Computers in Biology and Medicine 102 (2018) 40–50

Contents lists available at ScienceDirect

Computers in Biology and Medicine

journal homepage: www.elsevier.com/locate/compbiomed

Examining mesh independence for flow dynamics in the human nasal cavity T
∗
Kiao Inthavong , Annicka Chetty, Yidan Shang, Jiyuan Tu
RMIT University, School of Engineering, Australia

A R T I C LE I N FO A B S T R A C T

Keywords: Increased computational resources provide new opportunities to explore sophisticated respiratory modelling. A
Mesh independence survey of recent publications showed a steady increase in the number of mesh elements used in computational
Nasal cavity models over time. Complex geometries such as the nasal cavity exhibit sharp gradients and irregular curvatures,
y+ leading to abnormal flow development across their surfaces. As such, a robust method for examining the near-
y-plus
wall mesh resolution is required. The non-dimensional wall unit y+ (often used in turbulent flows) was used as a
Laminar flow
parameter to evaluate the near-wall mesh in laminar flows.
Mesh independence analysis from line profiles showed that the line location had a significant influence on the
result. Furthermore, using a single line profile as a measure for mesh convergence was unsuitable for re-
presenting the entire flow field. To improve this, a two-dimensional (2D) cross-sectional plane subtraction
method where scalar values (such as the velocity magnitude) on a cross-sectional plane were interpolated onto a
regularly spaced grid was proposed. The new interpolated grid values from any two meshed models could then
be compared for changes caused by the different meshed models. The application of this method to three-
dimensional (3D) volume subtraction was also demonstrated.
The results showed that if the near-wall mesh was sufficiently refined, then narrow passages were less reliant
on the overall mesh size. However, in wider passages, velocity magnitudes were sensitive to mesh size, requiring
a more refined mesh.

1. Introduction independence analysis used in computational modelling of human nasal

Objective tests of disordered nasal airflow, such as anterior rhino-
manometry, acoustic rhinometry and nasal peak inspiratory flow, suffer
from limited sensitivity, specificity and reliability, providing restricted
insight into the specific cause or site of obstruction. Assessments of
nasal morphology for abnormalities and diseases can be conducted
through 3D computational models [10,22,31,33,38], providing medical
practitioners with the insights necessary to make informed decisions
regarding surgical interventions.
To understand the precise causes of nasal obstruction and the effects
of corrective surgery, more qualitative and quantitative information is
required, including an understanding of the airflow pattern (laminar or
turbulent), localised velocity and pressure (at different flow rates and in
different parts of the nasal cavity) and wall shear stress. Computational
fluid dynamics (CFD) applied to the modelling of fluid flow in the nose
has been used to investigate airflow patterns [17] for common defor-
mities such as nasal bone fractures [5], septal deviation [2] and inferior
turbinate hypertrophy [4]. Increased computational power has fa-
cilitated enhanced studies involving more airway models [6,29,36,37].
With resources readily available, it is timely to explore mesh
cavity airflows. The accuracy requirements for computational model-
ling in clinical and engineering applications [9,18,28] are far less
stringent than in research applications. Thus arises the question of the
extent to which a mesh should refined be with respect to the level of
flow detail.
Early reported computational studies of airflow through a human
nasal cavity include a study by Ref. [7]; which used an idealised nose-
like shape (9888 mesh elements used) and a study by Ref. [17]; which
reconstructed the nasal cavity from computed tomography (CT) scans
of one chamber, abruptly ending at the choanae (76,950 mesh elements
used). A literature review was performed of computational studies of
the nasal cavity from the years 1993–2017 and the number of mesh
elements reported in the models was plotted (Fig. 1). Since 1993, the
number of studies published in journals has steadily increased, as has
computational model fidelity, multiphysics and complexity.
For clinical and engineering applications, a computational model
may not require high numbers of mesh elements when only general
flow features are of interest. These features include: the change of di-
rection of flow streamlines from vertical (as air enters the nostril inlet)
to horizontal (in the main nasal passage); flow acceleration through the

