cadavers, or (rarely) composed of synthetic
element solver is used to compute blood flow and pressure in
materials such as poly tetra-fluoro ethylene (PTFE).
the 3D (3 dimensional) model, and this solver is implicitly
coupled to the 0D LPN code at all inlets and outlets. By Percutaneous coronary intervention (PCI) with angi-
systematically parameterizing the graft geometry, we evalu- oplasty (inserting and inflating a balloon in the coro-
ate the influence of graft shape on the local hemodynamics, nary artery) and stenting (inserting and expanding a
and global circulatory dynamics. Virtual manipulation of permanent mesh structure) are alternative treatments,
graft geometry is automated using Bezier splines and control
which may be chosen when progression of disease is
points along the pathlines. Using this framework, we
quantify wall shear stress, wall shear stress gradients and less severe, the patient is not a candidate for surgery, or
oscillatory shear index for different surgical geometries. We when indicated by other clinical factors.
also compare pressures, flow rates and ventricular pressure– The operative mortality rate for CABG is approxi-
volume loops pre- and post-bypass graft surgery. We observe mately 3%. Following surgery, the patency and lon-
that PV loops do not change significantly after CABG but
gevity of a CABG depends on multiple factors
that both coronary perfusion and local hemodynamic
parameters near the anastomosis region change substantially. including the type of conduit used, the local hemody-
Implications for future patient-specific optimization of namic environment, and structural properties of the
CABG are discussed. walls. It is known that arterial grafts such as internal
mammary artery (IMA) and radial artery grafts pro-
Keywords—Hemodynamics, Bypass graft, Wall shear stress, vide, in general, better patency than vein grafts.
Coronary artery, Multiscale modeling. Approximately 15–30% of saphenous vein grafts
occlude within the first year of surgery, with the rate
increasing to over 50% after 10 years.32 Therefore,
Address correspondence to Alison L. Marsden, Department of
typically arterial grafts are preferred. However,
MAE, UCSD, 9500 Gilman Drive, La Jolla, CA 92093-0411, USA. because multiple vessels are bypassed in most CABG
Electronic mail: amarsden@eng.ucsd.edu, amarsden@ucsd.edu surgeries, vein grafts are usually used as well.
2228
0090-6964/12/1000-2228/0  2012 Biomedical Engineering Society
 Multiscale Modeling of Blood Flow for CABG Surgery 2229

The dominant complications post-CABG are reste- Modeling flow in the coronary arteries is particu-
nosis36 and graft occlusion with associated myocardial larly challenging due to the influence of ventricular
infarction,5 and are attributed primarily to deposition contraction, making traditional boundary conditions
of atherosclerotic plaque,11,28 which may subsequently inappropriate. Sankaranarayanan et al.43 used con-
rupture leading to thrombosis, and associated hypo- stant pressure boundary conditions to study flow in an
perfusion. Graft incompatibility, and hemodynamic idealized CABG. This requires knowledge of patient-
factors such as blood recirculation, wall shear stress specific coronary pressure information, which is not
(WSS), and wall shear stress gradients (WSSG) play an usually available from standard clinical data. Pressure
important role in the onset and development of plaque boundary conditions are also limited in their capabil-
(atherogenesis).2,22,23,26,27,38,54 Because these factors ity, since they cannot be used to predict a post-surgical
are difficult to predict a priori, surgeons use intuition outcome. Dur et al.8 investigated surgical planning for
and geometric guidelines to make decisions about graft CABG through shape optimization, but their studies
design, despite the well known links between hemo- were limited to hemodynamic efficiency. Recently,
dynamics and restenosis. While mean flow and pres- Kim et al.20,21 developed an algorithm that models
sure waveforms can be obtained from invasive coronary circulation and microcirculation using a
coronary catheterization, local hemodynamic infor- lumped parameter network (LPN). In this work, a set
mation cannot be obtained reliably using currently of ODEs describing the coronary LPN was solved
available methods. Hence, patient-specific modeling analytically to prescribe the relation between flow and
and accurate numerical simulations provide a means to pressure at the coronary boundary faces. Inlet
obtain data on hemodynamics and WSS that is cur- boundary conditions were switched from Neumann to
rently unavailable.15,34,50 Dirichlet when the aortic valve was closed, and the
It has been shown experimentally (in vivo) that authors employed a penalty method assuming a para-
anastomosis angle effects flow fields47 and that flow bolic flow profile to prevent divergence due to back
patterns effect graft patency.19 Using three angles, flow. Here, we build on this previous work by
Rittgers et al.39 did not observe a significant correla- employing an implicit numerical approach to solve the
tion between angle and IH on external iliac arteries for coupled PDE/ODE system. Using a finite element
dogs. Using three in vivo patient-specific cases, Gior- framework (FEM),51 we simultaneously solve the
dana et al.12 showed that 80 is better than 30 (since coupled boundary condition problem, modeling the
platelet activation is reduced with increased flow mix- effect of heart behavior and contraction in a closed
ing) and the low angles studied are predisposed to loop network. This work was also motivated by the
occlude earlier. These in vivo do not present a stan- closed loop LPN models of Migliavacca and col-
dardized or holistic case study to recommend a specific leagues,18,24,33 which have been applied to model single
anastomosis angle and further experiments are still ventricle physiology. This technique generalizes the
needed.13 Earlier computational studies of the influ- computational framework for cases in which analytical
ence of CABG on mechanical environment have lar- solutions are not possible, and eliminates the need to
gely focussed on idealized geometrical models, switch between two different boundary conditions (i.e.,
simplifying physical assumptions, and/or a limited Dirichlet and Neumann) for the same surface.
trial-and-error approach to identify BG shapes that In this work, we propose a general cost function for
improve outcome.1 In our previous work, we studied CABG design, but point out that the choice of cost
the influence of anastomosis angles and graft radius on function can change with the acquisition of future
flow in idealized BG models41 demonstrating the effect experimental data. The use of a formal cost function is
of geometrical parameters on local hemodynamics. meant to set the stage for future work using optimi-
Using regions of low WSS as a measure of undesirable zation algorithms for shape design. A cost function can
(stagnant) flow patterns, we observed that these be chosen for BG design using WSS, oscillatory shear
regions correlated with three known locations of BG index (OSI), gradient oscillatory number (GON),
failure (toe, heel, and arterial floor region) and that aneurysmal flow index (AFI), or a combination of
they could be minimized by changing the graft anas- them. Our cost function is based on a simple linear
tomosis angles.31,41 We also showed that uncertainty combination of various risk factors for IH, and further
quantification42 influences the outcome of surgical experimental studies are needed to refine this choice.
optimization. Holzapfel et al.14 showed the interplay Though exact numerical values that correspond to
between BG shape, suture length, residual stresses and diseased states are not fully understood, experimental
the structural stresses induced by CABG. They showed studies can provide guidance on acceptable values.7,29
that small variations of the arterial incision have big WSS values between 4 and 15 dynes/cm2 are, in gen-
effects on the size of the arterial opening and thereby, eral, acceptable. In coronary arteries, WSS between 10
anastomosis angles and WSS. and 25 dynes/cm2 are considered normal.40 WSSG
2230 SANKARAN et al.

