Journal of Biomechanics 48 (2015) 18171827

Contents lists available at ScienceDirect

Journal of Biomechanics
journal homepage: www.elsevier.com/locate/jbiomech
www.JBiomech.com

Quantication of wall shear stress using a nite-element method

in multidimensional phase-contrast MR data of the thoracic aorta
Julio Sotelo a,b,c, Jess Urbina a,d, Israel Valverde e,f, Cristian Tejos a,b,g, Pablo Irarrzaval a,b,g,
Daniel E. Hurtado c,g, Sergio Uribe a,d,g,n
a

Biomedical Imaging Center, Ponticia Universidad Catolica de Chile, Santiago, Chile

Electrical Engineering Department, Ponticia Universidad Catolica de Chile, Santiago, Chile
c
Structural and Geotechnical Engineering Department, Ponticia Universidad Catolica de Chile, Santiago, Chile
d
Radiology Department, School of Medicine, Ponticia Universidad Catolica de Chile, Santiago, Chile
e
Pediatric Cardiology Unit, Hospital Virgen del Roco, Seville, Spain
f
Laboratory of Cardiovascular Pathophysiology, Seville Biomedicine Institute, Hospital Virgen del Roco, Seville, Spain
g
Biological and Medical Engineering Institute, Schools of Engineering, Medicine and Biological Sciences, Ponticia Universidad Catolica de Chile, Santiago,
Chile
b

art ic l e i nf o

a b s t r a c t

Article history:
Accepted 27 April 2015

We present a computational method for calculating the distribution of wall shear stress (WSS) in the
aorta based on a velocity eld obtained from two-dimensional (2D) phase-contrast magnetic resonance
imaging (PC-MRI) data and a nite-element method. The WSS vector was obtained from a global leastsquares stress-projection method. The method was benchmarked against the Womersley model, and the
robustness was assessed by changing resolution, noise, and positioning of the vessel wall. To showcase
the applicability of the method, we report the axial, circumferential and magnitude of the WSS using invivo data from ve volunteers. Our results showed that WSS values obtained with our method were in
good agreement with those obtained from the Womersley model. The results for the WSS contour means
showed a systematic but decreasing bias when the pixel size was reduced. The proposed method proved
to be robust to changes in noise level, and an incorrect position of the vessel wall showed large errors
when the pixel size was decreased. In volunteers, the results obtained were in good agreement with
those found in the literature. In summary, we have proposed a novel image-based computational method
for the estimation of WSS on vessel sections with arbitrary cross-section geometry that is robust in the
presence of noise and boundary misplacements.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
2D CINE PC-MRI
Wall shear stress
Finite elements
Fluid mechanics
Flow quantication

1. Introduction
Two-dimensional cine phase-contrast magnetic resonance imaging (2D CINE PC-MRI) and three-dimensional (3D) CINE PC-MRI
have been used non-invasively to obtain qualitative and quantitative
information on the cardiovascular system. Several numerical procedures have recently been proposed to evaluate ow patterns,
determine the wall shear stress (WSS) distribution, and calculate
pressure difference maps (Oshinski et al., 1995; Tyszka et al., 2000;
Ebbers et al., 2001; Barker et al., 2010; Bock et al., 2010, 2011). These
methods have shown the potential of 2D and 3D CINE PC-MRI for
assessing different cardiovascular diseases (Wigstrm et al., 1999;
n
Correspondence to: Radiology Department, School of Medicine and Biomedical
Imaging Center. Ponticia Universidad Catlica de Chile. Marcoleta 367, Santiago,
Chile. Tel.: 56 2 23548272; fax: 56 2 23548468.
E-mail address: suribe@med.puc.cl (S. Uribe).

http://dx.doi.org/10.1016/j.jbiomech.2015.04.038
0021-9290/& 2015 Elsevier Ltd. All rights reserved.

Weigang et al., 2008; Boussel et al., 2009; Kafka and Mohiaddin,

2009; Markl et al., 2010; Cecchi et al., 2011; Frydrychowicz et al.,
2011; Francois et al., 2012). In particular, the evaluation of the WSS
distribution in the aortas of healthy volunteers (Stalder et al., 2008;
Frydrychowicz et al., 2009a) and patients has recently been reported
by several groups (Frydrychowicz et al., 2009b; Barker et al., 2010;
Harloff et al., 2010; Bieging et al., 2011). It has been shown that
different WSS-related parameters including the axial and circumferential components of WSS and the oscillatory shear index
have potential for assessing vascular function in several cardiovascular diseases such as atherosclerosis, aneurysms, stenosis and
restenosis (Cecchi et al., 2011).
The estimation of the WSS distribution from 2D CINE PC-MRI
goes back to the work of Oshinski et al. (1995), Morgan et al.
(1998a,b) and Oyre et al. (1998). In Oshinski et al. the WSS is
calculated from the product of the uid viscosity and the velocity
gradient at the wall, correcting for the wall position using MR data.