∗
Corresponding author.RMIT University, School of Engineering, PO Box 71, Bundoora, 3083, Australia.
E-mail address: kiao.inthavong@rmit.edu.au (K. Inthavong).

https://doi.org/10.1016/j.compbiomed.2018.09.010
Received 6 June 2018; Received in revised form 12 September 2018; Accepted 12 September 2018
0010-4825/ © 2018 Elsevier Ltd. All rights reserved.
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 1. Mesh cell elements reported in the literature between 1993 and 2017. Examples of some high meshed nasal models are highlighted [11,27]. The studies with
very large number of cells ([1,20] are linked to high fidelity turbulent flow using Large Eddy Simulations (LES) or Direct Numerical Simulations (DNS).

nasal valve; recirculating flow in the olfactory region; bulk flow velocity distributions and 3D volume distributions so that higher-order
through the middle nasal cavity and nasal cavity floor; and a 90° turn as grid independence analysis could be performed.
air enters the nasopharynx [34]. Many computational models apply
simplifications to physiological behaviour (e.g., rigid walls, dry, smooth 2. Method
walls and steady flows); therefore, higher-fidelity models may not yield
substantial increases in accuracy with additional simulation time. 2.1. Computational model: pipe geometry
Fig. 1 shows a large disparity in the reported mesh elements be-
tween 10,000 cells and 44 million cells, despite many publications re- Four computational pipe models with mesh variations were created
porting mesh-independent computational models. This is due to a for evaluating a laminar y+ parameter. Pipe flows provide solutions that
combination of the available computational power and the grid in- can be applied to complex internal flows that behave like passages, such
dependence test method used. To quantify errors of mesh in- as human airways. The pipe had a radius of 7.5 mm (R = 7.5 mm) and a
dependence/grid convergence, an estimation can be performed using length of 16R (120 mm). A structured mesh with varying densities was
Richardson extrapolation [23–25] and the grid convergence index used (see Table 1). The cross-sectional mesh is shown in Fig. 2a. A la-
(GCI), which was introduced by Ref. [26] as a method for uniform re- minar streamwise-periodic flow was applied to the pipe to produce a
porting of CFD results but has since evolved into an uncertainty esti- fully developed flow (Poiseuille flow) with different Reynolds numbers
mator. of 200, 500, 1000, 1300, 1600 and 2000.
Mesh independence analysis in CFD studies of airflow in the nasal
cavity has mainly applied convergence of a local parameter (such as
2.2. Laminar y+ for internal pipe flow
velocity) along a selected line profile [3,15] or convergence of global
parameters such as pressure drop or wall shear stress [19,21,32]. Early
The fully developed laminar velocity profile in a pipe is:
work by Ref. [14] relied on local velocity profiles for mesh in-
dependence, resulting in a mesh count of 586,000 tetrahedral elements. R − y 2⎞
u (y ) = 2U ⎛1 − ⎛
⎜ ⎞ ⎟
Subsequent articles recorded a count of 950,000 cells [13,15,16]. More ⎝ ⎝ R ⎠⎠ 1
recent work has shown approximately 4–5 million elements are re-
quired for grid independence [8,27,30,35]. where U is the mean velocity, R is the pipe radius and y is the distance
Accurately predicting airflow patterns in the nose to better under- from the wall. The non-dimensional wall distance and dimensionless
stand nasal airway disorders relies on the use of a high-quality mesh. velocity are defined as:
The near-wall mesh refinement in turbulent flow regimes
(typically > 20 L/min steady inhalation rate) is characterised by the Table 1
y+ parameter, which describes the first wall-adjacent cell height re- Pipe geometry mesh details.
lative to the different sub-layers of the turbulent boundary layer. For Dimension pipe M1 pipe M2 pipe M3 pipe M4
laminar flow regimes (typically < 15 L/min steady inhalation rate), no
A (no. faces) 4 6 8 12
parameter is used as the laminar boundary layer consists of a single
B (face size) 10e-4 6.5e-04 3.9e-04 2.6e-04
region and thus, studies do not report quantified measures of the near- C (Inflationa) None 3.25e-04 1.95e-04 1.30e-04
wall mesh. D (streamwise) 2.4e-03 1.5e-03 9.0e-04 6.0e-04
This paper evaluated the use of the y+ parameter to quantitatively Total cells: 10,560 43,824 130,560 407,232
describe the near-wall boundary layer of a laminar internal pipe flow
(Poiseuille flow). Additionally, the use of velocity line profiles as a Dimensions A to C are labelled in Fig. 2.
a
First cell height reported, inflation contained five cells in total with a
measure of grid independence was evaluated and extended to 2D planar
growth rate of 1.1.

