Biomech Model Mechanobiol (2018) 17:465–477

https://doi.org/10.1007/s10237-017-0972-9

ORIGINAL PAPER

Simulation of the human airways using virtual topology tools

and meshing optimization
A. Fernández-Tena1 · A. C. Marcos2 · R. Agujetas3 · C. Ferrera3

Received: 11 September 2017 / Accepted: 24 October 2017 / Published online: 5 November 2017
© Springer-Verlag GmbH Germany 2017

Abstract A method is proposed to improve the quality of 1 Introduction

the three-dimensional airway geometric models using a com-
mercial software, checking the number of elements, meshing
time, and aspect ratio and skewness parameters. The use of
real and virtual topologies combined with patch-conforming
and patch-independent meshing algorithms results in four
different models being the best solution the combination of
virtual topology and patch-independent algorithm, due to
an excellent aspect ratio and skewness of the elements, and
minimum meshing time. The result is a reduction in the com-
putational time required for both meshing and simulation due
to a smaller number of cells. The use of virtual topologies
combined with patch-independent meshing algorithms could
be extended in bioengineering because the geometries han-
dling is similar to this case. The method is applied to a healthy
person using their computed tomography images. The result-
ing numerical models are able to simulate correctly a forced
spirometry.
Keywords Human airways · CT images · CFD · Virtual
topology · Meshing optimization
B C. Ferrera
cfll@unex.es
1 the lung morphology in a simple way considered a symmet-
Universidad de Oviedo and Hospital Universitario Central de
Asturias, 33011 Oviedo, Spain
2 1963). This model indicates the bifurcation modes, designing
Dpto. de Expresión Gráfica, Universidad de Extremadura,
06006 Badajoz, Spain
3 that each airway generates two symmetrical branches (regu-
Dpto. de Ingeniería Mecánica, Energética y de los Materiales
and Instituto de Computación Científica Avanzada (ICCAEx),
Universidad de Extremadura, 06006 Badajoz, Spain
In the European Union and USA, total health spending will
increase as the population grows older (OECD/UE 2016;
Przywara 2010). A big percentage of these expenses is
due to respiratory diseases (Gibson et al. 2013), such as
asthma and COPD (chronic obstructive pulmonary diseases).
The increasing cost both in spending and human lives has
encouraged European and US administrations to invest in
developing different methods of drug-aerosol delivery (Lip
and Philip 2015; Ashurst et al. 2000). This delivery must
be targeted in order to increase the efficiency of the treat-
ment and to minimize its secondary effects (Kleinstreuer
et al. 2008). To reach this goal, it is crucial to understand
the airflow dynamics in the human airways. The Compu-
tational Fluid Dynamics (CFD) is a useful tool which can
provide us valuable data of the air movement in lungs. It is a
non-invasive technique, relatively inexpensive and can study
unusual or dangerous performances with the only require-
ment of a precise geometry. In order to get results, one must
accomplish three processes: (i) obtaining the topology of the
human airways, (ii) meshing the geometry, (iii) choosing a
simulation model and running to obtain the results.
Concerning the morphology of a lung, it consists of a series
of dichotomous bifurcations, each of them producing a new
generation, which results in a highly complex branching net-
work. As stated previously, it is quite difficult to describe
geometrically this network. The first attempt to characterize
rical bifurcation of the geometry at each bronchus (Weibel
the trachea as the first airway (generation 0) and assuming
lar dichotomy). So, the left and right main bronchi are order
1 and so on. There are at least 23 airway generations until