1818

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

Assuming an axial ow prole in a perfectly cylindrical vessel

(Poiseuille ow model), Oyre et al. tted a paraboloid to the
through-plane velocity prole measured at a boundary layer (Oyre
et al., 1998). From the t, they were able to compute the axial
velocity gradient, and in turn the WSS in the carotid arteries of
seven healthy volunteers. Whereas this approach may be valid for
small and rounded vessels, it cannot capture other ow patterns
commonly found in larger vessels, where the velocity prole does
not follow a parabolic distribution. Morgan et al. followed a different approach in which the tangential, radial and axial velocity
gradients were numerically estimated using a nite-difference
scheme to compute the WSS tensor at the left and right pulmonary
arteries (Morgan et al., 1998a,b). However, it is well known that
the nite-difference method cannot effectively handle complex
geometries such as those found in the cardiovascular system,
neither can it impose boundary conditions on irregular surfaces in
a direct manner (Zienkiewicz et al., 2005). To account for arbitrary
cross-section shapes, Stalder et al. (2008) used cubic B-spline
interpolations to smoothly describe the lumen contours as well as
to obtain a continuous and smooth description of the velocity
eld. This method has been the standard for quantifying the WSS
from 2D cine PC-MRI. However, as noted by Stalder et al., the
computation of WSS using B-spline interpolations on 2D CINE PCMRI and on reformatted slices from 3D CINE PC-MRI, both with 3D
velocity encoding, introduces important approximation errors due
to limited spatial resolution and numerical differentiation of the
velocity eld, which in the case of a Poiseuille ow can be as high
as 40% in the estimation of WSS.
In this work we propose a nite-element-based methodology
to compute the WSS at arbitrary plane sections of the thoracic
aorta from a velocity eld given by 2D CINE PC-MRI data. The
nite-element method has a well-established reputation for efciently representing 3D complex geometries, and has been successfully employed in patient-specic cardiovascular and cardiac
simulations (Taylor and Figueroa, 2009; Xiao et al., 2013; Hurtado
and Kuhl, 2014), providing in general a robust means for cardiovascular modeling and computation, with numerical convergence
that can be rigorously proven (Hurtado and Henao, 2014). In particular, nite-element methods have been used in computational
uid dynamic (CFD) simulations to obtain different hemodynamic
parameters on the basis of 3D models build from angiography
images and boundary conditions from 2D PC-MRI (LaDisa et al.,
2011; Goubergrits et al., 2014). Nevertheless, as far as we are
aware, nite elements have not been applied directly to process
the velocity data from 2D PC-MRI and in turn to obtain the WSS.
To estimate the velocity gradients, the domain of interest is
discretized using triangular elements, and the velocities at the
center of each voxel are interpolated using a conforming niteelement approximation of the velocity eld. In order to improve
the accuracy of the computed strain and stress elds, several a
posteriori stress-recovery methods have been proposed in the
literature (Zienkiewicz et al., 2005). Here, we adopt a global leastsquares stress projection method (Oden and Brauchli, 1971), which
has been shown to be super-convergent for linear elements
(Zienkiewicz and Zhu, 1992), exhibiting in some cases a better
performance than alternative methods (Heimsund et al., 2002).
We tested the proposed methodology using a Womersley ow
prole as a benchmark. The robustness of the method was assessed under different levels of resolution and noise and incorrect
positioning of the vessel wall. To showcase the applicability of the
method, we report the axial, circumferential and magnitude of the
WSS using in-vivo data.

2. Theory
2.1. Computation of the wall shear stress using a nite-element
method
The shear stress vector and magnitude at the vessel wall were
computed using the procedure described in the electronic supplementary material (see Appendix: nite element formulation),
which we briey summarize next. The velocity eld was obtained
at a discrete set of pixels using 2D CINE PC-MRI. Using linear triangular nite-element interpolations, the velocity-component
eld uiFEM (x,t ) is continuously described by the expression:

uiFEM (x, t ) =

N A (x ) viA (t ),

(1)

where NA (x ) is the nite-element shape function associated to

node A, viA (t ) is the i-th velocity component at the node A at time t ,
and is the set of all nodes of the triangular mesh used as discretization of the section under study. Based on Eq. (1), the shear
stress tensor components can be approximated using a global leastsquares stress projection method (Oden and Brauchli, 1971; Hinton
and Campbell, 1974), which consists in approximating the stress eld
S (x, t ) by:

S (x, t ) =

N A (x ) A (t ),

(2)

where A is the nodal smoothed value of the stress components obtained from a global least-squares minimization of the
stress L2 error. Once all the shear-component elds are obtained,
the shear stress tensor at any point in the domain of interest, and
particularly at the domain boundaries (i.e. vessel wall), can be

estimated using Eq. (2). Let n be the inward unit vector normal to
the vessel wall at a particular point of interest. Then, the WSS
vector corresponding to the shear stress tensor takes the form:

(3)

t = n .

For the purpose of this work, we consider the axial, circumferential and magnitude of the WSS vector projected over the

lumen contour t proy :

t proy = n

t
n .

(4)

From t proy = [t X , tY , t Z ], the axial component, t Z , represented the

projection in the longitudinal direction, and the circumferential
component represented the projection along the lumen circumference, which was calculated as t x 2 + t z 2 .
The method just described was implemented in Python language. It is important to mention that we dened the velocities in
the boundary of our mesh to be equal to zero, following a no-slip
boundary assumption.
3. Methods
3.1. Womersley ow model and robustness analysis
To evaluate the stability and robustness of the method, we generated synthetic
velocity proles using the Womersley model (Eq. (5)) (Womersley, 1955). A
detailed explication of the Womersley model is given in electronic supplementary
material (Womersley formulation see Appendix). From the Womersley model the
velocity inside a cylinder is given by:

vel (x , y , t ) =

) ,

J0 r (x , y ) i 3/2

A0
1
v
R 2 r (x , y )2 + Re PGm (t )
1
3/2
4
i

J
R
i
0

(5)

where vel (x, y, t ) is the velocity in the point (x, y ) (radius r ) at time t , in the
interior of a cylinder of length L and radius R ; is the blood density, is the viscosity

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

1819

Fig. 1. Analytic Womersley model. (a) The pressure difference which is the input of our model. (b) and (c) the mean velocity and WSS obtained by the Womersley model.
(d) The ow solution. (e) We observe four velocity proles at different times (0.077 s, 0.205 s, 0.256 s, and 0.333 s for the cardiac phases 3, 9, 11, and 14 respectively) and the
2D triangular mesh.
and v = / is the kinematic viscosity. A0 is the mean value of the pressure gradient
PGm (t ) . J is the Bessel function of the rst kind and order . The parameter
= R /v is a non-dimensional parameter known as the Womersley number, with
frequency .
For obtaining physiological peak ow and velocities similar to those in the
normal aorta, but not necessarily physiological waveform along time, we considered a cylinder of length 10 cm, diameter 2 cm, with a pressure gradient range
7 6.4 mm Hg, 40 cardiac phases and a heart rate of 60 beats per minutes. These
values yielded a time average (average along the time) of mean velocity equal to
45.29 cm/s, mean ow equal to 141.3 ml/s. The WSS can be obtained by

differentiating Eq. (5) at the vessel wall, as shown in Eq. (6). Using this equation we
obtained a time average of axial WSS of 0.85 N/m2.