values of 400 dynes/cm2 or higher are considered 2. Build and mesh a 3D computational model of
abnormal leading to excessive cell proliferation. Values the aorta with major branch vessels, coronaries
of oscillation in WSS over 5 dynes/cm2 are undesir- and BGs
able, which roughly translates to values of OSI greater 3. Run coupled Navier–Stokes simulations using
than 0.25. Acceptable ranges for other derived vari- specialized lumped parameter boundary con-
ables, GON and AFI, have not yet been experimentally ditions, incorporating available clinical data
determined. We also develop tools to automatically such as blood pressure, heart rate, stroke vol-
post-process hemodynamic quantities and test differ- ume, and ventricular PV loop.
ent graft designs for the CABG surgery. 4. Post-process the solutions to evaluate quanti-
The three goals of this paper are (1) to present a ties of interest such as WSS, WSSG, and OSI.
computational framework that could be used (with
We describe each of these steps in the following
proper validation) in future work to predict coronary
sections.
flow based solely on non-invasive clinical measure-
ments, (2) to investigate the dependence of local
hemodynamics on anastomosis angle, and (3) to
compare the pre- and post-surgical flow conditions Patient-Specific Model Construction
using multiscale modeling. The paper is organized as We obtained post-operative anatomic data from a
follows. In ‘‘Patient-Specific Simulations of Coronary clinically indicated CT angiogram (CTA) of a 63 year
Flow’’ section, we provide details of patient-specific old male patient (with IRB approval) who underwent
geometric modeling. We then provide details of a triple BG surgery at UCSF (Fig. 1). The patient
boundary condition implementation and post- received three BGs: two native saphenous vein grafts
processing. In ‘‘Parameterizing Surgical Geometry’’ (one to the right coronary artery and one to an obtuse
section, we describe a spline-based method to param- marginal branch of the left circumflex artery) and one
eterize surgical geometry. Finally, we present patient- IMA graft (to the left anterior descending artery)
specific modeling results. (Fig. 2). The saphenous vein grafts originate from two
locations in the ascending aorta, distal to the aortic
valves. Blood flow was reinstated distal to two stenoses
PATIENT-SPECIFIC SIMULATIONS OF in the left coronary artery (one IMA and one saphe-
CORONARY FLOW nous vein) and one stenosis in the right coronary artery
(one saphenous vein).
In this section, we describe a general framework for
A 3D iso-contour of the pixel intensities obtained
hemodynamic simulations in CABG patients. The
from the image slices is shown in Fig. 2. Three steps
features of our modeling process, in the order per-
were used to reconstruct 3D geometric models from
formed, are listed below:
the CTA: (i) creation of centerline paths for vessels of
1. Obtain computed tomography scans for a interest including coronary arteries, aorta, brachioce-
patient who underwent CABG surgery phalic artery, IMA, etc. (ii) segmentation of the vessel

FIGURE 1. Two sections of the CT scan of a patient with multiple CABG and the constructed model (in red) running through the
slices.
 Multiscale Modeling of Blood Flow for CABG Surgery 2231

FIGURE 2. (Left) 3D isocontour plot of the major cardiovascular blood vessels and the internal mammary and saphenous vein
bypass grafts obtained from the CT scan and (right) a 3D reconstructed model of the aorta, major branching vessels, coronary
arteries and bypass grafts.

lumen through a combination of pixel intensity thres- assumed to be rigid, though we use capacitance ele-
holding and level set method, and (iii) lofting of the ments in each of the outlets to capture the global effect
segments to construct the 3D geometry. Models are of flexible walls. Newtonian constitutive behavior of
created using a customized version of the open sourced the fluid is assumed, with viscosity of blood set to
Simvascular software package (simtk.org),48,51 as 0.04 g/cm and density set to 1.06 g/cm3. A fourth
shown in Fig. 2. The reconstructed arteries are dis- order Runge–Kutta time stepping method is used for
cretized into tetrahedral elements using the commercial the 0D model, which is coupled to the 3D domain. We
software Meshsim (Simmetrix, Inc, Clifton Park, NY). use a recently developed backflow stabilization tech-
nique9 to prevent divergence of the numerical solution
in the presence of backflow. At the aortic inlet and all
Finite Element Modeling
the outlets of the model, a pressure (Neumann) BC is
The strong form of the governing Navier–Stokes prescribed, from the solution to a lumped parameter
equations is given by heart model, as explained in the next section. Pressures
  in the coronary LPN model are also connected via a
q~ v;t þ ~
v  5~
v  f ¼  5 p þ 5  s 8x 2 X
closed loop network to the heart model.
8x 2 @g X : vi ¼ gi
8x 2 @h X : pn þ sji nj ¼ hi ;
Multi-Scale Modeling
where X represents the 3D domain, q is density of We adapt a recently developed coupling framework
blood, ~
v is blood velocity, p is pressure, n is the normal (with applications in pediatric surgery18) to numeri-
to the vessel wall, s is stress on the wall, f represents cally couple the pressure and flow resulting from finite
body forces, t is time and x represents space. The element solutions at the inlet and outlet boundaries of
Neumann and Dirichlet boundaries are denoted by the 3D model to the 0D lumped parameter model.
@h X and @g X respectively, and values for traction and Using Neumann boundary conditions, pressure is
velocity are denoted by h and g, respectively. passed from the 0D to the 3D model and flow is passed
The corresponding weak form is given by from the 3D to the 0D model. Previous coupling
Z Z methods have used an explicit approach, but the semi-
 
w ~
q~ v;t þ ~
v  5~
v  f dX ¼ w ~  ð 5 p þ 5  sÞdX: implicit method that we use here10 is more robust and
has a less restrictive time-step choice. In the semi-
ð1Þ
implicit method, the tangent matrix in the finite ele-
A stabilized finite element technique using the gen- ment equation is calculated just once at the begin-
eralized-a method16,49 is employed with linear basis ning of the simulation, as opposed to a fully implicit
elements using an in-house custom solver. Walls are method where it is updated at each iteration. Numer-
2232 SANKARAN et al.