123
466
the alveolar sacs. In this model, it was not taken into account
the spatial disposition of the branches; therefore, it is consid-
ered a model that only allows modelling the lung function in
one dimension. More general approaches for complex rep-
resentation have been proposed. The algorithmic techniques
(Kitaoka et al. 1999) carry out the design of a realist model
of the lung, following a set of basic rules.
Although relevant information can be extracted from sim-
ulations made with the previous models (Fernández-Tena
and Casan Clarà 2015; Fernández-Tena et al. 2017a), real-
istic models are required to get results applicable to clinical
situations. For this reason, idealized shapes are giving way
to patient-specific geometries reconstructed from medical
imaging techniques, such as magnetic resonance (MR) and
computer tomography (CT). In this case, one of the major
problems is the reconstruction of the intricate local anatomy
delimiting the fluid domain, which is precisely the main cause
of complex flows in lungs. Nowadays, upper airways (Lin
et al. 2007), until the sixth generation (Choi 2007), carti-
laginous rings (Zhang and Finlay 2005), mucus (Inthavong
et al. 2010; Islam et al. 2015) and vascular models (Zheng
2014) have been included. A combination from CT images
and idealized models is presented in Marchandise et al.
(2013) and Miyawak et al. (2017), using an open-source
software. Both cases propose a fully automatic procedure
for the mesh generation of tubular geometries such as blood
vessels or airways based on centrelines. Therefore, progress
in studying the airways as a result of modern imaging
techniques (MRI, SPECT—single photon emission CT) is
opening doors to wider use of CFD simulations as: compar-
ing with real measurements (De Backer et al. 2010; Rochefort
et al. 2007), considering unsteady conditions (Taherian et al.
2011), assessing the effects of different drugs in asthma (De
Backer et al. 2008) and COPD (De Backer et al. 2012) and
detecting the local airways resistance in patients with sev-
eral treatments (Vos et al. 2013) and analyse the influence
in the deposition of drugs of rings present in both trachea
and the first generation airways (Kannan et al. 2017b). Nov-
els models, as the quasi-3D wire (Kannan et al. 2017a), are
also introduced obtaining results faster than traditional CFD
schemes. In spite of all these advances, the presence of differ-
ent scales and the intrinsic fractal character (Weibel 2009),
force any new study to refine the segmentation of the CT
images to obtain a geometry that closely matches the real
one. So the resolution of CT images must be increased dur-
ing the segmentation process in order to get a geometry with
the most important details and easy to manipulate and to
calculate.
Another important feature is the airways wall deforma-
tion. It is well known that airways significantly expand and
contract during the inhalation and exhalation phases, respec-
tively. These deformations in the airways wall modify the
flow behaviour. The analyses of the fluid structure interac-
123
A. Fernández-Tena et al.
tion (FSI) in complexes geometries are scarce. Jahangiri et al.
2015 conducted an FSI research on stenosed arteries and
assessed that the error assuming laminar flow instead of tur-
bulent flow is bigger than considering a rigid wall instead
of a flexible one. The assumption of a rigid wall can lead
to a good correlation between experiments and simulations
(Agujetas et al. 2017). Moreover, an FSI analysis of flow in
human airways would require the segmentation of multiple
images during the respiratory cycle and a huge computational
cost. For all these reasons, in this work, we neglect the defor-
mations associated with the wall and concentrate our efforts
in a fixed geometry.
Once the geometry is obtained, it is necessary to create
a numerical mesh and to adapt it to the geometry with-
out adding irregularities. For a complex geometry, there are
different ways of meshing: (a) increasing the number of
elements in non-structured (Kannan et al. 2016; Kolanjiyil
et al. 2017; Vos et al. 2013; Zhang et al. 2012) or struc-
tured (Hörschler et al. 2003) meshes, (b) developing a specific
meshing algorithm (Dyedov et al. 2009) or (c) using com-
mercial algorithms combined with virtual topologies. These
ways have both advantages and disadvantages. Introducing
tiny elements both adapts the numerical mesh to the geome-
try and increase their quality. However, the computing time
grows as the number of elements is higher. Constructing an
algorithm can produce a very good mesh, but it is quite dif-
ficult and limited to experts in mesh development. Virtual
topologies can reduce the number of elements to be pro-
cessed by merging the smaller triangular surfaces resulting
from the segmentation process, but this methodology has
only been successfully applied to geometries involved in
industrial processes (Streinbenner et al. 2000). As virtual
topologies are included in both commercial and open source
software packages (Fedorov et al. 2012) and do not alter
the original geometry (Foucault et al. 2007), its application
in the study of a complex geometry could provide valuable
results. A successful application of this methodology could
avoid the disadvantages of increasing the computational time
or the development of a specific algorithm for each case of
complex geometry.
After the mesh generation is completed, we must perform
the simulation. The use of a direct numerical simulation
(DNS) or a large Eddy simulation (LES) would provide
exact or very accurate predictions, respectively. These simu-
lations have been used to solve idealized stenotic geometries
(Varghese et al. 2007a, b; Tan et al. 2011). The computing
effort is huge for low Reynolds numbers but, to the best
of our knowledge, DNS or LES have not been applied to
a complex geometry. Although some authors question their
applicability to complex flows (Scotti and Piomelli 2002), the
Reynolds-averaged Navier–Stokes (RANS) equations have
been applied to get results without consuming too much com-
puting time (Kleinstreuer 2006). SST-k-ω turbulent model
Simulation of the human airways using virtual topology tools and meshing optimization
Fig. 1 Pulmonary CT images performed with a 64 slices volumetric scanner (Light speed VCT, GE medical systems). (Left) coronal plane, (centre)
axial plane, (right) sagittal plane. Measurements guidelines: collimation: 0.6 mm; pitch: 1.1; rotation time: 0.6 s; tube voltage: 120 kV
Fig. 2 Representation of
triangulation process of an
airway branch bifurcation
(Menter et al. 2006) can produce useful results in clinical
situations when it simulates a real geometry (Agujetas et al.
2017).
This work seeks a reduction in the time required for
the complete process that includes the reconstruction of the
geometry from CT images and its meshing. Our main objec-
tive is to assess the use of virtual topologies in combination
with the patch-independent algorithm to mesh a real human
airways and reflect their advantages. The lower number of
elements will simplify the mesh, and it will have a positive
impact in the time required for the simulation.
The paper is organized as follows. The different methods
employed in this work, i.e. the segmentation and the mesh-
ing processes, the experiments performed and the simulations
executed are described in Sect. 2. The results of the applica-
tion of those methods are discussed and presented in Sect. 3.
Finally, the paper closes in Sect. 4 with some conclusions.
2 Methods
2.1 Numerical model
The numerical model will be reconstructed from the airways
CT images (Fig. 1) of a patient without chronic obstruc-
467
tive pulmonary disease (COPD) obtained at the Hospital
IC in Badajoz-Spain. The images were exported to DICOM
(Digital Imaging and Communication in Medicine) format.
Their resolution is 512 × 512 pixels, and the pixel size is
680 µm. Each pixel has a grey intensity value according to
the Hounsfield greyscale (Hounsfield 1973). There is a direct
relationship between the density of each anatomical structure
and the grey value assigned to each pixel in the image. As all
human tissues are classified under the Hounsfield scale, the
3D Slicer software (3D Slicer 2014) was used to group similar
grey values, identifying the threshold between the different
tissues and extracting the human airways. Once the segmen-
tation process has finished, the cloud of points (Fig. 2a)
extracted from the images can be joined with lines (Fig. 2b),
being generated triangular faces with a size inversely pro-
portional to the curvature of the different parts (Fig. 2c). In
addition, the scale of each generation is greater than the next
one. This factor is very important as this difference increases
sharply between non-consecutive generations. With these
image resolutions, the 3D Slicer software can rebuild the
lung up to the fourth generation.
One crucial aspect in CFD studies is the mesh resolution. A
reduced size of the elements (high number of cells) increases
the computational time and does not always lead to a better
solution. The influence of the edge length in the meshing
123
468
Fig. 3 Influence of the edge length in the meshing process
Fig. 4 a Sphere with multiple elements, b meaningful and c full reduction of the number of faces in a mesh using virtual topology