(R , t ) =

PGm (t ) R
A0 R
+ Re

2
i

3/2
i
v

J1 R

J R
0

(
(

)
)

3/2
i
v

3/2
i
v

(6)

The time evolution of the pressure difference considered in this analysis

together with the corresponding mean velocity, WSS and ow are shown in Fig. 1.

1820

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

3.1.1. Effect of resolution and noise

The circular domain was discretized with triangular meshes of seven different
element sizes (element area7standard deviation 2 70.61 mm2, 1.5 7 0.43 mm2,
17 0.25 mm2, 0.5 7 0.14 mm2, 0.17 0.028 mm2, 0.05 7 0.014 mm2, 0.01 7
0.004 mm2). Additionally, the velocity values were perturbed in the X, Y
and Z directions by three different Gaussian noises with mean 0 and standard
deviation of 0.25%, 0.5% and 1% of the peak velocity value, which was 1.65 m/s
(Fig. 2a and b).

3.1.2. Effects of position inaccuracy

We assessed the effects of an incorrect positioning of the vessel wall using
the two strategies sketched in Fig. 2c and d. Specically, we increased and
decreased the circle mesh by 0.5 mm, 1 mm and 2 mm for the same Womersley
model. We performed this experiment for the seven different element sizes with
the condition of zero noise. In all cases, we calculated the time-averaged ow
and WSS to assess the agreement between our method and the analytical
solution.

3.2. In-vivo MR acquisition

2D CINE PC-MRI datasets were acquired in the thoracic aortas of ve volunteers
(mean age 27 years; range 2433 years; one female) at ve planes transecting the
thoracic aorta as schematically illustrated in Fig. 3. Data were acquired on a clinical 1.5T
MR system (Philips Achieva, Best, The Netherlands) using an RF-spoiled gradient-echo
sequence during free breathing and retrospective electrocardiogram gating as in Markl
et al. (2007) and Uribe et al. (2009). The acquisition parameters for the in-vivo
experiments were: temporal resolution 38 ms, VENC 200 cm/s, and 25 cardiac phases, a
ip angle of 15 and an acquired and reconstructed resolution of 1.67  1.67 mm2 and
0.83  0.83 mm2 respectively.
3.2.1. Segmentation and mesh generation
Data processing included the segmentation of the lumen and the generation of
the nite-element mesh for each planar section under study (Fig. 3b). For each
cardiac phase of in-vitro and in-vivo data, the vessel lumen of the anatomical image
was segmented using the ChanVese algorithm (Chan and Vese, 2001). No

Fig. 2. (a) The seven different mesh resolutions used for the Womersley analysis. (b) Three different Gaussian noise levels with mean 0 and standard deviation of 0.25%, 0.5%
and 1% of the peak velocity value that was 1.65 m/s; the velocity prole corresponds to cardiac phase number 11 at time 0.256 s. (c,d) The procedure for assessing the effect
of an incorrect position of the segmentation in the analysis of WSS: (c) erosion and (d) dilatation.

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

1821

Fig. 3. (a) Schematic illustration of the location of analyzed planes. These planes were positioned in the ascending aorta (AO1), at the beginning and the end of aortic arch
(AO2AO3) and at the descending aorta (AO4AO5), with the innermost curvature reference positions marked by dots. (b) The mesh generation process. From the 2D PC-MRI
we draw a region of interest (ROI) in the T1 magnitude image (anatomical); in this case the ROI was drawn in section AO1. The vessel lumen of the region was then
segmented using a ChanVese algorithm; subsequently, each node (pixel in the 2D image) of the domain of interest was identied, and the triangular mesh was generated.

smoothing lters were applied to the 2D CINE PC-MRI datasets before the segmentation process. The nite-element mesh was generated by creating triangular
elements, where each node of the mesh was located in the center of pixels in the
segmented region, covering the entire lumen (Fig. 3b).
3.2.2. Effects of MRI resolution
An experiment similar to the one performed in the Womersley model was
carried out in vivo. In one volunteer we acquired ve datasets with different
resolutions at the AO1 level, 0.7  0.7 mm, 1.0  1.0 mm, 1.25  1.25 mm,
1.5  1.5 mm and 2.0  2.0 mm. Additionally, for each dataset of this volunteer the
WSS values were obtained using ve different element sizes with an element area
of 0.49 mm2, 1 mm2, 1.56 mm2, 2.25 mm2 and 4.0 mm2. We analyzed the WSS
values only in the systolic phase. When the center of the element did not match the
pixel center the velocity data were interpolated using a cubic interpolation.

Table 1
WSS contour mean and standard deviation, for the cardiac phases (13, 16, 29) and
different element sizes.
ES

(T*)

N1

N2

N3

N4

CP13

0.01
0.05
0.1
0.5
1.0
1.5
2.0

 7.40
 7.40
 7.40
 7.40
 7.40
 7.40
 7.40

 6.8 70.09
 6.3 70.18
 5.8 70.25
 4.0 70.52
 3.1 70.45
 2.5 70.54
 1.8 70.44

 6.8 70.18
 6.3 70.20
 5.8 70.26
 4.0 70.52
 3.1 70.45
 2.5 70.54
 1.8 70.45

 6.8 70.34
 6.2 70.24
 5.8 70.28
 4.0 70.54
 3.1 70.44
 2.5 70.55
 1.8 70.44

 6.8 7 0.65
 6.3 7 0.40
 5.8 7 0.29
 4.0 7 0.53
 3.17 0.47
 2.5 7 0.52
 1.8 7 0.47