41
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 2. Nasal cavity geometry model with dimensions for overall length, height, and width. Six planes were created (two in each of the anterior, main passage, and
posterior regions) and labelled as planes-a, -b, -c, -d, -e, -f. The unstructured surface mesh for model nose M3 (2-million cells) is shown and an inferior view of the
nostril inlet is given to show the mesh quality with prism layers.

ρu *y u 2.3. Computational model: nasal cavity geometry

y+ = ; u+ =
μ u* 2
where ρ is the fluid density, μ is dynamic viscosity and u∗ is the friction A human nasal cavity geometry was meshed using Ansys-Fluent
velocity (taken at the first cell), defined as: v18.2, which provided high-quality tetrahedral cells with prism layers.
For the mesh independence evaluation, the model was meshed with five
τw du levels of refinement, which were labelled Nose M1 (coarsest mesh, 0.5
u* = with τw = μ
ρ dy 3 million cells) through to Nose M5 (finest mesh, 10 million cells). The
level of skewness was used as the main criteria for mesh quality, where
Substituting Eqn. (1) and its derivative (i.e., velocity gradient) into
the most refined model, Nose M5, had a maximum skewness of 0.64. A
y+ and u+ produces:
steady laminar flow of 15 L/min (mass flow rate = 3.0625e–4 kg/s) was
R−y 2

y+ =
2ρy Uμ
(R − y ) u+ =
( ( ))
U 1− R
defined at the outset.
The pressure-based coupled solver was used where the system of
μ R2ρ Uμ
(R − y ) equations comprising the continuity equation:
R2ρ 4
∂
When y′ = y / R is set and the terms that comprise the Reynolds (ui ) = 0
number are collected, the non-dimensional wall units simplify to: ∂x i 6

y+ = y' 2Re (1 − y') 5 and the momentum equation:

This is the solution profile for a fully developed laminar Poiseuille ∂ui ∂p ∂ ⎛ ∂ui ⎞
ρui =− + ⎜μ
flow. The term y′ has range of y′ = 0 (wall boundary) to y′ = 1 (pipe
⎟
∂x j ∂x i ∂x j ⎝ ∂x j ⎠ 7
centre/radius).

42
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 3. (a) Normalised velocity u+ profile as a function of wall units y+ for a laminar pipe flow in the near wall region of different meshed pipe models (b) CFD results
for determining the y+ profile for different flow rates. The markers for the largest first cell height ( y′ = 0.055) are for pipe M1 and moving closest to the wall for
models pipe M2 ( y′ = 0.022), pipe M3 ( y′ = 0.013), pipe M4 ( y′ = 0.008) are shown.

Fig. 4. Velocity contours in three planes for models, nose M1, nose M3, and nose M5 which shows the variation between models of different mesh density. Plane-b,
and plane-c are found where two nasal chambers exist, while plane-d is in the nasopharynx region where the chambers have merged.

Were solved together. Spatial discretisation used the second-order used double precision to avoid errors that may arise from high aspect
accurate upwind scheme where quantities at cell faces were computed ratio cells from the prism layers.
using a multidimensional linear reconstruction. The Green-Gauss node-
based gradient of a variable was used to discretise the convection and
diffusion terms in the flow conservation equations. The simulations

43
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 5. Multiple velocity line profiles on cross-section (a)(b) Plane-b; (c)(d) Plane-c; (e)(f) Plane-d for direct comparison between models (a)(c)(e) Nose M1 (dots), and
Nose M3 (lines); and (b)(d)(f) Nose M3 (dots), and Nose M5 (lines).