process can be seen in the Fig. 3. This figure shows two sim-
ilar geometries (a and b). However, if the edge is zoomed, the
Fig. 3c is a perfect circle and the Fig. 3d is a 36-sided polygon.
If the edge length is very small one obtains a mesh with mul-
tiple elements (Fig. 3e). The detailed geometry is difficult
to manipulate and a very fine mesh needs a lot of calcula-
tions increasing the time to converge a solution. Kannan et al.
2017b pointed that the convergence of the numerical algo-
rithms on such fine meshes can be bad and proposed a model
to overcome this issue. On the other hand, if one decreases
the number of elements, then the mesh is considerably sim-
plified (Fig. 3f). In this case, the perfect geometry (Fig. 3a)
is treated numerically as a simplified geometry (Fig. 3b) so
a loss in the accuracy is due to a bad mesh. So instead of
wasting resources meshing a very precise geometry with big
elements we must mesh the simplified one. Nevertheless, as
stated in the introduction, it is a usual practice to increase
the resolution either the geometry or the mesh with commer-
cial algorithms. We propose applying a virtual topology to
address the problem of meshing the human airways. The goal
of any simulation is to reduce the number of elements and
edges while getting a model closer to the real one. Edges have
123
A. Fernández-Tena et al.
a strong influence on the meshing process as they increase its
complexity. The use of a virtual topology can reduce them
when they are not necessary, as it can see in Fig. 4. This figure
is an example of the generation of different virtual surfaces
over the different faces that form a sphere. Figure 4a shows a
sphere meshed with multiple faces while Fig. 4b, c show the
same sphere meshed with significant and full reduction of
the number of faces using virtual topology. Nevertheless, the
underlying geometry remains intact. Nowadays, commercial
software (Ansys 2015) provides meshing tools to perform
these operations.
Figure 5 (Ansys 2015) shows the method of generating
virtual surfaces, which consists of joining the triangular sur-
faces as a function of the Gauss Curvature Angle that form
two contiguous surfaces (an angle between 0◦ and 180◦ can
be selected). The larger the angle of curvature selected, the
more surfaces will be joined on fewer virtual surfaces. The
best results are achieved for angles between 20◦ and 120◦ . If
the angle is set below this range, few faces will be merged,
while if the value is above this range could result in prob-
lems with the mesh. Figure 6 (Ansys 2015) shows the virtual
Simulation of the human airways using virtual topology tools and meshing optimization
Fig. 5 Definition of the Gauss curvature angle (Ansys fluent)
surfaces made in a machining geometry using three different
Gauss Curvature Angles (25◦ , 60◦ and 120◦ ).
In lung geometry, edges are not important like in machin-
ing; then, it is essential to reduce its number. The Ansys
meshing tool allows this process to be performed automat-
ically, but the best results are obtained in a manual way
(without the requirement to define a normal for a multiple
Fig. 6 Gauss curvature angle at a 25◦ , b 60◦ and c 120◦ (Ansys fluent)
Fig. 7 Virtual topologies generated over the segmented geometry. a Automatic generated mesh with the smallest and b biggest element sizes, c
manually generated mesh to increase the size of the surfaces
469
joined surface) because the order in which the surfaces are
joined determines the maximum size that can be obtained
for each virtual surface. In addition, it also allows to leave
small surfaces or significant edges to be taken into account
by meshing algorithms. Figure 7 shows three different virtual
topologies, two of them are automatically generated with dif-
ferent surfaces size (Fig. 7a has smaller surfaces than Fig. 7b),
and the third shows the case made manually (Fig. 7c). With
the manual process, the lung surface can be separated by
levels (Fig. 7c), which will allow to complete studies, such
as particles deposition (Fernández-Tena et al. 2016), deter-
mining the generation of the lung where the particles are
deposited. Simultaneously, the number of edges and, as a
consequence, the mesh that is going to be constructed is
less complex, and a reduced time is needed to perform this
process. This case gets the best results because the num-
ber of surfaces can be reduced, respecting important details
in the surface, such as the presence of a polyp in a branch
(Fig. 8). The absence of cartilaginous rings in the airways of
this patient can also be observed. As the patient is a small
child, the rings have not been developed yet.
123
470
Fig. 8 Detail of the virtual surfaces generated around a polyp in a
branch
Fig. 9 Detail of the interior, with the tetrahedral and the boundary layer
meshing
Tetrahedra have been selected due to the complexity of
the bronchial tree, so we are going to test methods com-
patible with that element. These methods are named patch-
conforming and patch-independent and they employ the
biggest elements possible to save processing time. The main
difference between both methods; it is founded on how they
capture the irregularities in the surface. The study of the tur-
bulent flow present in high frequency breathing imposes the
meshing of the boundary layer, so the way the surface details
are meshed can influence and hinder the boundary layer
mesh. Figure 9 shows a branch, including its bifurcation, with
this type of mesh, an external boundary layer mesh composed
of triangular prisms and a tetrahedral mesh in the interior. The
patch-conforming algorithm takes into account all edges and
faces of surface, while the patch-independent algorithm cre-
ates a mesh that does not necessary conform to every edge
or face (they can be defeatured within a tolerance margin).
Patch-conforming needs to have a large number of elements
to avoid deforming cells while patch-independent, with fewer
elements, provides a good model with small inaccuracies and
it is not limited by the underlying geometry. In this test, the
chosen parameters in the construction of the mesh are the
same in both algorithms: maximum and minimum length of
each side are 0.8 and 0.4 mm, the algorithm used to place the
cells is the proximity function, and five inflation layers are
123
A. Fernández-Tena et al.
used as the mesh close to the geometrical surface must be
structured and there must be a good transition.
The quality of the mesh is assessed using the skewness and
aspect ratio parameters (Ansys 2015). The skewness mea-
sures how much one face of the mesh is close to an ideal cell
with equal sides or angles. The aspect ratio is a measure of
the stretching of a cell, being computed as the ratio of the
maximum value to the minimum value of any of the follow-
ing distances: the normal distances between the cell centroid
and face centroids and the distances between the cell centroid
and nodes.
The combination of topologies and meshing algorithms
results in four different models: real topology and patch-
conforming, real topology and patch-independent, virtual
topology and patch-conforming and virtual topology and
patch-independent. Table 1 shows the results, performed on
a PC with an Intel Core i7-6700HQ processor at 2.6 GHz
and 16 Gb of RAM. Figures 10 and 11 show the skewness
distribution over the wall and the interior of a human lung
for the four models. For a better understanding of the mesh
(Fig. 10), in each sketch, there are two labels showing where
the maximum (red colour) and minimum (blue colour) skew-
ness values are placed. In Fig. 11 one can observed the detail
of the two types of meshing: boundary layer at the wall and
tetrahedra inside.
2.2 Experimental
In addition to the geometry, the particle deposition is influ-
enced by the mass median aerodynamic diameters, the
difference in the velocity of the air in the different branches,
the humidity in the lungs, the tidal volume, the flow rate and
the different typology of diseases affecting the respiratory
system (Nahar et al. 2013). Studies ex vivo, in vitro and in
vivo can be found in the bibliography. Ex vivo studies are
mostly employed in animals. In vitro are limited to a cul-
ture of lung cells. Both type of studies are focused on the
analysis of drug absorption and transport. So, the best option
to localize the position of the particles are in vivo studies
(Nahar et al. 2013) because they not only provide the posi-
tion in the different regions of the lung but also they assess
the clinical effects of the drugs. Concerning the transport of
the drugs, in vivo studies have been performed in animals.
Concerning the position of the particles, there are studies
in humans (Stahlhofen et al. 1989) where the particles are
located by labelling the aerosol with a radiotracer and imag-
ing the total lung deposition and the fraction remaining in the
airways. Nevertheless, these types of test in humans require a
very expensive equipment, a specialized staff to perform the
experiments and a specific permission from the authorities
to perform experiments in humans with radioactive particles.
As our group do not have access to the experimental equip-
ment necessary to perform these test, a good alternative could
Simulation of the human airways using virtual topology tools and meshing optimization 471