CP16

0.01
0.05
0.1
0.5
1.0
1.5
2.0

 5.82
 5.82
 5.82
 5.82
 5.82
 5.82
 5.82

 5.7 70.02
 5.5 70.07
 5.4 70.11
 4.3 70.36
 3.7 70.37
 3.2 70.47
 2.6 70.43

 5.7 70.10
 5.5 70.10
 5.4 70.11
 4.3 70.36
 3.7 70.38
 3.2 70.47
 2.6 70.43

 5.6 70.63
 5.5 70.31
 5.4 70.24
 4.3 70.38
 3.7 70.41
 3.2 70.48
 2.6 70.42

 5.6 7 0.63
 5.5 7 0.31
 5.4 7 0.24
 4.3 7 0.38
 3.7 7 0.41
 3.2 7 0.48
 2.6 7 0.43

CP29

0.01
0.05
0.1
0.5
1.0
1.5
2.0

5.48
5.48
5.48
5.48
5.48
5.48
5.48

5.2 70.04
4.9 70.09
4.7 70.13
3.7 70.30
3.2 70.27
2.8 70.34
2.4 70.29

5.2 70.37
4.9 70.16
4.7 70.17
3.7 70.28
3.2 70.28
2.8 70.33
2.4 70.30

5.2 7 0.68
4.9 7 0.33
4.7 7 0.23
3.7 7 0.31
3.2 7 0.28
2.8 7 0.33
2.4 7 0.30

3.3. Statistical analysis

Over the boundary of each segmented image we calculated the WSS contour
mean and standard deviation as follows:

WSS =

C t w dL
C

WSS =

tw

2
WSS d L ,

(7)

where t w could represent the axial, circumferential and magnitude of WSS

obtained from our method and C is the perimeter length of the vessel contour C .
For each section of the in-vivo data we calculated the axial, circumferential and
magnitude of WSS contour mean, standard deviation and time to peak. To study
the intra- and inter-observer reproducibility of the proposed approach, analysis of
the data was performed twice by one observer (JS), and once by another observer
(SU). For this analysis we computed the mean difference (MeanD ) between the
observers as follows:

MeanD =

N
a
b
cp
=1 t cp t cp

),

5.2 70.12
4.9 70.12
4.7 70.15
3.7 70.31
3.2 70.27
2.8 7 0.34
2.4 70.29

ES characteristic element size in mm. T theoretical WSS in N/m2. CP cardiac

phase. N1 noise level of 0%. N2 noise level of 0.25%. N3 noise level of 0.5%.
N4 noise level of 1%.

(8)

b
a
where tcp
is the WSS contour mean value processed by one observer a and tcp
is the equivalent value processed by the other observer b , for each cardiac phase
(cp 1,..,40).

4. Results
4.1. Robustness analysis
In Table 1 we show the results of the axial WSS contour mean
(WSS ) and standard deviation (WSS ) for three representative

cardiac phases. We selected two phases that showed a peak forward and backward ow (Fig. 4a), and one cardiac phase that
showed a peak in WSS, as indicated in Fig. 4b. In our simulation
these values corresponded to the times of 0.307 s (cardiac phase
13, peak WSS), 0.384 s (cardiac phase 16, peak retrograde ow)
and 0.717 s (cardiac phase 29, peak forward ow). As observed in
Table 1, the WSS values obtained with our method were similar to
the theoretical ones for the smallest elements. Notably, our results
also show that the levels of noise had a negligible effect on the
results for any cardiac phase.

1822

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

Fig. 4. (a) Flow and (b) WSS contour mean as function of time for different element sizes. The analytic values are shown in black; we observed that as the element mesh size
is smaller the ow and WSS contour mean value converge to the theoretical one. (c) Different velocity proles for cardiac phases 13, 16 and 29, at time 0.307 s, 0.384 s and
0.717 s respectively.

4.1.1. Effect of resolution and noise

The resulting ow estimations and the axial WSS contour mean
of the Womersley model for different element sizes and zero noise
levels are depicted in Fig. 4a and b. We note from Fig. 4 that the
distribution shape of the WSS obtained follows the theoretical
distribution, delivering contour mean values that were in excellent
agreement with those obtained from the analytical model when
the element size was small, independently of the cardiac phase.
However, for large elements, the WSS presented larger differences
with respect to the theoretical model. In particular, for cardiac
phase number 13 (Table 1), the ow prole exhibited a great slope
near the vessel wall, which is difcult to capture with large elements. When noise was increased to 1% of the maximum velocity
value, the time-average WSS contour mean did not change with
respect to the zero noise level. The analytical time-averaged WSS
contour mean was 0.85 N/m2.

smaller elements outside the vessel had a velocity equal to zero.

On the other hand, when the position of the wall was located
inside the real vessel contour (i.e. erosion of the vessel wall), the
WSS contour means were in better agreement as the element size
decreased. However, when the element size was o0.1 mm, the
WSS started to deviate from the theoretical value. This can be
explained by the fact that the zero velocity condition at the edge of
the segmentation increases the slope of the velocity at the wall
when the vessel is eroded, which can be better captured by a
smaller element.
In Fig. 6 we show the ow and WSS contour means for different
element sizes subjected to changes of position of the vessel wall.
In general, we found that the time average of the WSS contour
mean increases when the vessel wall is moved inwards (erosion)
and decreases when it is moved outwards (dilatation).
4.2. In-vivo MR acquisition

4.1.2. Effects of position inaccuracy

In Fig. 5, we analyzed the effect of incorrect positioning of the
wall in the estimation of the WSS distribution for different element sizes for cardiac phase 13. When the vessel wall was dilated,
the WSS tended to zero for small element sizes. In this case,

The time-averaged ow rates in volunteers from 2D PC-MRI were

72710.37 ml/s, 72712.93 ml/s, 54714.23 ml/s, 5076.56 ml/s and
4878.39 ml/s for AO1, AO2, AO3, AO4 and AO5 respectively, where
7 indicates one standard deviation. Plots of ow and WSS contour