ical stability, accuracy and further details of the cou- TABLE 1. Parameters of the heart model in c.g.s. units.
pling technique using explicit, semi-implicit and fully
Component Rp Rd L C
implicit methods are described in Esmaily Moghadam
et al.10 ra-rv 5.0 – 1.0 –
The 0D LPN model includes blocks for the four- rv-pa 10 – 0.55 –
chambered heart model, the pulmonary arteries, the pa 12 108 – 0.005
la-lv 5.0 – 5.0 –
coronary model accounting for micro-coronary arter-
lv-aor 10.0 – 0.69
ies and coronary veins, and RCR circuits that model
the downstream resistances and capacitances of all
other outlets. The LPN circuit model is governed by a
set of ODEs that can be numerically solved, given  
1 Y1  Y2
boundary conditions imposed by the flow at the inlets Y01¼ Qcor 
Ca Ra þ Ral
and outlets of the 3D model.  
0 1 Y1  Y2 Y2  Pla
The heart model circuit has four one-way valves Y2 ¼  þ P0lv :
which are modeled using an indicator function, I ðxÞ Cim Ra þ Ral Rv þ Rvl
which is 0 if x £ 0, and 1 otherwise. The circuit also The governing ODEs at the RCR outlets are
contains resistors (P / @V@t ), capacitors (P  V), induc-  
2
tances (P / @@tV2 ) and elastance (P  f(V, t)), where P 0 1 Z1  Pla
Z1 ¼ Qbr 
and V represent pressure and volume of blood. For the C R
Runge–Kutta method, the derivatives of variables with Y2  Pla Z1  Pla
respect to time are used to update the variables at each Qt ¼ þ :
Rv þ Rvl R
time step.
The set of ODEs governing the closed circuit heart A schematic of the LPN is shown in Fig. 3. The Xs
model is presented below. The variables are denoted as refer to either flow rates, volumes, or pressures. In the
Xi and their time derivatives as X0i . Qt is the flow rate figure, X2, X4, X7, X9 are the flow variables denoted by
through the circuit, P is pressure, R is resistance, L is arrows, X1, X3, X6, X8 are variables corresponding to
inductance and C is capacitance. Subscripts ra, rv, ventricular or atrial volume, and the rest (X5, X10) are
la, lv, and p refer to the right atrium, right ventricle, pressures. Y is used for pressures at the coronary
left atrium, left ventricle and pulmonary systems, outlets and Z is used for pressures at all the other
respectively. Subscript pa represents pulmonary artery, outlets.
pp and pd represent pulmonary proximal and distal The pressures in the atria and ventricles are com-
resistances, Caorta refers to aortic capacitance, Qaorta puted using the relation Pi = Ei 9 (Vi 2 Vi,0) where i
refers to flow in the inlet to aorta, and Ei represents the denotes the chamber (la, lv, ra, or rv), and Vi,0 is the
elastance function which is different for each of the stress free volume.35 The resistances of the coronary
four chambers. The parameter values are given in arteries act against the coronary-ventricular pressure
Table 1. difference, while the resistances in the RCR blocks act
against the right atrial pressure (which is typically very
X01 ¼ Qt  X2 low). Further, the atrial and ventricular pressures at
X02 ¼ 1=Lrarv  ðPra  Prv  Rrarv X2 Þ each time step are calculated as a product of the elas-
X03 ¼ X2  X4 tance with the volume of the corresponding vessels.
  The net flow into the right atrium Qt is computed
X04 ¼ 1=Lrvpa  Prv  X5  ðRrvpa þ Rp pÞX4
  from the set of ODEs governing the LPN circuits. The
X5  Pla dPla coronary bed boundary conditions are comprised of a
X05 ¼ 1=Cp  X4  þ
Rpd dt coronary arterial circulation and microcirculation
X06 ¼ X4  X7 ð2Þ
(capillaries), and coronary venus circulation and
X07 ¼ 1=Llalv  ðPla  Plv  Rlalv X7 Þ microcirculation that drains the blood into sinus of
Valsalva, feeding into the right atrium (Fig. 3). The
X08 ¼ X7  X9 capacity of the coronary arteries to store blood
1 depends on the ventricular pressure (this is indicated
X09 ¼  ðPlv  X10  Rlvao X9 Þ
Llvaor by the line connecting coronary capacitance to left
1 ventricle). This coupling enables us to reproduce real-
X010 ¼  ðX9  Qaorta Þ
Caorta istic coronary waveforms, in which peak flow occurs
Xi ¼ X0i ¼ 0 if Xi <0; i ¼ 2; 4; 7; 9 during diastole, and minimum flow occurs during
systole. Traditional boundary conditions such as RCR
The governing ODEs at the coronary outlets are fail to accurately capture this behavior. Pressure vol-
 Multiscale Modeling of Blood Flow for CABG Surgery 2233

FIGURE 3. Schematic of coupling the 3D domain to the 0D lumped parameter model, including analytical RCR and coronary BCs
and heart model.

120
left ventricle 450 Q la-lv (cc/s)
right ventricle
100 Q ra-rv (cc/s)
400
Vlv (cc)
Pressure (mm Hg)

350 Vrv (cc)

80
300

60 250
200
40
150
100
20
50
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time (s) time (s)

FIGURE 4. (Left) Pressure tracings of the left and right ventricle and (right) atrio-ventricular flow rates and ventricular volumes,
obtained from the fourth order Runge–Kutta method.

ume curves, pressure tracings, volume and flowrate are conditions were achieved post surgery, and fit our
shown in Fig. 4. parameters accordingly. The elastance function for the
left and right ventricles are derived from the normal-
Parameters for the Heart Model ized curve of Senzaki et al.20,35,44 The values for other
parameters of the heart model are also determined
Since post-CABG clinical data was unavailable
from literature values for a normal adult patient.20 It is
for this patient, we assumed normal hemodynamic
2234 SANKARAN et al.

important to note that if data is available, our com- each branch,21 we have Ri = k/Ai where k = RbrAbr,total.
putational framework has the capacity to incorporate The resistance of each branch is then Rp,i = kp 9 Ri and
it, either iteratively or through a rigorous optimization Rd,i = kd 9 Ri, where subscripts p and d refer to proxi-
procedure. mal and distal respectively. The constants kp and kd are
0.09 and 0.91, respectively.20 The RCR values for each of
the outlets are given in Table 2.
Algorithm for Choosing Boundary Condition
At the coronary outlets, we split the resistance for
Parameters
the left and right coronaries to match literature data.17
P 
The total resistance for all the outlets is Rtotal ¼ Q : Given a flow split c between left and right coronaries,
This is split into the total resistance for the coronary we have Rright left left right
cor = cRcor, while Qcor = cQcor . We
outlets (cor) and the rest (br). Since the circuits are in choose c = 7/3 to enforce a typical 70–30% flow split
1þc
1
parallel, we have Rtotal ¼ R1cor þ R1br : We prescribe the between LCA and RCA.17 Hence, Rleft cor ¼ c Rcor and
right
average flow split between the coronaries and the Rcor = (1 + c)Rcor. The resistance for each branch
aorta. Let this fraction be represented as b, i.e., is then Rcor = Ra,cor + Ra-l,cor + Rv-l,cor + Rv. We
Qcor = bQbr. The net coronary and branching resis- choose typical values for splitting the coronary resis-
tances are then shown to be Rcor ¼ 1þb b Rtotal and tance,6,20 Ra,cor = 0.32Rcor, Ra-l,cor = 0.52Rcor and
Rbr = (1 + b)Rtotal. In this work, b was chosen to be Rv + Rv-l = 0.16Rcor. The values of the coronary BC
0.04 based on typical literature values.4 Capacitance parameters are given in Table 3.
does not affect the mean flow and is tuned to adjust the The capacitance values for each outlet are chosen to
range and variation of the pressure values. The be proportional to their area. For each outlet, the
capacitance values and flow rate are chosen to be capacitance value is split according to typical values
proportional to the outlet areas,53 Ci  Ai. We first fix Ca = 0.11Caorta and Cim = 0.89Caorta.
the capacitance of the aortic outlet and then compute
the proportionality constant to fix the capacitance of
Computation of Hemodynamic Parameters
all other outlets. The capacitance of the aortic outlet is
iterated on, using trial and error, to match the typical In this section, we explain the methodology used to
blood pressure of 120/80. For instance, a higher compute the WSS and WSSG.
capacitance value will shift the peak towards the right.
The net resistance of the branches, Rbr, is split for each Consistent Flux Calculation for Computing WSS
outlets as follows. Given the net resistance and area of
Wall shear stress is computed in a post-processing
step from the velocity (~ v) and pressure (~
p) fields of the
TABLE 2. The proximal resistance, capacitance and distal Navier–Stokes solution. Rearranging the weak form in
resistances (Rp, C and Rd) for each of the RCR (non-coronary) Eq. P (1), with velocity
outlets in c.g.s. units. P and pressure solutions
v~i ¼ A NA v~iA and p~ ¼ A NA p~; we have:
Outlets Area (cm2) Rp (9103) C (91026) Rd (91026) Z Z
 