Table 1 Comparison between the four different meshes

Model Number of elements Meshing time (min)

Real topology and patch-conforming 462,279 <5

Real topology and patch-independent 797,180 120
Virtual topology and patch-conforming 978,816 30
Virtual topology and patch-independent 544,946 <5

Model Aspect ratio

Min Max Mean Standard deviation

Real topology and patch-conforming 1.18E+00 7.72E+03 5.47E+00 1.35E+01

Real topology and patch-independent 1.18E+00 7.75E+01 5.00E+00 3.29E+00
Virtual topology and patch-conforming 1.17E+00 2.67E+01 5.17E+00 2.84E+00
Virtual topology and patch-independent 1.18E+00 3.19E+01 4.92E+00 3.00E+00

Model Skewness
Min Max Mean Standard deviation

Real topology and patch-conforming 8.96E−04 1.00E+00 3.39E−01 1.60E−01

Real topology and patch-independent 1.43E−03 9.92E−01 2.44E−01 1.46E−01
Virtual topology and patch-conforming 5.99E−05 9.60E−01 2.36E−01 2.03E−01
Virtual topology and patch-independent 3.44E−04 9.56E−01 2.24E−01 1.50E−01

be a forced spirometry. These test is also an in vivo test and
can be used to validate the models here proposed.
In this test, the patient only has to make several inhalations
and exhalations with a flowmeter placed in his mouth. To
obtain spirometry data to be used as a reference for compar-
ison, several forced spirometry tests will be conducted in the
patient, and the spirometry that best matched the ERS quality
criteria (Miller et al. 2005) is chosen. The spirometry data
is obtained by means of a spirometer model CPFS/D USB,
with a MedGraphics preVentTM pneumotachometer and
BREEZESUITETM diagnostic software (Medical Graphics
Corporation 2004, 350 OakGrove Parkway, St. Paul, Min-
nesota 55127-8599). From the obtained data, the relationship
between flow rate and volume versus time was represented
in Fig. 12. It can be seen that after 3 normal breathing cycles,
the portion corresponding to the manoeuvre of the forced
spirometry itself is between 11.14 and 16.70 s. The forced
spirometry has a range of flow rate of − 312 L/min (22.7 m/s)
and 348 L/min (−20.6 m/s), in a time of 5.56 s, being the rate
of change of the flow rate (velocity) very dependent with the
considered time interval.
2.3 Simulations
To know the goodness of the model, two types of simula-
tions (steady and transient cases) have been performed. In
each case, the RANS equations are integrated in the unsteady
and incompressible regime with the finite volume method
(Versteeg and Malalasekera 2007) implemented in the com-
mercial software Fluent (Ansys 2015). The walls are rigid,
and the fluid is Newtonian. The turbulent stresses are calcu-
lated with the SST-k-ω model (Menter et al. 2006; Wilcox
2006). Air was the working fluid with a constant density of
1.225 kg/m3 and dynamic viscosity of 1.7894×10−5 kg/m/s.
On the other side, the non-slip boundary condition is pre-
scribed at the airway inner surface. Some values of 5% for
the turbulence intensity at both the inlet and outlet sections,
and 10% for the turbulent viscosity ratio at the inlet section
are fixed. The wall turbulence conditions are the standard
(Ansys 2015) and all y+ values fell into the laminar sublayer,
most of them within the interval 0 < y+ < 1. The numerical
integration of the hydrodynamic equations is carried out with
the pressure-based solver, and the gradients in the cell centre
and faces are evaluated with the Green–Gauss node-based
scheme and a multidimensional Taylor expansion, respec-
tively. The spatial discretization of the pressure equation is
conducted with the second-order approximation, while the
momentum equations are discretized with the second-order
upwind scheme. The velocity-pressure coupling is conducted
with the SIMPLEC procedure. Normalized residuals levels
(measured by a reduction in all flow residuals by at least
three orders of magnitude, as well as a relative change in
measured parameters, pressure and flow rate in select air-
ways) were examined, accepting a convergence with criteria
of 10−5 residuals.
In steady mode, two simulations were executed to know
the velocities and pressure drop in the branches. Each sim-
ulation corresponds to the inhalation and exhalation phases