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

1823

Fig. 5. Resulting WSS for different positions of the vessel wall for cardiac phase 13. The rst row shows the WSS obtained when the segmentation curve is placed correctly at
the vessel wall. The second row shows the WSS obtained when the segmentation curve is erroneously moved outwards by 0.5 mm. The third row shows the corresponding
results when the segmentation curve is erroneously moved inwards by 0.5 mm. The columns show seven different spatial resolutions of the mesh.

means as a function of time are shown in Fig. 7. Additionally, in this

gure we show the mean and standard deviation (in parenthesis)
between volunteers of ow, time to peak, and WSS for all sections.
Table 2 shows the result of the intra- and inter-observer
variability study in volunteers. We found a high correlation with R2
values of 0.95, 0.99 and 0.94 for axial, circumferential and magnitude of WSS for the intra-observer variability. For the interobserver analysis we obtained R2 values of 0.92, 0.96 and 0.88 for
axial, circumferential and magnitude of wall shear stress. The
intra-observer analysis showed that the mean difference for the
axial, circumferential and magnitude of WSS were 0.025 7
0.040 N/m2, 0.008 70.010 N/m2, and 0.025 70.040 N/m2, respectively. The inter-observer analysis showed a mean WSS stress
difference of 0.032 70.047 N/m2, 0.0117 0.012 N/m2, and 0.031 7
0.047 N/m2 for the axial, circumferential and magnitude of WSS,
respectively.
4.2.1. Effects of MRI resolution
Table 3 shows the WSS contour means for the data acquired in
one volunteer with different resolutions and analyzed with different element sizes. We observed the same trend as that found in
the Womersley model. For the following analysis let's consider our
best estimation of the WSS when the pixel size was 0.7  0.7 mm2
and the element size was 0.49 mm2. When the acquired resolution
matched the element size (bold numbers in the diagonals), we
found that larger pixel sizes delivered underestimated values of

WSS. In the same manner, we observed that when the MR data

were analyzed with element sizes larger than the pixel resolution
(data below the diagonal) the WSS values were underestimated.
On the other hand when the MR data were analyzed with an
element size smaller than the pixel resolution (data above the
diagonal), underestimated WSS values were less. However, WSS
values were overestimated when the element size was 0.49 mm2
and the acquired pixel resolution was larger than 0.7  0.7 mm.

5. Discussion
We have proposed a novel approach for the determination of
WSS distributions using nite-element and stress-recovery
methods based on 2D cine PC-MRI. The proposed method has
distinctive theoretical advantages. First, the nite-element
approach allows for effortless and automatic-meshing procedures
based on segmented images. Second, it delivers an estimate of the
shear stress, or any spatial derivative of the velocity eld, that
converges to the exact solution as the element size of the mesh
decreases, as shown for the case of the Womersley model. This
trend is consistent with the well-known convergence properties of
nite-element approximations (Zienkiewicz et al., 2005). Third,
the proposed method delivers a continuous eld for the shear
stress tensor inside the lumen of the vessel under analysis.

1824

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

Fig. 6. Time average ow and WSS contour means, for different element size subjected to changes of position of the vessel wall (dilation and erosion). The theoretical value
(TV) of ow volume and time average of WSS contour mean values were 141.3 ml/s and 0.85 N/m2, respectively. D1, displacement of 0 mm; D2, displacement of 0.5 mm; D3,
displacement of 1 mm; D4, displacement of 2 mm.

Experiments using the Womersley model as a benchmark

allowed us to analyze the performance of the method for different
ow proles (different cardiac phases), resolution, noise and
incorrect positioning of the vessel walls.
The proposed approach delivered axial WSS contour mean
values that were similar to theoretical values for small element
sizes and for any ow prole. However, estimated WSS values tend
to be less accurate as the element sizes increased. This result was
corroborated with in-vivo data acquired with different resolutions
and analyzed with different element sizes. One way to improve the
stability of our method is to improve the approximation properties
of the nite-element interpolation by increasing the continuity of
the derivatives across elements, together with least-squares ttings instead of direct velocity interpolations.
Using the Womersley model we also demonstrated that the
proposed method has remarkable stability for different levels of
noise (Table 1). This is a relevant and desirable feature, considering
the fact that images obtained with PC-MRI techniques are normally affected by a low velocity-to-noise ratio.
Our results showed that the WSS values were affected by the
position of the wall. In this work we forced an incorrect positioning of the vessel wall greater than one pixel as was analyzed
by Petersson et al. (2012). We found greater dependency and
errors compared to the methods studied by Petersson. In volunteers we performed the segmentation for each cardiac phase in the
anatomical image provided by 2D PC-MRI data. The 2D anatomical
image provided well-dened contours and therefore positioning of
the vessel wall did not affect the WSS obtained in vivo. Nevertheless, incorrect positioning of the vessel wall may be particularly
relevant when analyzing 3D PC-MRI data. Those datasets do not
have enough quality to dene precisely the boundary of the vessel
wall for each cardiac phase. To improve the quality of the 3D PCMR images, it is common to average all cardiac phases of 3D