~  tds ¼ w
w ~  q v~;t þ v~  5~
v  f dX
O1 0.86 0.64 278.66 6.44
O2 0.28 1.93 92.10 19.50 Cg
Z Z
O3 0.36 1.53 116.25 15.45
O4 0.10 5.50 32.46 55.32 þ 5~
w : ð~
pI þ sÞdX  ~  tds
w ð3Þ
O5 0.07 7.86 22.11 81.24 Ch
O6 0.05 11.00 16.98 105.78
O7 0.02 27.52 6.46 278.00 The weighting function, w ~; is non-zero only at the
O8 0.27 2.04 88.43 20.31 surfaces where the WSS is to be computed. Choosing w
~
O9 4.74 0.12 1540.00 1.17
to be the surface shape functions, we solve for the

TABLE 3. The arterial, micro-arterial and venal resistances (Ra, Ra-l, Rv + Rv-l) and arterial and intramyocardial capacitance
(Ca and Cim respectively) for the coronary outlets in c.g.s. units.

Outlets Area (cm2) Ra (9103) Ra-l (9103) Rv + Rv-l (9103) Ca (91026) Cim (91026)

O10 0.15 35.81 58.63 18.53 0.37 3.15

O11 0.04 137.87 225.72 71.33 0.09 0.818
O12 0.01 435.62 713.21 225.37 0.03 0.26
O13 0.04 124.28 203.48 64.30 0.11 0.91
O14 0.06 89.15 145.95 46.12 0.15 1.27
O15 0.32 16.25 26.61 8.41 0.82 6.94
 Multiscale Modeling of Blood Flow for CABG Surgery 2235

boundary tractions. Two additional quantities of RT

interest are derived from the WSS solution, s(x, t). j 0 Gdtj
GON ¼ 1  R T ð6Þ
First, OSI is defined as
0 jGjdt
" R Tc #
0 sðx; tÞdt where G ¼ 5si ssi :
OSIðxÞ ¼ 0:5 1  R Tc
Details of how the GON is computed are given in
0 jsðx; tÞjdt
the Appendix.
and is an indicator of the directionality of flow. OSI
has minimum value of 0, corresponding to uni-
directional traction, and maximum value of 0.5, cor- PARAMETERIZING SURGICAL GEOMETRY
responding to zero mean traction with equal time
spent in both directions. Second, we compute the We parameterize the surgical graft shape using cubic
AFI,30 originally developed to measure likelihood of splines. There are three BGs in the patient. Initially, we
aneurysm formation, by quantifying the temporal identify the center pathline of each vessel of interest
fluctuations of the direction of WSS. The AFI is de- and identify a set of control points along the pathline
fined as the instantaneous projection of the wall shear (7 for this patient). These pathlines are parameterized
direction on the mean wall shear direction. Previous by a variable t which varies from 0 to 1 along the
studies support the relationship between endothelial length of the path. Piecewise cubic splines were chosen
response and temporal fluctuations in WSS.37 Since it to interpolate the paths, after second order polyno-
is well known that endothelial cells orient in the mials were found to be inadequate. To compute the
direction of mean WSS,30 the AFI quantifies the time coefficients, the spline is made to pass through all the
averaged tensile/compressive force that endothelial control points and the slope is made continuous at
cells are subject to. the control points. The number of equations is one less
than the number of control points.
Wall Shear Stress Gradients Different BG shapes are obtained by manipulating
Wall shear stress gradients quantify the spatial varia- the location of the control points. Let r0 represent the
tions in hemodynamic forces on the vessel wall.25,37 location of the control point for the original BG. We
WSSG quantifies the magnitude of shear-stress gradients specify a maximum deviation dm as well as a direction
using the average wall shear direction and its in-plane n and define the possible new shapes of the BG as
(plane of the wall) normal. Off-diagonal coupled- r ¼ r0 þ kdm n where k is between 21 and 1 and rep-
derivatives, such as gradient of mean wall shear in a resents the extremes of the BG shape. The direction of
direction normal to the wall, are neglected since they have motion n is chosen perpendicular to the pathlines (at
been shown to have little clinical relevance.25 The for- the control point) in an arbitrary plane.
mulation for computation of WSSG is given below. Let t be the parameter that represents distance along a
From the velocities, tractions are defined as ti = saphenous vein graft and s be the parameter that repre-
rijnj and shear stresses as si (x, t) = ti 2 (tk nk) ni. Let sents distance along a coronary artery. The anastomosis
0 0
s
s1(x) be a unit vector in the direction of mean wall angle between these arteries is given by cosðhÞ ¼ kt0tkks 0k
@x ðtÞ @x ðtÞ @x ðtÞ
shear traction and let s2(x) represent a direction where t0 ¼ @t1 i þ @t2 j þ @t3 k represents the tangent
orthogonal to s1 in the plane of the wall (i.e., nw is the to the centerline path of the saphenous vein and s0 ; which
normal to the wall, s2 = nw 9 s1). The WSS is is defined similarly, represents tangent to the coronary
decomposed into two components, ðss1 ; ss2 Þ ¼ artery. A different anastomosis angle can be tested by
ðs  s1 ; s  s2 Þ and the WSSG is defined as25,38: modifying the control points. The endpoints are fixed,
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and if the control point closest to the anastomosis is
u 
u 2  2 ! moved by a value k normal to the spline, the rest of the
@ @
sG ðx; tÞ  t ss þ ss ð4Þ control points move by xk ðtÞ ¼ kk  tmax t
where kk is a
@s1 1 @s2 2 vector normal to the cubic spline (i.e.) kk = k 9 nk, where
nk is the normal to the spline. A plot of k vs. anastomosis
where the individual terms are computed as angle h is shown in Fig. 5. The plot shows that a full range
@
@si ssi ¼ si  rssi ; i ¼ 1; 2: Finally, the mean WSS gra- of realistic anastomosis angles can be obtained by this
dient is computed as: parameterization.
ZT For the patient-specific model, parameterization of
1 the BGs is shown in Fig. 5. A set of seven control
sG ðxÞ ¼ sG ðx; tÞdt ð5Þ
T points is used for each of the BGs. Different anasto-
0
mosis angles are then obtained by manipulating the
45
The GON is defined from the WSSG as: location of these control points. We automated the
2236 SANKARAN et al.