123
472 A. Fernández-Tena et al.

Fig. 10 Skewness distribution over the wall of a human airways for the different applied models: a real topology and patch-conforming, b real
topology and patch-independent, c virtual topology and patch-conforming, d virtual topology and patch-independent

of a person whose breathing condition is a mixed one, with
a value of 28 L/min, between sedentary and heavy-activity
(Hofmann et al. 1989; Sbirlea-Apiou et al. 2007). A uniform
velocity profile compatible with the measured flow rate and
normal to the inlet is imposed at that section. Pressure out-
let boundary conditions are set at the outlets sections. They
were run to full convergence (Fig. 13), indicating that a steady
state condition had been achieved, showing the Fig. 15 the
obtained results (average velocity and pressure drop in each
generation).
For transient mode, we think that the best method is to
simulate a forced spirometry, which has already been tested
in previous works by the authors (Fernández-Tena 2014;
Fernández-Tena et al. 2017b). The simulation consists of
reproducing the forced spirometry (Fig. 12) with the same
conditions that in the experimental test. The boundary con-
ditions imposed to the model were the flow rate varying
over time at the inlet (generation 0), and a static pressure
(at the inhalation) and a total pressure (at the exhalation) at
the outlet (generation 4). The rate of change of the flow rate
is very dependent with the considered time interval, which
suggests that time points corresponding to simulation time
steps should be clustered closer together when the rate of
change in mass flow rate is large and more sparsely when

123
Simulation of the human airways using virtual topology tools and meshing optimization 473

Fig. 11 Skewness quality in the interior of the human airways: a real topology and patch-conforming, b real topology and patch-independent, c
virtual topology and patch-conforming, d virtual topology and patch-independent

the rate of change is small. For this reason, to reproduce the

variation of flow rate over time, it was used a special adaptive
time stepping (Fernández-Tena et al. 2017b), with a constant
step in the relevant dependent variable (flow rate, pressure)
rather than a constant step in independent variable (time). To
check that these obtained pressure values are correct what it
does is to repeat the spirometry test but, in this case, using
the previously obtained unsteady total pressure (sum of static
and dynamic pressure) as inlet boundary condition (genera-
tion 4) and pressure outlet as outlet condition (generation 0).
Figure 16 shows experimental flow rate at generation 0 and Fig. 12 Forced spirometry, with a flow rate range of − 312 L/min and
the obtained numerical pressure at generation 4. Figure 17 348 L/min in 5.56 s
shows experimental and the obtained numerical flow rate at
generation 0.
ters. The meshing time and the number of elements used are
the variables to be optimized. The skewness quality of the
3 Results and discussion four meshes from a global point of view is good (Fig. 10);
however, it is possible that some of those four meshes may
3.1 Meshing have several deformed cells that affect to the results of
the simulation. Concerning the real topology and patch-
The quality of the mesh, as mentioned in the Sect. 2.1, is conforming model (Figs. 10a, 11a), the Table 1 shows quite
tested using the aspect ratio and the skewness as parame- good results as it has minimum number of cells, minimum