datasets and then perform the segmentation on a single average

image as described previously (Bock et al., 2010).
In volunteers, axial, circumferential and magnitude of WSS
contour mean values showed, in general, a good agreement with
the values found in the literature (Stalder et al., 2008). We obtained
a time-average WSS across the ve volunteers equal to
AO10.3370.10 N/m2, AO20.3470.05 N/m2, AO3 0.447
0.09 N/m2, AO40.4070.13 N/m2 and AO50.4270.08 N/m2,
while Stalder et al. obtained WSS values in similar positions equal to
AO10.4370.08 N/m2,
AO20.4470.09 N/m2,
AO3 0.467
0.08 N/m2, AO40.5070.08 N/m2 and AO50.5670.08 N/m2.
Although we did not perform a one-to-one comparison with the
method proposed by Stalder et al., we observed differences in the
results obtained that were negligible.
One limitation of our study is that we did not provide a one-toone comparison with another method; however, this was not the
purpose of this work. The objective of the present study was to
present a new method, showing its advantages and disadvantages
through extensive tests in a realistic analytical model corroborated
with in-vivo data. Furthermore, since a gold standard method for
in-vivo WSS measurements is not currently available, it is difcult
to ascertain whether the results obtained are more or less accurate
than those with other methods.
The proposed methodology also suffers from certain limitations.
First, an extrinsic limitation is the resolution of the MR images, which
yielded underestimated values of WSS. Although a similar peak ow
was used for the Womersley model, we obtained different WSS
values compared to those in volunteers. Some of these differences can
be explained by the use of different resolutions. For instance, we
obtained a WSS5.82 N/m2 with an innite resolution for the
Womersley model and WSS 1.59 N/m2 in volunteers using a resolution of 0.83  0.83 mm. Some other effects including noise, ow
prole and velocity accuracy could also inuence these differences.

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

1825

Fig. 7. Plots of the ow, WSS axial (WSS Axi.), WSS circumferential (WSS Cir.) and WSS magnitude (WSS Mag.) for the sections AO1 (a), AO3 (b) and AO5 (c). Additionally,
(d) shows the time average and standard deviation (in parentheses) between volunteers of different variables for each section.

Second, throughout this work, linear interpolation functions within

each element have been employed in all calculations. This can result
in a poor approximation scheme if the domain discretization is too

coarse. Third, the nite-element representation of the vessel wall

consists in the union of straight segments, with normal boundary
vectors that are discontinuous at the element boundary nodes. Since

1826

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

Table 2
Inter-observer and intra-observer analysis for volunteer data. We show the mean absolute error and standard deviation ( 7 ) between observers for the ow, area, mean
velocity, time to peak. Additionally, we show the mean absolute error and standard deviation for the time average of axial (Axi.), circumferential (Cir.), and magnitude (Mag.)
of WSS.
AO1

AO2

AO3

AO4

AO5

Inter-observer

Flow (ml/s)
Area (mm2)
Mean velocity (m/s)
Time to peak (ms)
Mean diff Axi. WSS (N/m2)
Mean diff Cir. WSS (N/m2)
Mean diff Mag. WSS (N/m2)

2.17 1.4
16.5 7 5.2
0.7 7 0.4
0.0 7 0.0
0.0277 0.046
0.0137 0.015
0.029 7 0.047

1.4 70.5
11.9 74.4
0.6 70.4
0.0 70.0
0.029 70.043
0.014 70.014
0.030 70.044

2.17 1.6
10.7 7 9.5
1.2 7 0.6
0.0 7 0.0
0.0377 0.054
0.0117 0.009
0.036 7 0.053

0.9 7 0.6
5.2 7 2.5
0.5 7 0.2
0.0 7 0.0
0.0217 0.031
0.0077 0.009
0.022 7 0.030

1.0 7 1.0
6.17 5.5
1.0 7 0.6
0.0 7 0.0
0.0467 0.062
0.0127 0.013
0.0407 0.063

Intra-observer

Flow (ml/s)
Area (mm2)
Mean velocity (m/s)
Time to peak (ms)
Mean diff Axi. WSS (N/m2)
Mean diff Cir. WSS (N/m2)
Mean diff Mag. WSS (N/m2)

2.6 7 2.1
23.7 7 10.4
0.7 7 0.3
0.0 7 0.0
0.0277 0.044
0.0097 0.011
0.028 7 0.044

1.9 71.0
16.1 77.4
0.9 70.4
0.0 70.0
0.020 70.030
0.009 70.011
0.020 70.032

2.8 7 1.6
15.17 7.6
1.2 7 0.7
0.0 7 0.0
0.026 7 0.043
0.0097 0.009
0.026 7 0.041

1.2 7 1.4
7.4 7 8.1
0.6 7 0.5
0.0 7 0.0
0.0197 0.030
0.0077 0.008
0.0197 0.029

1.3 7 0.8
9.9 7 5.3
1.17 0.2
0.0 7 0.0
0.034 7 0.051
0.0077 0.009
0.032 7 0.052

Table 3
Axial (Axi.), circumferential (Cir.) and magnitude (Mag.) of WSS contour mean, for
the systolic phase of the data acquired in one volunteer with different resolutions
and analyzed with different element sizes. For this experiment, we select as a
reference the WSS value obtained with spatial resolution of 0.7x0.7 mm and element size of 0.49 mm2. These values are marked with * in the table.
MRI resolution (mm)
0.7  0.7 1.0  1.0 1.25  1.25 1.5  1.5 2.0  2.0
Element
size
(mm2)

WSS
Axi.

WSS
Cir.

WSS
Mag.

0.49
1.00
1.56
2.25
4.00

1.33*
1.11
1.04
0.93
0.81

1.59
1.30
1.13
1.01
0.85

1.63
1.28
1.15
1.03
0.82

1.50
1.21
1.01
0.94
0.78

1.55
1.21
1.01
0.88
0.75

0.49
1.00
1.56
2.25
4.00

0.09*
0.05
0.04
0.04
0.04

0.11
0.06
0.05
0.04
0.04

0.09
0.06
0.05
0.05
0.05

0.07
0.04
0.04
0.04
0.04

0.08
0.04
0.03
0.02
0.02

0.49
1.00
1.56
2.25
4.00

1.34*
1.11
1.05
0.93
0.81

1.59
1.30
1.13
1.01
0.85

1.64
1.28
1.15
1.03
0.82

1.50
1.21
1.01
0.94
0.78

1.55
1.22
1.01
0.88
0.75

the WSS is linearly proportional to the normal vector, a discontinuity

may appear in the WSS value across elements, which may not be
realistic. Again, as the mesh size is decreased, this limitation is
diminished, but the accuracy of the estimation of normal vectors will
depend largely on the image pixel resolution. Finally, we have only
shown results from 2D PC-MRI data; we could also apply the developed technique to 2D reformatted data from 3D PC-MRI.
We are currently developing a similar methodology that works
directly on velocity data obtained from 3D PC-MRI to obtain the
distribution of WSS along an entire vessel of interest. Additionally,
we plan to build upon the proposed methodology to study the
distribution of the oscillatory shear index and three-band diagrams, which may yield additional information on the dynamic
structure of the WSS distribution (Gizzi et al., 2011).
Conict of interest
None