110
IMA
100 Saphenous
RCA
90 Saphenous
LCA

80

70

θ
60

50

40

30

20

10
-1 -0.5 0 0.5 1
λ

FIGURE 5. (Left) Five different bypass grafts for different values of the parameter k corresponding to (from left)
21, 20.5, 0, 0.5, 1, and (right) a plot showing how different anastomosis angles can be achieved by changing the control points (k)
of the saphenous graft from 21 to 1.

process of building a geometry given a trial anasto- 1

Normalized flow
mosis angle. Through a simple inversion process, we Normalized pressure
0.9 Normalized energy
obtain the location of the control points for a given Normalized value
trial anastomosis angle. This is then converted into a 0.8
new pathline, and segmentations constructed from CT 0.7
are translated accordingly. These segmentations are
then lofted together and blended to form the new BG 0.6

geometry. 0.5

0.4
RESULTS
0.3
800,000 2,000,000 4,000,000
Convergence Test
number of elements
To ensure convergence, we performed mesh-
independence and time-step independence tests. We FIGURE 6. Convergence plots for normalized (outlet aver-
aged) velocity, pressure and energy with computational
gradually refined the mesh density and compared meshes of size 800,000, 2,000,000, and 4,000,000.
pressure, velocity, energy, and WSS. Figure 6 depicts
the change in simulation results with mesh density. We
conclude that the mesh density with 4 million elements
Coronary Flow and Pressure Profiles
is sufficient for our simulations.
A patient-specific measured heart-rate of 61 beats The multiscale model results in a coronary flow
per minute is used for the cardiac cycle. We used a time peak at diastole, agreeing with clinical observation, as
step size of 0.5 ms for the multiscale model to ensure expected. The mean flow ratio matches the ratio of the
convergence. We used 4 Newton–Raphson lineariza- prescribed resistance between the coronary and aortic
tion iterations per time step for all the models. We circulation, as expected. The flows were all measured
simulate flow in the first model geometry for 10 cardiac distal to the anastomosis and branching locations in
cycles to ensure convergence, and then perform sub- five arteries in the left coronary system (one diagonal,
sequent simulations for other geometries using the first LAD after diagonal bifurcation, two obtuse marginals,
case as the initial condition. Only 3–4 cardiac cycles are LCX after second obtuse marginal bifurcation) and
subsequently needed for the results to stabilize. When one artery in the right coronary system. We iterated
we restart the simulations, we use 5 Newton–Raphson the total capacitance of the coronary circuit to match
iterations for the first 100 timesteps and then con- the position and ratio of pressure peaks to match
tinue with 4 Newton–Raphson non-linear iterations, to typical clinical data. An LCA-RCA flow split of
ensure convergence. 70–30% was achieved, as shown in Fig. 7.
 Multiscale Modeling of Blood Flow for CABG Surgery 2237

Homeostatic States Before and After Surgery of ~4.71. In the aorta, maximum Reynold’s number is
~3200 and Womersley number is around ~12.53. As
Using the coupled multiscale model, we compute
shown in Fig. 9, coronary flow is partially restored
the flow conditions pre- and post-CABG surgery and
after the CABG is performed. Prior to CABG surgery,
compare quantities of interests, keeping all LPN
we assume that the bypassed right coronary artery and
boundary condition parameters fixed. The coronary
left obtuse marginal have a 75% stenosis and create the
arteries have a mean and maximum Reynold’s number
corresponding model keeping the rest of the patient-
of ~424 and ~256 respectively, and Womersley nomber
specific geometry intact. This changes the effective
hemodynamic resistance of the model. We then mea-
6
sure how the cardiac output and other quantities of
Qtotal
interest in the LPN and the 3D model change with
5
QLCA
inclusion of BG resistance.
4 QRCA Pressure-volume curves of the left and right ventri-
flow rate (ml/s)

cle before and after the surgery are compared in Fig. 8.

3
As expected, the cardiac output is not significantly
2 increased by the reduction of hemodynamic resistance
1
induced by the BG. There are two factors that con-
tribute to this. First, the coronary flow is a very small
0 fraction of the full aortic flow, and second, the model
−1 resistance is dominated by the rest of the circuit,
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 implying that addition of the BG results in insignificant
time (s) changes to the net resistance.
FIGURE 7. Coronary flow profile shows a peak in the dias- We compare the coronary perfusion (measured
tole with a flow split of 70–30% LCA to RCA. distal to the BG) before and after CABG surgery in

140 140
right ventricle
120 right ventricle 120 left ventricle
left ventricle
100 100
P (mm Hg)

P (mm Hg)

80 80

60 60

40 40

20 20

0 0
20 40 60 80 100 120 140 160 180 20 40 60 80 100 120 140 160 180
V (cc) V (cc)

FIGURE 8. Comparison of pressure–volume loops for the left and right ventricle before (left) and after surgery (right).

1.6
6
1.4
Coronary flow rate (ml/s)

5 pre surgery
1.2 post surgery
flow rate (ml/s)

1 4

0.8 3
0.6 RCx after CABG
LAD after CABG
2
0.4 75% stenosed RCx
75% stenosed LAD 1
0.2
0 0

-0.2 -1
0 0.125 0.25 0.375 0.5 0.625 0.75 0.875 1.0 0 0.25 0.5 0.75 1.0
time (s) time (s)

FIGURE 9. A comparison of (left) coronary artery flow rate in the stenosed arteries and (right) net coronary perfusion, before and
after CABG measured distal to the bypass grafts.
2238 SANKARAN et al.

Fig. 9. After CABG surgery, blood is drawn through value, we start to see an effect on coronary flow since
the BG which offers lower resistance than the stenosed the resistance varies with the radius.4
coronary arteries. The figure clearly shows an
improvement in the coronary flow-rate, and hence
Influence of Shape on Quantities of Interest
perfusion, after the BG is inserted in the two stenosed
coronary arteries. We also confirm the diastolic flow Figure 5 shows the shape of the BG for values of
peak in the coronary arteries both pre- and post- h = 70, 62, 50, 35, 10. A wide range of physically
surgery. Prior to implanting the BG, the model resis- realistic anastomosis angles is achieved with this
tance of the stenosed coronary arteries dominates the parameterization. In this section, we restrict our dis-
downstream boundary conditions. After the surgery, cussion to the parameterization of the saphenous vein
the BGs impose a lower resistance to flow and hence BG, which is anastomosed to the left circumflex, LCx.
provides a conduit for almost all coronary flow in the Volume renderings of the velocity magnitude at
stenosed arteries. The flow in the remaining coronaries diastole for three BG shapes, h = 70, 50, 10 are
is unaffected. shown in Fig. 10. Coronary flow is maximum in dias-
In addition, we changed the BG radius to half its tole. The figure shows that there is a change in the
original value, keeping the rest of the geometry intact. aortic flow resulting from different bypass shapes,
We observed that changing the graft radius in this while the flow in the non-stenosed coronary arteries
range had little effect on the cardiac output or coro- remains almost the same. Figure 11 shows the differ-
nary flow. This is because downstream coronary artery ence in flow characteristics for different graft angles at
resistance is still much greater than the BG, which has the anastomosis region. A lower anastomosis angle
a very small contribution to the overall coronary implies a longer suture region, and hence we observe
resistance. However, dropping the radius below this that (a) a higher bypass angle leads to rapid flow

FIGURE 10. Three-dimensional volume rendered plots of magnitude of blood velocity for three different shapes of the saphenous
vein bypass graft corresponding to h 5 70°, 50°, 10°.