123
474
Fig. 13 Convergence: a real topology and patch-conforming, b real topology and patch-independent, c virtual topology and patch-conforming, d
virtual topology and patch-independent
meshing time and acceptable mean values. Nevertheless, if
one looks carefully the minimum and maximum values, there
are several cells that are quite deformed and are worsen-
ing the mesh. Also, the algorithm fails to create boundary
layer meshing over the entire surface, as can be seen in the
Fig. 11a. The real topology and patch-independent model
(Figs. 10b, 11b) has the next worse results. The mean val-
ues are improved slightly in spite of an increment of nearly
75% in the number of cells. The consequence is a drastic rise
of the time needed to complete the process. The quality of
the mesh is further improved when using virtual topologies.
The combination between a virtual geometry and the patch
conforming algorithm (Figs. 10c, 11c) yields a larger num-
ber of cells and a higher meshing time than the combination
123
A. Fernández-Tena et al.
of the same algorithm with a real geometry. Despite that,
the most deformed cells disappear as the maximum value of
the aspect ratio decreases drastically. Finally, the best results
are achieved combining a virtual topology and the patch-
independent algorithm (Figs. 10d, 11d). Both mean aspect
ratio and skewness values are further improved, while the
number of elements and the meshing time is reduced.
3.2 Simulation
The same tendency appears when simulations are conducted.
As it can be seen in Fig. 13, the case (a) Real topology and
patch-conforming does not converge due to presence of cells
with very poor Aspect Ratio and Skewness. Convergence
Simulation of the human airways using virtual topology tools and meshing optimization
Fig. 14 Convergence of the selected variable p_0 for three different
meshes
Fig. 15 Flow rate and pressure drop in the conductive region during
steady state simulations of inhalation and exhalation
Fig. 16 Unsteady inlet flow rate imposed at generation 0 in the simu-
lation and static pressure difference obtained in generation 4
in the other three cases is good, being the fastest, 10 min,
(d) Virtual topology and patch-independent. Cases (b) Real
topology and patch-independent and (c) Virtual topology and
patch-conforming spend 50 and 90% more time. Because the
results of the three cases have differences lower than 0.1%, in
order to avoid confusion and from now to on, we focus only
the best case, (d) Virtual topology and patch-independent.
Concerning the dependence of the values of a variable
on the number of the elements of the cell, we have set the
inlet pressure (p_0) as the control parameter. We have run
three different simulations. The first one for the optimized
mesh with 544946 cells, a second one duplicating the num-
ber elements and a third one quintuplicating the number of
475
Fig. 17 Flow rate and pressure drop in the conductive region during
steady state simulations of inhalation and exhalation
elements of the optimized mesh. Figure 14 shows that after
a few iterations the convergence value of 25 Pa is reached
by the different models. The variation observed in the pres-
sure drop was less than 0.67% between the 544,946 cells
model and the model five times higher and 0.45% with the
twice superior model. Therefore, the small improvement in
the accuracy of the result with the increase in the mesh does
not compensate the computational time spent.
In steady mode (Fig. 15), velocities and pressures have
similar values in both cases, inhalation and exhalation. As
expected, the velocity increases in the first three generations
of the airways as the total cross-sectional area decreases,
and then it gradually decreases in the remaining generations,
since the total cross-sectional area of the lung is globally
increasing. Static pressure values obtained during the inhala-
tion and exhalation cases have a similar tendency but with
differences in the upper airway generations. During inhala-
tion, there is initially a more rapid increase in pressure as
generation number increases than in the case of expiration.
This difference occurs in the region for which global mean
velocity is increasing with increasing generation number. The
trend reverses after generation 4, i.e. when the global mean
velocity begins to decrease (Fernández-Tena et al. 2017a).
These results agree with those obtained by Hofmann et al.
(1989) and Sbirlea-Apiou et al. (2007). For velocity, the mean
difference over all generations for both inhalation and exha-
lation is 0.8%. For average pressure, the mean difference is
1.1%. The results of the steady simulations verify that the
model is able to reproduce physiologically realistic flow fea-
tures of the lung airway during both steady inhalation and
exhalation.
In transient mode (Fig. 16), during the exhalation cycle,
which occurs during the first 4.6 seconds of the test, the
pressure at generation 4 increases sharply in accord with the
sudden exhalation of breath, then relaxes towards zero as the
exhalation rate decreases. Similarly, during the inhalation
phase, the pressure decreases sharply during rapid inhala-
123
476
tion and relaxes towards zero as the flow rate decreases. The
results of the flow rate obtained at generation 0 for both
cases (Fig. 17), the numerical and experimental tests, are
very similar. That is, the maximum flow rate occurs at time
0.17 s, with a predicted value of 344.561 L/min, compared to
an experimentally obtained value of 345.936 L/min. There-
fore, the maximum error is less than 0.4%, which suggests
that the method presented is capable of faithfully reproduc-
ing physiologically realistic results under realistic ventilation
conditions.
4 Conclusions
To summarize, the use of virtual topologies to rebuild a
human lung from CT images of a patient was made. The
three-dimensional geometry was meshed using a combina-
tion of real and virtual topologies with patch-conforming