3. Results y+, where the viscous dominant sub-layer (i.e., laminar layer) occurs
when y+ < 5. In this sub-layer, u+/ y+ ∼ 1. Computational results from
3.1. y+ for laminar pipe flow the pipe simulations of normalised velocity and wall units for all
Reynolds numbers were plotted in Fig. 3a to verify that u+/ y+ ≅ 1 for all
In turbulent flows, the near-wall boundary layer is characterised by cases, suggesting it captures the laminar boundary layer profile

44
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 6. Comparison of velocity profiles at plane-d for (a) Nose M3 (dots) to Nose M4 (lines) and (b) Nose M4 (dots) to Nose M5 (lines) where grid convergence occurs at
Nose M4.

Fig. 7. Changing the near wall mesh refinement on the mesh independent model, Nose M4 (4.8million cells) where (a) inferior view of model showing the nostrils;
and the (b) original Nose M4 (unchanged) model with refined near wall mesh. Subsequent relaxation of the near wall mesh refinement was created where (c)
y+ + + +
ave = 0.27; (d) yave = 0.53; (e) yave = 0.98; (f) yave = 1.

correctly. Fig. 3b shows that y+ increases with increased flow rate (e.g.,
Re = 200 to Re = 2000, these being the plotted lines obtained from
Eqn. (5)). The markers plotted were individual results obtained from
each pipe model and for the entire range of Reynolds numbers. These
results verify that the y+ value can be used as a quantitative parameter
for examining near-wall meshing of laminar flows where the profile
characteristics are consistent with laminar boundary layers, such as the
behaviour found in the viscous dominant sub-layers of turbulent flows.
3.2. Nasal cavity flow and mesh independence
Velocity contours in three planes (Plane-b, Plane-c and Plane-d)
showed that recirculating flow patterns were captured in the fine
Meshed model, Nose M5 (Fig. 4). In the coarse meshed model, Nose M1,
the flow was more diffusive. Although only three planes are given for
brevity, it shows that where the left and right chambers were separate,
the velocity contours between each nasal cavity model were similar.
However, where the two chambers merged into a single passage (Plane-
d), there was greater variation between the contours of the coarse- and
fine meshed models.
In reported mesh independence tests or CFD model comparisons, a
single line profile of a flow variable is used to demonstrate convergence
or model performance. Multiple line profiles were extracted and com-
pared between Nose M1 (0.5 million cells) and Nose M3 (2.3 million
cells) (Fig. 5a, c and 5e), and between Nose M3 (2.3 million cells) and
Nose M5 (10.0 million cells) (Fig. 5b, d and 5f). At Plane-b, the profile
along Line-3 (and, to a lesser extent, Line-2 and Line-5) showed rea-
sonable mesh convergence between Nose M1 and Nose M3. However,