FIGURE 11. Plot shows the velocity profile at two planes (shown in d) for three different bypass grafts corresponding to k 5 21, 0, 1
(angles 70°, 50°, and 10°) and the increase in blood flow velocity (hence volume flow rate) resulting from the bypass. Since each
anastomosis angle corresponds to a different length of suture, the anastomosis area is much larger in some configurations.
 Multiscale Modeling of Blood Flow for CABG Surgery 2239

FIGURE 12. Comparison of (top) WSS and (bottom) OSI for five different bypass graft shapes corresponding to anastomosis
angles of (from left) 70°, 62°, 50°, 35°, and 10°.

1 (h = 70) has the optimal configuration when consid-

ering area of low WSS. This is because the arterial
0.95 floor region contributes most to the total area of low
WSS, and low WSS in this region is minimized in
Cost function

0.9 designs with higher anastomosis angle. The trend in

ALWSS GON is the result of two competing factors: a higher
0.85
OSI anastomosis angle results in (1) a higher contribution
GON
AFI
to the GON via WSSG, and (2) a lower contribution to
0.8 Cost function the GON via variance in the WSS direction. The plot
shows that the latter term is dominant, resulting in the
0.75 optimal GON at h = 70. If we consider a cost func-
10 20 30 40 50 60 70
tion defined as J ¼ 0:25  ðAðs  1Þ þ OSI þ GON þ
θ
AFIÞ; we conclude from Fig. 13 that h = 70 is the
FIGURE 13. Figure plots different objective functions in the optimal configuration. The first term indicates the area
parameter space. It depicts that the optimal shape parameter of the lumen wall that is subject to low shear stress.
for different quantities of interest as well as the cost function.
k 5 0 represents the geometry corresponding to the actual The appropriateness of this choice of cost function and
surgery performed. determination of whether the cost function and
resulting optimal angle should be patient-specific will
mixing, and (b) low BG angles ensure a streamlined be the subject of future research. The optimal bypass
mixing of blood at the location of the anastomosis and shape ultimately ought to involve a detailed combi-
larger areas of low WSS. Plots of WSS and OSI for five natorial optimization involving several different clini-
different BG shapes are shown in Fig. 12. The WSS cally motivated cost functions and constraints.
plots reinforces earlier observations27,28 showing that
regions of critical hemodynamic importance in BGs
are the toe, heel, and arterial floor regions. High
Sensitivity to Flow Split
anastomosis angle results in reduced energy efficiency
and larger flow separation regions. Lower anastomosis To test the sensitivity of simulation outcomes to the
angles, however, result in higher areas of low WSS at total coronary flow (which we assumed was 4% of the
the heel and toe regions of the BG. aortic flow), we performed two additional sets of
Figure 13 shows the variation of different quantities simulations with total coronary flow set to 3% and 5%
of interest with the BG parameterization. All quanti- of the aortic flow. The resistances were adjusted
ties are normalized to a value of 1 based on the max- accordingly, and the total flow rate was maintained.
imum value across the five designs. Areas of low wall We observed that the coronary flow profile shifted in
shear, OSI and GON have the same minima corre- magnitude by ±25% keeping the shape of the flow
sponding to h = 70 (an anastomosis angle of roughly waveform shape the same. However, we observed a
70). Overall, OSI and AFI are the least sensitive with change of less than 1% in the cost function values since
respect to shape. If we consider the sum of heel, toe the cost function values are generally normalized by
and anastomosis floor regions, the higher impact angle their mean values.
2240 SANKARAN et al.

CONCLUSIONS 70 would be hemodynamically optimal for this

patient. Hence, our results provide further evidence
We expanded upon an implicitly coupled multiscale that performing patient-specific optimization on the
framework to perform CFD simulations in a patient surgical geometry should be considered and could lead
who underwent CABG. The described method pro- to improved longevity of the graft.
vides a means to perform realistic simulations of flow Limitations of this work include the use of rigid
in the coronary arteries and BGs by modeling the walls, Newtonian flow, and assumptions of coronary
entire circulatory system as a closed loop. This tech- flow distribution. In a recent study,46 the difference in
nique is useful for virtual surgical parameterization in WSS between FSI and rigid wall simulations in coro-
which the surgical geometry could affect the boundary nary arteries was found to be less than 5%, however
conditions, and this is implicitly captured by the de- studies in other arteries have found larger differences.3
scribed technique. When available, clinical data was The physical location of the surgery prevented us from
incorporated into our computational framework and exploring higher anastomosis angles without adding
nominal literature-based parameter values were additional geometric parameters. It remains to be seen
assumed otherwise. how the choice of cost function presented here impacts
We used cubic splines with seven control points to long term clinical outcomes, or if different patients
parameterize the BG shapes, and this method could need different weighting functions based on a priori
potentially be used in the parameterization of other clinical factors. In the future, we plan to perform a
surgical procedures, or in a formal optimization pro- detailed study of flow in coronary arteries in multiple
cedure. We are able to virtually manipulate surgical patients and perform statistical analyses to test the
shape, anastomosis angles and correlate hemodynam- generality of our observations. We also plan to per-
ics with different surgical shapes. form patient-specific optimization on different patients
Our framework reproduces the clinically observed to test whether a one-size-fits-all solution is sufficient
coronary flow peak during diastole and also incorpo- or if different patients should have different optimal
rates data such as flow split between coronary arteries BG shapes.52 Further validation with post surgical
and aorta, when available. We observe 3D flow profiles clinical data is needed. The effects of uncertainty in
at the anastomotic region for high anastomosis angles. simulation inputs on their predictive capability should
We observed that performing the BG surgery increases also be analyzed using formal uncertainty quantifica-
coronary perfusion, as expected. The increase in flow tion tools.42 We also plan to incorporate the effects of
rate in the stenosed arteries is observed to be as much exercise and autoregulative mechanisms of the heart.
as 200%. While previous studies have also examined
the effect of geometry on BG hemodynamics,8,43 this
was the first to do it using only CT and non-invasive APPENDIX
clinical measurements, without the need to assume
pressure and/or flow waveforms in the coronaries. Here, we provide details on computing the WSS
We observed that local variables such as WSS and gradients used in this study. For each element e, let the
WSSG are affected by the BG shape, and different cost normal be denoted by ne and the mean shear stress be
functions have different optima. Local velocities in the denoted by se ¼ k s k se1 : The direction se2 is computed
e se1 ne
aorta are also observed to be affected by changes in BG as s2 ¼ kse ne k : The components of elemental shear
1
shape. Global quantities such as energy efficiency stress in these two directions are computed by taking a
and PV loops are unaffected by the resistance of the dot product with the directions. We can compute the
inserted BG and its shape. In patients with impaired gradients as follows:
left ventricular systolic function prior to surgery, these
5si sesi ¼ si  5se  si : ð7Þ
quantities will also depend on the ventricular con-
tractility, which often improves after BG surgery. Our The gradient of the shear stresses are evaluated at each
computational framework has the potential to model tetrahedral element as:
this through a change in elastance function, whenever
such data is available. X
3
5sei;j ¼ NA;j siA : ð8Þ
We conclude that BG shape has a strong influence A¼1
on local quantities of interest that would potentially
affect patient longevity, and that different hemody- The corresponding MWSSG is represented as sG;e : We
namic cost functions have different optima. Normal- then derive the nodal values as
izing each cost function to a value of 1 and assuming Z Z
that each cost function is weighted equally, we observe s ðxÞwðxÞdX ¼ sG;e wðxÞdX:
G ð9Þ
that for the range considered, an anastomosis angle of X X
 Multiscale Modeling of Blood Flow for CABG Surgery 2241