and patch-independent algorithms giving rise to four models:
(a) Real topology and patch-conforming, (b) Real topology
and patch-independent, (c) Virtual topology and patch-
conforming, (d) Virtual topology and patch-independent.
These models were tested:
– Using skewness and aspect ratio as parameters. The
virtual topology and the patch-independent algorithm
provide the lowest values for both skewness and aspect
ratio, confirming that this combination can be applied to
intricate geometries. Editing a virtual topology is allowed
in commercial software, so we respect the real geometry
composed of triangles provided by the CT. Simultane-
ously, we reduce the number of cells keeping a very good
quality in the mesh and save a lot of computing time.
– Using steady and transient simulations. In steady mode,
the case (a) Real topology and patch-conforming does
not converge due to presence of cells with poor Aspect
Ratio and Skewness, and the fastest is the case (d) Virtual
topology and patch-independent. In transient mode, the
cases (b), (c) and (d) have a good behaviour since they
reproduce correctly a forced spirometry.
The use of virtual topologies could be extended in biomed-
ical engineering because the geometries handling is similar
to this case (Agujetas et al. 2017).
Acknowledgements Thanks to Dr. Alejo, Servicio Radiología (Hospi-
tal IC, Badajoz), who provided the images of this study. This work was
financially supported by Junta de Extremadura under Project “Ayudas
para la realización de actividades de investigación y desarrollo tec-
nológico, de divulgación y de transferencia de conocimiento por los
Grupos de Investigación de Extremadura (GR150014)” and Sociedad
Asturiana de Patología Respiratoria under project Experimental and
123
A. Fernández-Tena et al.
numerical study of a three-dimensional model of an asthmatic patient
airways reconstructed from CT or MR images.
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of
interest.
References
3D Slicer V4.4.0 (2014). https://www.slicer.org
Agujetas R, Ferrera C, Marcos AC, Alejo JP, Montanero JM (2017)
Numerical and experimental analysis of the transitional flow across
a real stenosis. Biomech Model Mechanobiol 16(4):1447–1458.
https://doi.org/10.1007/s10237-017-0898-2
Ansys version 16.2 (2015) ANSYS Inc
Ashurst I, Malton A, Prime D, Sumby B (2000) Latest advances in the
development of dry powder inhalers. Pharm Sci Technol Today
3:246–256
Choi L (2007) Simulation of fluid dynamics and particle transport in
realistic human airways. Master’s thesis, School of Aerospace,
Mechanical and Manufacturing Engineering, RMIT University,
Melbourne, Australia
De Backer JW, Vos WG, Devolder A, Verhulst SL, Germonpré P, Wuyts
FL, Parizel PM, De Backer W (2008) Computational fluid dynam-
ics can detect changes in airway resistance in asthmatics after acute
bronchodilation. J Biomech 41:106–113
De Backer JW, Vos WG, Vinchurkar SC, Claes R, Drollmann A, Wul-
frank D, Parizel PM, Germonpré P, De Backer W (2010) Validation
of computational fluid dynamics in CT-based airway models with
SPECT/CT. Radiology 257:854–862
De Backer LA, Vos W, De Backer J, Van Holsbeke C, Vinchurkar S,
De Backer W (2012) The acute effect of budesonide/formoterol
in COPD: a multi-slice computed tomography and lung function
study. Eur Respir J 40:298–305
Dyedov V, Einstein DR, Jiao X, Kuprat AP, Carson JP, Pin F (2009)
Variational generation of prismatic boundary-layer meshes for
biomedical computing. Int J Numer Meth Eng 79:907–945
Fedorov A et al (2012) 3D slicer as an image computing platform for the
quantitative imaging network. Magn Reson Imag 30:1323–1341
Fernández-Tena A (2014) Clinical applications of fluid dynamics mod-
els in respiratory disease. Ph.D. thesis, University of Oviedo,
Spain. http://digibuo.uniovi.es/dspace/handle/10651/29057
Fernández-Tena A, Casan Clarà P (2015) Use of computational
fluid dynamics in respiratory medicine. Arch Bronconeumol
2015(51):293–298
Fernández-Tena A, Fernández J, Casan P (2016) Particle deposition in
healthy and bronchoconstricted lung. Eur Respir J 48(60):PA4402
Fernández-Tena A, Fernández J, Álvarez E, Casan P, Walters K (2017a)
Design of a numerical model of lung by means of a special bound-
ary condition in the truncated branches. Int J Numer Methods
Biomed Eng 33(6):e2830
Fernández-Tena A, Marcos AC, Martínez C, Walters DK (2017b) A
new adaptive time step method for unsteady flow simulations in a
human lung. Comput Methods Biomech Biomed Eng 20(8):915–
917
Foucault G, Cuilliére JC, Francoise V, Leon JC, Maranzana R (2007)
Adaptation of cad model topology for finite element analysis.
Comput Aided Des 40:176–196
Gibson GJ, Loddenkemper R, Sibille Y, Lundbäck B, Fletcher M (2013)
Lung health in europe: facts and figures. European Lung Founda-
tion, Sheffeld
Simulation of the human airways using virtual topology tools and meshing optimization
Hofmann W, Martonen TB, Graham RC (1989) Predicted deposition
of nonhygroscopic aerosols in the human lung as a function of
subject age. J Aerosol Med 1989(2):49–68
Hörschler I, Meinke M, Schröder W (2003) Numerical simulation of the
flow field in a model of the nasal cavity. Comput Fluids 32:39–45
Hounsfield GN (1973) Computerized transverse axial scanning (tomog-
raphy): part I. Description of system. Br J Radiol 46:1016–1022
Inthavong K, Choi L, Tu J, Ding S, Thien F (2010) Micron particle depo-
sition in a tracheobronchial airway model under different breathing
conditions. Med Eng Phys 32:1198–1212
Islam MS, Saha SC, Sauret E, Gu Y, Ristovski Z (2015) Numerical
investigation of aerosol particle transport and deposition in real-
istic lung airway. In: Liu GR, Das R (eds) The 6th international
conference on computational methods (ICCM2015), ICCM. Sci-
entech Publisher llc, USA, Auckland, New Zealand
Jahangiri M, Saghafian M, Sadeghi MR (2015) Numerical study of
turbulent pulsatile blood flow through stenosed artery using fluid–
solid interaction. Comput Math Methods Med 2015:515613