Acknowledgment
The authors acknowledge the support of the Interdisciplinary
Research Fund VRI# 44/2011 from the Ponticia Universidad
Catlica de Chile, Anillo ACT 079, and FONDECYT through Grants
#11100427, #11121224 and #1141036. Julio Sotelo thanks CONICYT
and the Ministry of Education of Chile, with his higher education
program, for the graduate scholarship for doctoral studies.

Appendix A. Supplementary material

Supplementary data associated with this article can be found in
the online version at. http://dx.doi.org/10.1016/j.jbiomech.2015.04.
038

References
Barker, A.J., Lanning, C., Shandas, R., 2010. Quantication of hemodynamic wall
shear stress in patients with bicuspid aortic valve using phase-contrast MRI.
Ann. Biomed. Eng. 38, 788800.
Bieging, E.T., Frydrychowicz, A., Wentland, A., Landgraf, B.R., Johnson, K.M., Wieben,
O., Francois, C.J., 2011. In vivo three-dimensional MR wall shear stress estimation in ascending aortic dilatation. J. Magn. Reson. Imaging 33, 589597.
Bock, J., Frydrychowicz, A., Stalder, A.F., Bley, T.A., Burkhardt, H., Hennig, J., Markl,
M., 2010. 4D phase contrast MRI at 3T: effect of standard and blood-pool
contrast agents on SNR, PC-MRA, and blood ow visualization. Magn. Reson.
Med. 63, 330338.
Bock, J., Frydrychowicz, A., Lorenz, R., Hirtler, D., Barker, A.J., Johnson, K.M., Arnold,
R., Burkhardt, H., Hennig, J., Markl, M., 2011. In vivo noninvasive 4D pressure
difference mapping in the human aorta: phantom comparison and application
in healthy volunteers and patients. Magn. Reson. Med. 66, 10791088.
Boussel, L., Rayz, V., Martin, A., Acevedo-Bolton, G., Lawton, M., Higashida, R., Smith,
W.S., Young, W.L., Saloner, D., 2009. Phase-contrast magnetic resonance imaging measurements in intracranial aneurysms in vivo of ow patterns, velocity
elds, and wall shear stress: comparison with computational uid dynamics.
Magn. Reson. Med. 61, 409417.
Cecchi, E., Giglioli, C., Valente, S., Lazzeri, C., Gensini, G.F., Abbate, R., Mannini, L.,
2011. Role of hemodynamic shear stress in cardiovascular disease. Atherosclerosis 214, 249256.
Chan, T., Vese, L., 2001. Active contours without edges. IEEE Trans. Image Process.
10, 266277.
Ebbers, T., Wigstrm, L., Bolger, A.F., Engvall, J., Karlsson, M., 2001. Estimation of
relative cardiovascular pressures using time-resolved three dimensional phase
contrast MRI. Magn. Reson. Med. 45, 872879.
Francois, C.J., Srinivasan, S., Schiebler, M.L., Reeder, S.B., Niespodzany, E., Landgraf,
B.R., Wieben, O., Frydrychowicz, A., 2012. 4D cardiovascular magnetic resonance velocity mapping of alterations of right heart ow patterns and main
pulmonary artery hemodynamics in tetralogy of Fallot. J. Cardiovasc. Magn.
Reson. 14, 112.