intramyocardial compliance on the basis of the waterfall

Discretizing this equation and choosing the test func- model. Ann. Biomed. Eng. 13(5):385–404, 1985.
tions as shape functions, we have 7
DePaola, N., M. A. Gimbrone, Jr., P. F. Davies, and C. F.
Dewey, Jr. Vascular endothelium responds to fluid shear
Anel G nel
e¼1 ðNA ; NB ÞsB ¼ Ae¼1 ðNA ; 1Þs
G;e ; ð10Þ stress gradients. Arter. Thromb. Vasc. Biol. 12:1254–1257,
1992.
where A denotes the assembly matrix. Integrating this 8
Dur, O., S. Coskun, K. Coskun, D. Frakes, L. Kara, and
equation, we observe that: K. Pekkan. Computer-aided patient-specific coronary
0 10 G 1 0 1 artery graft design improvements using CFD coupled
2 1 1 s1 1 shape optimizer. Cardiovasc. Eng. Technol. 2(1):35–47,
jJ j
e @ AB GC
jJ j
e @ A G;e 2011.
Anel
e¼1 1 2 1 @ s 2 A ¼ A nel
e¼1 1 s : 9
24 6 Esmaily Moghadam, M., Y. Bazilevs, T.-Y. Hsia, I.
1 1 2 s3G 1 Vignon-Clementel, A. Marsden, and Modeling of Con-
genital Hearts Alliance (MOCHA). A comparison of outlet
ð11Þ
boundary treatments for prevention of backflow diver-
This equation is found from the least squares method gence with relevance to blood flow simulations. Comput.
Mech. 48(3):277–291, 2011.
(minimizing the difference between sG and sG,e). For 10
Esmaily Moghadam, M., I. Vignon-Clementel, R. Figliola,
the purpose of visualization, we approximate the LHS and A. Marsden. A modular numerical method for implicit
system to choose nodal values that are weighted 0D/3D coupling in cardiovascular finite element simula-
averages of the elemental solutions. tions. J. Comput. Phys., 2011, in review.
11
Giddens, D., C. K. Zarins, and S. Glagov. The role of fluid
mechanics in localization and detection of atherosclerosis.
ACKNOWLEDGMENTS J. Biomech. Eng. 115:588–594, 1993.
12
Giordana, S., S. Sherwin, J. Pelro, D. Doorly, J. Crane,
The authors acknowledge funding from an Ameri- K. Lee, N. Cheshire, and C. Caro. Local and global geo-
can Heart Association postdoctoral fellowship, the metric influence on steady flow in distal anastomoses of
Leducq Foundation, a Burroughs Wellcome Fund peripheral bypass grafts. J. Biomech. Eng. 127:1087–1098,
2005.
Career Award at the Scientific Interface, and NIH 13
Haruguchi, H., and S. Teraoka. Intimal hyperplasia and
grant RHL102596A for funding this work. We thank hemodynamic factors in arterial bypass and arteriovenous
Hyun Jin Kim for expertise on coronary modeling, and grafts: a review. J. Artifi. Organs 6:227–235, 2003.
14
Francesco Migliavacca for expertise on multiscale Holzapfel, G. A., T. Gasser, and T. Ogden. A new con-
modeling. Dr. Guccione acknowledges funding from stitutive framework for arterial wall mechanics and a
comparative study of material models. J. Elast. 61:1–48,
the National Institute of Health, grant numbers
2000.
R01HL077921 and R01HL086400. 15
Inzoli, F., F. Migliavacca, and G. Pennati. Numerical
analysis of steady flow in aorto-coronary bypass 3D model.
J. Biomech. Eng. 118:172–179, 1996.
16
Jansen, K. E., C. H. Whiting, and G. M. Hulbert. A gen-
eralized-alpha method for integrating the filtered Navier
REFERENCES
Stokes equations with a stabilized finite element method.
Comput. Methods Appl. Mech. Eng. 190(3–4):305–319,
1
Agoshkov, V., A. Quarteroni, and G. Rozza. A mathe- 2000.
17
matical approach in the design of arterial bypass anasto- Johnson, K., P. Sharma, and J. Oshinski. Coronary artery
mosis using unsteady stokes equations. J. Sci. Comput. flow measurement using navigator echo gated phase
38:139–161, 2006. contrast magnetic resonance velocity mapping at 3.0 t.
2
Bassiouny, H., S. White, S. Glagov, E. Choi, D. P. Gid- J. Biomech. 41:595–602, 2008.
18
dens, and C. K. Zarins. Anastomotic intimal hyperplasia: Katia, L., R. Balossino, F. Migliavacca, G. Pennati,
mechanical injury or flow induced. J. Vasc. Surg. 10:326– E. Bove, M. de Leval, and G. Dubini. Multiscale modeling
337, 1989. of the cardiovascular system: application to the study of
3
Bazilevs, Y., M.-C. Hsu, D. Besnon, S. Sankaran, and pulmonary and coronary perfusions in the univentricular
A. Marsden. Computational fluid-structure interaction: circulation. J. Biomech. 28:1129–1141, 2004.
19
Methods and application to a total cavopulmonary con- Keynton, R., M. Evancho, R. Sims, N. Rodway, A. Gobin,
nection. Comput. Mech. 45(1):77–99, 2009. and S. Rittgers. Intimal hyperplasia and wall shear in
4
Bogren, H., R. Klipstein, D. Firmin, R. Mohiaddin, arterial bypass graft distal anastomosis: an in vivo model
S. Underwood, R. Rees, and D. Longmore. Quantitation study. J. Biomech. Eng. 123:464–473, 2001.
20
of antegrade and retrograde blood-flow in the human aorta Kim, H., I. E. Vignon-Clementel, J. Coogan, C. Figueroa,
by magnetic-resonance velocity mapping. Am. Heart J. K. Jansen, and C. Taylor. On coupling a lumped parameter
117(6):1214–1222, 1989. heart model and a three-dimensional finite element aorta
5
Bryan, A., and G. D. Angelini. The biology of saphenous model. Ann. Biomed. Eng. 37(11):2153–2169, 2009.
21
vein graft occlusion: etiology and strategies for prevention. Kim, H., I. Vignon-Clementel, A. Figueroa, J. Ladisa,
Curr. Opin. Cardiol. 9:641–649, 1994. K. Jansen, J. Feinstein, and C. Taylor. Patient-specific
6
Burattini, R., P. Sipkema, G. Vanhuis, and N. Westerhof. modeling of blood flow and pressure in human coronary
Identification of canine coronary resistance and arteries. Ann. Biomed. Eng. 38(10):3195–3209, 2010.
2242 SANKARAN et al.
22
Kohler, T., T. R. Kirkman, L. W. Kraiss, B. K. Zierler, and D. Giddens. Coronary artery wall shear stress is associated
A. W. Clowes. Increased blood flow inhibits neointimal with progression and transformation of atherosclerotic
hyperplasia in endothelialized vascular grafts. Circ. Res. plaque and arterial remodeling in patients with coronary
69:1557–1565, 1991. artery disease. Circulation 124:779–788, 2011.
23 41