Kannan R, Guo P, Przekwas A (2016) Particle transport in the human
respiratory tract: formulation of a nodal inverse distance weighted
Eulerian–Lagrangian transport and implementation of the Wind–
Kessel algorithm for an oral delivery. Int J Numer Meth Biomed
Eng 32(6). https://doi.org/10.1002/cnm.2746
Kannan R, Singh N, Przekwas A, Delvadia R, Tian G, Walenga R
(2017a) Pharmaceutical aerosols deposition patterns from a dry
powder inhaler: Euler Lagrangian prediction and validation. Med
Eng Phys 42:35–47
Kannan R, Chen ZJ, Singh N, Przekwas A, Delvadia R, Tian G, Walenga
R (2017b) A quasi-3D wire approach to model pulmonary airflow
in human airways. Int J Numer Methods Biomed Eng 33(7):e2838
Kitaoka H, Takaki R, Suki B (1999) A three-dimensional model of the
human airway tree. J Appl Physiol 87(6):2207–2217
Kleinstreuer C (2006) Biofluid dynamics: principles and selected appli-
cations. CRC Press, Boca Raton
Kleinstreuer C, Zhang Z, Donohue JF (2008) Targeted drug-aerosol
delivery in the human respiratory system. Annu Rev Biomed Eng
10:195–220
Kolanjiyil AV, Kleinstreuer C, Sadikot RT (2017) Computationally effi-
cient analysis of particle transport and deposition in a human
whole-lung-airway model. Part II: dry powder inhaler application.
Comput Biol Med 84:247–253
Lin C, Tawhai MH, Merryn H, McLennan G, Hoffman EA (2007)
Characteristics of the turbulent laryngeal jet and its effect on air-
flow in the human intra-thoracic airways. Respir Physiol Neurobiol
157:295–309
Lip K, Philip C (2015) Pharmaceutical aerosol electrostatics: a field
with much potential for development. Ther Deliv 6:105–107
Marchandise E, Geuzaine C, Remacle JF (2013) Cardiovascular and
lung mesh generation based on centerlines. Int J Numer Method
Biomed Eng 2013(29):665–682
Menter F, Langtry R, Völker S (2006) Transition modelling for general
purpose CFD codes. Flow Turbul Combust 77:277–303
Miller MR, Hankinson JATS, Brusasco V, Burgos F, Casaburi R, Coates
A, Crapo R, Enright P, Van der Grinten CP, Gustafsson P et al
(2005) Standardisation of spirometry. Eur respir J 26(2):319–338
Miyawaki S, Tawhai MH, Hoffman EA, Wenzel SE, Lin CL (2017)
Automatic construction of subject-specific human airway geom-
etry including trifurcations based on a CT-segmented airway
skeleton and surface. Biomech Model Mechanobiol 16:583–596
Nahar K, Gupta N, Gauvin R, Absar S, Patel B, Gupta V, Khademhos-
seini A, Ahsan F (2013) In vitro, in vivo and ex vivo models for
studying particle deposition and drug absorption of inhaled phar-
maceuticals. Eur J Phar Sci 49:805–818
OECD/UE (2016) Health at a glance: Europe 2016. state of health in
the EU cycle. Technical report OECD/UE
477
Przywara B (2010) Projecting future health care expenditure at Euro-
pean level: drivers, methodology and main result. European
Commission, Economic and Financial Affairs Publications, Brus-
sels
Rochefort L, Vial L, Fodil R, Mâıtre X, Louis B, Isabey D, Caillibotte G,
Thiriet M, Bittoun J, Durand E, Sbirlea-Apiou G (2007) In vitro
validation of computational fluid dynamic simulation in human
proximal airways with hyperpolarized 3 He magnetic resonance
phase-contrast velocimetry. J Appl Physiol 102:2012–2023
Sbirlea-Apiou G, Katz I, Caillibotte G, Martonen T, Yang Y (2007)
Deposition mechanics of pharmaceutical particles in human air-
ways. In: Hickey AJ (ed) Inhalation aerosols. Physical and
biological basis for therapy, 2nd edn. Informa Healthcare USA,
New York, pp 1–30
Scotti A, Piomelli U (2002) Turbulence models in pulsating flows.
AIAA 40:537–544
Stahlhofen W, Rudolf G, James AC (1989) Intercomparison of
experimental regional aerosol deposition data. J Aerosol Med
1989(2):285–308
Streinbenner JP, Wyman NJ, Chawner JR (2000) Fast surface meshing
on imperfect cad models. In: 9th international meshing roundtable,
Sandia National Laboratories, New Orleans, USA, pp 33–41
Taherian S, Rahai HR, Waddington T (2011) CFD modelling and
analysis of pulmonary airways/particles transport and deposition.
In: 41st AIAA fluid dynamics conference and exhibit, American
Institute of Aeronautics and Astronautics. American Institute of
Aeronautics and Astronautics, Honolulu, USA, pp 2011–3270
Tan FPP, Wood NB, Tabor G, Xu XY (2011) Comparison of LES
of steady transitional flow in an idealized stenosed axisymmet-
ric artery model with a RANS transitional model. J Biomech Eng
133:051001
Varghese SS, Frankel SH, Fischer PF (2007a) Direct numerical simu-
lation of stenotic flows. Part 1. Steady flow. J Fluid Mech 582:253
Varghese SS, Frankel SH, Fischer PF (2007b) Direct numerical simula-
tion of stenotic flows. Part 2. Pulsatile flow. J Fluid Mech 582:281
Varghese SS, Frankel SH, Fischer PF (2008) Modelling transition to
turbulence in eccentric stenotic flows. J Biomech Eng 130:014503
Versteeg HK, Malalasekera W (2007) An Introduction to Computational
Fluid Dynamics. Pearson Education Limited, England
Vos W, De Backer J, Poli G, De Volder A, Ghys L, Van Holsbeke C,
Vinchurkar S, De Backer L, De Backer W (2013) Novel func-
tional imaging of changes in small airways of patients treated with
extrafine beclomethasone/formoterol. Respiration 86:393–401
Weibel ER (1963) Morphometry of the human lung. Academic, New
York
Weibel ER (2009) What makes a good lung? The morphometric basis
of lung function. Swiss Med Wkly 139:375–386
Wilcox DC (2006) Turbulence modelling for CFD. DCW Industries,
Flintridge
Zhang Y, Finlay WH (2005) Measurement of the effect of cartilaginous
rings on particle deposition in a proximal lung bifurcation model.
Aerosol Sci Technol 39:394–399
Zhang Z, Kleinstreuer C, Hyun S (2012) Size-change and deposition
of conventional and composite cigarette smoke particles during
inhalation in a subject-specific airway model. J Aerosol Sci 45:34–
52
Zheng J (2014) Numerical simulation of nanoparticle transportation and
deposition in pulmonary vasculature. Master’s thesis, Department
of Mechanical Engineering and Mechanics, Lehigh University,
Lehigh, USA
123