J. Sotelo et al. / Journal of Biomechanics 48 (2015) 18171827

Frydrychowicz, A., Stalder, A.F., Russe, M.F., Bock, J., Bauer, S., Harloff, A., Berger, A.,
Langer, M., Hennig, J., Markl, M., 2009a. Three-dimensional analysis of segmental wall shear stress in the aorta by ow sensitive four-dimensional-MRI. J.
Magn. Reson. Imaging 30, 7784.
Frydrychowicz, A., Berger, A., Stalder, A.F., Markl, M., 2009b. Preliminary results by
ow sensitive magnetic resonance imaging after Tiron David I procedure with
an anatomically shaped ascending aortic graft. Interact. Cardiovasc. Thorac.
Surg. 9, 155158.
Frydrychowicz, A., Francois, C.J., Turski, P.A., 2011. Four-dimensional phase contrast
magnetic resonance angiography: potential clinical applications. Eur. J. Radiol.
80, 2435.
Gizzi, A., Bernaschi, M., Bini, D., Cherubini, C., Filippi, S., Melchionna, S., Succi, S.,
2011. Three-band decomposition analysis of wall shear stress in pulsatile ows.
Phys. Rev. E 83 (3), 31902.
Goubergrits, L., Riesenkampff, E., Yevtushenko, P., Schaller, J., Kertzscher, U., Berger,
F., Kuehne, T., 2014. Is MRI-based CFD able to improve clinical treatment of
coarctations of aorta? Ann. Biomed. Eng. 43 (1), 168176.
Harloff, A., Nubbaumer, A., Bauer, S., Stalder, A.F., Frydrychowicz, A., Weiller, C.,
Hennig, J., Markl, M., 2010. In vivo assessment of wall shear stress in the
atherosclerotic aorta using ow-sensitive 4D MRI. Magn. Reson. Med. 63,
15291536.
Heimsund, B., Tai, X., Wang, J., 2002. Superconvergence for the gradient of nite
element approximations by L2 projection. SIAM J. Numer. Anal. 40, 12631280.
Hinton, E., Campbell, J.S., 1974. Local and global smoothing of discontinuous nite
element functions using a least squares method. Int. J. Numer. Methods Eng. 8,
461480.
Hurtado, D.E., Kuhl, E., 2014. Computational modelling of electrocardiograms:
repolarisation and T-wave polarity in the human heart. Comput. Methods
Biomech. Biomed. Eng. 17 (9), 986996.
Hurtado, D.E., Henao, D., 2014. Gradient ows and variational principles for cardiac
electrophysiology: toward efcient and robust numerical simulations of the
electrical activity of the heart. Comput. Methods Appl. Mech. Eng. 273,
238254.
Kafka, H., Mohiaddin, R.H., 2009. Cardiac MRI and pulmonary MR angiography of
sinus venous defect and partial anomalous pulmonary venous connection in
cause of right undiagnosed ventricular enlargement. Am. J. Roentgenol. 192,
259266.
LaDisa Jr., J.F., Figueroa, C.A., Vignon-Clementel, I.E., Kim, H.J., Xiao, N., Ellwein, L.M.,
Chan, F.P., Feinstein, J.A., Taylor, C.A., 2011. Computational simulations for aortic
coarctation: representative results from a sampling of patients. J. Biomech. Eng.
133 (9), 091008.
Markl, M., Harloff, A., Bley, T.A., Zaitsev, M., Jung, B., Weingan, E., Langer, M.,
Hennig, J., Frydrychowicz, A., 2007. Time resolved 3D MR velocity mapping at
3T: improved navigator-gated assessment of vascular anatomy and blood ow.
J. Magn. Reson. Imaging 25, 824831.
Markl, M., Wegent, F., Zech, T., Bauer, S., Strecker, C., Schumacher, M., Weiller, C.,
Hennig, J., Harloff, A., 2010. In vivo wall shear stress distribution in the carotid
artery: effect of bifurcation geometry, internal carotid artery stenosis, and
recanalization therapy. Circ. Cardiovasc. Imaging 3, 647655.
Morgan, V.L., Graham Jr., T.P., Roselli, R.J., Lorenz, C.H., 1998a. Alterations in pulmonary artery ow patterns and shear stress determined with three-

1827

dimensional phase contrast magnetic resonance imaging in Fontan patients. J.

Thorac. Cardiovasc. Surg. 116, 294304.
Morgan, V.L., Roselli, R.J., Lorenz, C.H., 1998b. Normal three-dimensional pulmonary
artery ow determined by phase contrast magnetic resonance imaging. Ann.
Biomed. Eng. 26, 557566.
Oden, J.T., Brauchli, H.J., 1971. On the calculation of consistent stress distributions in
nite element applications. Int. J. Numer. Methods Eng. 3, 317325.
Oshinski, J.N., Ku, D.N., Mukundan, S., Loth, Jr, Pettigrew, RI., F., 1995. Determination
of wall shear stress in the aorta with the use of MR phase velocity mapping. J.
Magn. Reson. Imaging 5, 640647.
Oyre, S., Ringgaard, S., Kozerke, S., Paaske, W.P., Erlandsen, M., Boesiger, P., Pedersen, E.M., 1998. Accurate noninvasive quantitation of blood ow, cross-sectional lumen vessel area and wall shear stress by three-dimensional paraboloid
modeling of magnetic resonance imaging velocity data. J. Am. Coll. Cardiol. 32,
128134.
Petersson, S., Dyverfeldt, P., Ebbers, T., 2012. Assessment of the accuracy of MRI wall
shear stress estimation using numerical simulations. J. Magn. Reson. Imaging
36, 128138.
Stalder, A.F., Russe, M.F., Frydrychowicz, A., Bock, J., Hennig, J., Markl, M., 2008.
Quantitative 2D and 3D phase contrast MRI: optimized analysis of blood ow
and vessel wall parameters. Magn. Reson. Med. 60, 12181231.
Taylor, C.A., Figueroa, C.A., 2009. Patient-specic modeling of cardiovascular
mechanics. Annu. Rev. Biomed. Eng. 11, 109134.
Tyszka, J.M., Laidlaw, D.H., Asa, J.W., Silverman, J.M., 2000. Three-dimensional,
time-resolved (4D) relative pressure mapping using magnetic resonance imaging. J. Magn. Reson. Imaging 12, 321329.
Uribe, S., Beerbaum, P., Sorensen, T.S., Rasmusson, A., Razavi, R., Schaeffter, T., 2009.
Four-dimensional (4D) ow of the whole heart and great vessels using realtime respiratory self-gating. Magn. Reson. Med. 62, 984992.
Wigstrm, L., Ebbers, T., Fyrenius, A., Karlsson, M., Engvall, J., Wranne, B., Bolger, A.
F., 1999. Particle trace visualization of intracardiac ow using time-resolved 3D
phase contrast MRI. Magn. Reson. Med. 41, 793799.
Weigang, E., Kari, F.A., Beyersdorf, F., Luehr, M., Etz, C.D., Frydrychowicz, A., Harloff,
A., Markl, M., 2008. Flow-sensitive four-dimensional magnetic resonance
imaging: ow patterns in ascending aortic aneurysms. Eur. J. Cardio-Thorac.
Surg. 34, 1116.
Womersley, J.R., 1955. Method for the calculation of velocity, rate of ow and
viscous drag in arteries when the pressure gradient is known. J. Physiol. 127,
553563.
Xiao, N., Humphrey, J.D., Figueroa, C.A., 2013. Multi-scale computational model of
three-dimensional hemodynamics within a deformable full-body arterial network. J. Comput. Phys. 244, 2240.
Zienkiewicz, O.C., Zhu, J.Z., 1992. The superconvergent patch recovery and a posteriori error estimates. Part 1: the recovery technique. Int. J. Numer. Methods
Eng. 33, 13311364.
Zienkiewicz, O.C., Taylor, R.L., Zhu, J.Z., 2005. The Finite Element Method: its Basis
and Fundamentals, 6th ed. Butterworth-Heinemann, London, UK.