Ku, D. Blood flow in arteries. Ann. Rev. Fluid Mech. Sankaran, S., and A. Marsden. The impact of uncertainty
29:399–434, 1997. on shape optimization of idealized bypass graft models in
24
Lagan, K., G. Dubini, F. Migliavacca, R. Pietrabissa, unsteady flow. Phys. Fluids 22:121902, 2010.
42
G. Pennati, A. Veneziani, and A. Quarteroni. Multiscale Sankaran, S., and A. Marsden. A stochastic collocation
modelling as a tool to prescribe realistic boundary condi- method for uncertainty quantification and propagation in
tions for the study of surgical procedures. Biorheology cardiovascular simulations. J. Biomech. Eng. 133:031001,
39:359–364, 2002. 2011.
25 43
Lei, M., J. Archie, and C. Kleinstreur. Computational de- Sankaranarayanan, M., D. Ghista, C. Poh, T. Seng, and
sign of a bypass graft that minimizes wall shear stress G. Kassab. Analysis of blood flow in an out-of-plane
gradients in the region of the distal anastomosis. J. Vasc. CABG model. Am. J. Physiol. Heart Circ. Physiol.
Surg. 25(4):637–646, 1997. 291:H283–H295, 2006.
26 44
Liu, M., G. S. Roubin, and S. King. Restenosis after cor- Senzaki, H., C. H. Chen, and D. A. Kass. Single beat
onary angioplasty: potential biologic determinants and role estimation of end-systolic pressure-volume relation in
of intimal hyperplasia. Circulation 79:1374–1387, 1989. humans: a new method with the potential for noninvasive
27
Loth, F., P. F. Fischer, and H. S. Bassiouny. Blood flow in application. Circulation 94(10):2497–2506, 1996.
45
end-to-side anastomosis. Ann. Rev. Fluid Mech. 40:367– Shimogonya, Y., T. Ishikawa, Y. Imai, N. Matsuki, and
393, 2008. T. Yamaguchi. Can temporal fluctuation in spatial wall
28
Loth, F., S. A. Jones, C. K. Zarins, D. P. Giddens, R. F. shear stress gradient initiate a cerebral aneurysm?
Nassar, S. Glagov, and H. S. Bassiouny. Relative contri- A proposed novel hemodynamic index, the gradient
bution of wall shear stress and injury in experimental inti- oscillatory number (GON). J. Biomech. 42(4):550–554,
mal thickening at PTFE end-to-side arterial anastomoses. 2009.
46
J. Biomech. Eng. 124:44–51, 2002. Siogkas, P. K., A. I. Sakellarios, T. P. Exarchos,
29
Malek, A. M., S. L. Alper, and S. Izumo. Hemodynamic K. Stefanou, D. I. Fotiadis, K. Naka, L. Michalis,
shear stress and its role in atherosclerosis. J. Am. Med. N. Filipovic, and O. Parodi. Blood flow in arterial seg-
Assoc. 282(21):2035–2042, 1999. ments: rigid vs. deformable walls simulations. J. Serbian
30
Manthaa, A., C. Karmonikc, G. Benndorfb, C. Strotherc, Soc. Comput. Mech. 5(1):69–77, 2011.
47
and R. Metcalfea. Hemodynamics in a cerebral artery Staalsen, N., M. Ulrich, J. Winther, E. Pedersen, T. How,
before and after the formation of an aneurysm. Am. J. and H. Nygaard. The anastomosis angle does change the
Neuroradiol. 27:1113–1118, 2006. flow fields at vascular end-to-side anastomoses in vivo.
31
Marsden, A., J. Feinstein, and C. Taylor. A computational J. Vasc. Surg. 21:460–471, 1995.
48
framework for derivative-free optimization of cardiovas- Taylor, C., M. Draney, J. Ku, D. Parker, B. Steele,
cular geometries. Comput. Methods Appl. Mech. Eng. K. Wang, and C. Zarins. Predictive medicine: computa-
197(21–24):1890–1905, 2008. tional techniques in therapeutic decision-making. Comput.
32
Mehta, D., M. B. Izzat, A. J. Bryan, and G. D. Angelini. Aided Surg. 4(5):231–247, 1999.
49
Towards the prevention of vein graft failure. Int. J. Cardiol. Taylor, C. A., T. J. R. Hughes, and C. K. Zarins. Finite
62(1):S55–S63, 1997. element modeling of blood flow in arteries. Comput.
33
Migliavacca, F., and G. Dubini. Computational modeling Methods Appl. Mech. Eng. 158(1–2):155–196, 1997.
50
of vascular anastomoses. Biomech. Model. Mechanobiol. Walker, J., M. B. Ratcliffe, P. Zhang, A. W. Wallace,
3:235–250, 2005. B. Fata, E. W. Hsu, D. Saloner, and J. M. Guccione. MRI-
34
Moore, J., D. A. Steinman, S. Prakash, K. W. Johnson, based finite-element analysis of left ventricular aneurysm.
and C. R. Ethier. A numerical study of blood flow patterns Am. J. Physiol. 289:H692–H700, 2005.
51
in anatomically realistic and simplified end-to-side anasto- Wilson, N., K. Wang, R. Dutton, and C. A. Taylor. A
moses. J. Biomech. Eng. 121:265–272, 1999. software framework for creating patient specific geometric
35
Neal, M., and J. Bassingthwaighte. Subject-specific model models from medical imaging data for simulation based
estimation of cardiac output and blood volume during medical planning of vascular surgery. Lect. Notes Comput.
hemorrhage. Cardiovasc. Eng. 7(3):97–120, 2007. Sci. 2208:449–456, 2001.
36 52
Nikkari, S., and A. W. Clowes. Restenosis after vascular Yang, W., I. Vignon-Clementel, G. Troianowski,
reconstruction. Ann. Med. 26:95–100, 1994. V. Mohan Reddy, J. Feinstein, and A. Marsden. Hepatic
37
Ojha, M. Spatial and temporal variations of wall shear blood flow distribution and performance in conventional
stress within an end-to-side arterial anastomosis model. and novel Y-graft Fontan geometries: a case series com-
J. Biomech. 26(12):1377–1388, 1993. putational fluid dynamics study. J. Thorac. Cardiovasc.
38
Ojha, M. Wall shear stress temporal gradient and anasto- Surg. 143(5):1086–1097, 2012.
53
motic intimal hyperplasia. Circ. Res. 74:1227–1231, 1994. Zamir, M., P. Sinclair, and T. H. Wonnacott. Relation
39
Rittgers, S., P. Karayannacos, J. Guy, R. Nerem, G. Shaw, between diameter and flow in major branches of the arch of
J. Hostetler, and J. Vasko. Velocity distribution and intimal the aorta. J. Biomech. 25:1303–1310, 1992.
54
proliferation in autologous vein grafts in dogs. Circ. Res. Zarins, C., M. A. Zatina, D. P. Giddens, D. N. Ku, and
42(6):792–801, 1978. S. Glagov. Shear stress regulation of artery lumen diameter
40
Samady, H., P. Eshtehardi, M. McDaniel, J. Suo, in experimental atherogenesis. J. Vasc. Surg. 5:413–420,
S. Dhawan, C. Maynard, L. Timmins, A. Quyyumi, and 1987.