45
K. Inthavong et al.
Fig. 8. Gaussian density estimation function for (a) y+ values and (b) wall shear stress on the boundary wall surface of nasal cavity geometry M4 (approximately
4.8million cells), affected by near wall mesh refinements.
there is significant flow variation along Line-4. For Plane-c, the largest
flow variation occurred at Line-2 and Line-4. For Plane-d, the flow
differed significantly for all lines. This is due to the larger open area,
where the mesh variation was substantial compared to the anterior
planes that exhibited narrowed regions. The results show that the use of
line profiles—in particular, a single line profile—is a limited method for
justifying mesh convergence over the entire computational domain.
It was apparent that Plane-d required the most mesh refinement due
to the mixing of the two flow streams at the nasopharynx, caused by the
two chambers merging. Plane-b and Plane-c were not shown because
the velocity profiles had already reached mesh independence for the
model Nose M3. Line velocity profiles at Plane-d for Nose M3 to Nose
M4 (Fig. 6a) and Nose M4 to Nose M5 (Fig. 6b) show that the velocity
profiles converge between Nose M4 and Nose M5.
3.3. Effect of near-wall mesh in laminar flow
The results showed that Nose M4 had converged velocity profiles
across three planes over the nasal cavity. The effect of the near-wall
mesh was evaluated by changing the prism layer distribution (Fig. 7).
The redistribution of mesh elements changed the average y+ values
where the unmodified model (Fig. 7b) displayed y+ave = 0.13. Four
additional models were created with increasing y+ave values. The re-
distribution across the entire surface as a distribution function is shown
in Fig. 8a. Despite the larger y+ variation, the resulting wall shear stress
(Fig. 8b) remained consistent across all models.
Line velocity profile comparisons taken at the same locations (the
planes from Fig. 5) were referenced against the unmodified model that
displayed y+ave = 0.13. When y+ ave increased to 0.27, very little change
occurred in all profiles. For y+ave = 0.53, significant differences in the
profiles were observed in Plane-d but not in Plane-b or Plane-c. As y+ave
increased further, variations between velocity profiles began to appear
in Plane-b and Plane-c. However, the variations were not significant
relative to the near-wall mesh coarsening. This suggests that the near-
wall mesh was sufficiently resolved if y+ ave = 0.27. Further, it sig-
nificantly influenced velocity profiles in large channels of the nasal
cavity but was less significant for narrow channels of the nasal cavity.
3.4. Multidimensional grid independence
The line profile comparison is one-dimensional; a comparison on
cross-sectional planes is a 2D representation. The velocity (or any
scalar) of a plane was extracted and interpolated [12] onto an
46
Computers in Biology and Medicine 102 (2018) 40–50
800 × 800 uniformly spaced grid for any two meshed models. The in-
terpolation used the ‘nearest’ method in Python's SciPy library. The
variation between two meshed models was defined as the difference
between the interpolated grid values, given as:
N
1 abs(qi − pi )
σ=
N
∑ (qi + pi )/2
× 100
i=1 8
where p and q are scalar values (e.g., velocity) from cell i, taken from
the coarser- and finer-meshed models respectively. N is the total
number of cells in the grid. To constrain the large σ (residual) due to a
small qi , the denominator used the average value between the two
models: (qi + pi )/2 .
The six planes from Fig. 2b were evaluated where pi and qi were the
velocity magnitude values on the planes in any two models. The σ
values for each plane pair between Nose M2 and Nose M3 are shown as
contours in Fig. 9a. Plane-e exhibited the largest local discrepancy due
to the intense mixing caused by the merging of the left and right
chambers. Plane-a and Plane-b also displayed some local discrepancies;
however, these were less severe. The averaged residual for each plane
was plotted in Fig. 9b, which shows that σ increased from Nose M1 to
Nose M2 but then decreased monotonically. The averaged σ reduced to
less than 5% between Nose M4 and Nose M5. This was sufficiently small
to satisfy mesh independence.
If the interpolation from 2D planes to the airway volume is ex-
tended, then the mesh independence subtraction method becomes 3D.
Data from each nasal cavity model was interpolated onto a 100 × 100 x
100 uniformly spaced grid. The σ for each pair of models is shown in
Fig. 10, where the σ initially increased between Nose M1 and Nose M2
but decreased thereafter. This profile was the same as that of the 2D
plane subtraction method where the σ values were nearly identical,
suggesting that using six planes is sufficient to obtain an overall σ re-
presentation of the entire volume.
4. Discussion
Mesh independence based on convergence of flow parameters on a
single line is a limited approach as it ignores the rest of the domain,
which can lead to regions of poorly converged mesh. However, mesh
independence based on a single averaged value of global flow para-
meters misses locally poor mesh convergence, which is averaged out.
This study proposes that one of the following should be used to ensure
sufficient mesh independence:
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 9. Comparisons of line velocity profiles on cross-sectional planes (Fig. 5) showing the effect of different numbers of near wall mesh elements. The horizontal axis,
is the x-coordinate values (m), while the vertical axis is velocity (m/s). The profiles are compared against the reference velocity profiles of the unmodified model that
has y+ave = 0.13 (lines).

multiple line profiles across multiple cross-section planes
a comparison of a flow parameter at each mesh element region over
the entire domain, using multiple planar subtractions or volume
subtraction.
The CFD results also showed that the near-wall mesh for laminar
flows can adopt the y+ parameter to ensure a sufficient mesh is applied
at the wall.
The velocity line profiles and contour planes for narrow passages
did not vary significantly between meshed models. However, cross-

47
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 10. (a) Residual values from subtraction of planes between two models Nose M2 and Nose M3.

sectioned regions that exhibited a larger passage (e.g., nasopharynx) space to fill and the mesh cells varied in size between models (see
showed greater variations in flow behaviour between the different Table 2). Therefore, mesh refinement in larger passage regions is es-
meshed models. While this variation was partly due to the increased sential, especially where rapid changes in flow behaviour (such as
mixing of two streams of fluids, the nasopharynx (Plane-e) had more mixing) exist and maintain overall importance in the analysis.

48
K. Inthavong et al. Computers in Biology and Medicine 102 (2018) 40–50

Fig. 11. 3D volume velocity magnitude residual change between two subsquent mesh models.

Table 2 comparison of different computational model performances.

Nasal cavity geometry mesh details.
Nose M1 Nose M2 Nose M3 Nose M4 Nose M5
Conflicts of interest

min-scoped-size 0.14 0.09 0.06 0.045 0.04 None Declared.

Max skewness 0.7427 0.754 0.722 0.684 0.642
wall 1st-cell-height 0.04 0.04 0.03 0.025 0.02
Total cells: 511,312 1,092,521 2,324,473 4,892,911 10,019,305
Acknowledgement

The authors acknowledge the financial support for the research,

A variety of tests were presented that showed that a mesh with
approximately 4–5 million cells provided good overall mesh in-
dependence. Recent work has shown that this number of elements was
needed for grid independence [8,27,30,35]. Coarse Meshed models can
provide sufficient gross flow features if the near-wall mesh maintains
y+ < 0.27. These models may be acceptable for rapid-solution turn-
around times and should be considered before committing to a higher-
resolution analysis. This is particularly the case for the anterior nasal
cavity and situations where rapid results take precedence over accu-
racy. These findings have strong relevance for large scale model ana-
lysis and provide the option of rapid turnaround of results.
5. Conclusion
This paper presented the use of the y+ parameter for evaluating the
near-wall mesh resolution in laminar flows. It also demonstrated that if
the near-wall mesh was refined effectively, then the anterior half of the
nasal cavity became less reliant on the overall mesh size as the passa-
geways were very narrow and the flow field was heavily influenced by
the bounded walls. Flow variations were most sensitive in the naso-
pharynx, where the two chambers merged into a single passageway.
This area required more refined meshing in the inner flow region (bulk
flow region).
Mesh independence using single line profiles of a flow parameter to
justify convergence is a limited approach as it ignores the flow in the
rest of the domain, which can lead to regions of poorly converged mesh.
The extension of the line profiles to 2D planes and 3D volumes was used
to evaluate the velocity magnitude in different meshed models of the
nasal cavity. This new method proved capable of determining local
regions of discrepancy when each mesh was refined.
The methods presented in this study are expected to provide a tool
for quick, easy and reliable evaluation of mesh independence or
authorship, and/or publication of this article from the Australian
Research Council (grant no. DP160101953).
References
[1] H. Calmet, A.M. Gambaruto, A.J. Bates, M. Vazquez, G. Houzeaux, D.J. Doorly,
Large-scale CFD simulations of the transitional and turbulent regime for the large
human airways during rapid inhalation, Comput. Biol. Med. 69 (2016) 166–180.
[2] X.B. Chen, H.P. Lee, V.F. Chong, Y. Wang de, Assessment of septal deviation effects
on nasal air flow: a computational fluid dynamics model, Laryngoscope 119 (2009)
1730–1736.
[3] X.B. Chen, H.P. Lee, V.F. Chong, Y. Wang de, A computational fluid dynamics model
for drug delivery in a nasal cavity with inferior turbinate hypertrophy, J. Aerosol
Med. Pulm. Drug Deliv. 23 (2010) 329–338.
[4] X.B. Chen, H.P. Lee, V.F. Chong, Y. Wang de, Impact of inferior turbinate hyper-
trophy on the aerodynamic pattern and physiological functions of the turbulent
airflow - a CFD simulation model, Rhinology 48 (2010) 163–168.
[5] X.B. Chen, H.P. Lee, V.F. Chong, Y. Wang de, Assessments of nasal bone fracture
effects on nasal airflow: a computational fluid dynamics study, American journal of
rhinology & allergy 25 (2011) e39–43.
[6] J.W. De Backer, W.G. Vos, A. Devolder, S.L. Verhulst, P. Germonpré, F.L. Wuyts,
P.M. Parizel, W. De Backer, Computational fluid dynamics can detect changes in
airway resistance in asthmatics after acute bronchodilation, J. Biomech. 41 (2008)
106–113.
[7] D. Elad, R. Liebenthal, B.L. Wenig, S. Einav, Analysis of air flow patterns in the
human nose, Med. Biol. Eng. Comput. 31 (1993) 585–592.
[8] D.O. Frank-Ito, M. Wofford, J.D. Schroeter, J.S. Kimbell, Influence of mesh density
on airflow and particle deposition in sinonasal airway modeling, J. Aerosol Med.
Pulm. Drug Deliv. 29 (1) (2015) 46–56.
[9] C.B. Frederick, P.R. Gentry, M.L. Bush, L.G. Lomax, K.A. Black, L. Finch,
J.S. Kimbell, K.T. Morgan, R.P. Subramaniam, J.B. Morris, J.S. Ultman, A hybrid
computational fluid dynamics and physiologically based pharmacokinetic model for
comparison of predicted tissue concentrations of acrylic acid and other vapors in
the rat and human nasal cavities following inhalation exposure, Inhal. Toxicol. 13
(2001) 359–376.
[10] G.J.M. Garcia, J.D. Schroeter, J.S. Kimbell, Olfactory deposition of inhaled nano-
particles in humans, Inhal. Toxicol. 27 (2015) 394–403.
[11] K. Inthavong, Q.J. Ge, X.D. Li, J.Y. Tu, Detailed predictions of particle aspiration
affected by respiratory inhalation and airflow, Atmos. Environ. 62 (2012) 107–117.
[12] K. Inthavong, Y. Shang, J. Tu, Surface mapping for visualization of wall stresses
during inhalation in a human nasal cavity, Respir. Physiol. Neurobiol. 190 (2014)

49
K. Inthavong et al.
54–61.
[13] K. Inthavong, Z. Tian, J. Tu, CFD simulations on the heating capability in a human
nasal cavity, 16th Australasian Fluid Mechanics Conference (AFMC), School of
Engineering, The University of Queensland, 2007, pp. 842–847.
[14] K. Inthavong, Z.F. Tian, H.F. Li, J.Y. Tu, W. Yang, C.L. Xue, C.G. Li, A numerical
study of spray particle deposition in a human nasal cavity, Aerosol Sci. Technol. 40
(2006) 1034–1045.
[15] K. Inthavong, Z.F. Tian, J.Y. Tu, W. Yang, C. Xue, Optimising nasal spray para-
meters for efficient drug delivery using computational fluid dynamics, Comput.
Biol. Med. 38 (2008) 713–726.
[16] K. Inthavong, J. Wen, J. Tu, Z. Tian, From CT scans to CFD modelling–fluid and heat
transfer in a realistic human nasal cavity, Engineering Applications of
Computational Fluid Mechanics 3 (2009) 321–335.
[17] K. Keyhani, P.W. Scherer, M.M. Mozell, Numerical simulation of airflow in the
human nasal cavity, J. Biomech. Eng. 117 (1995) 429–441.
[18] J.S. Kimbell, R.P. Subramaniam, Use of computational fluid dynamics models for
dosimetry of inhaled gases in the nasal passages, Inhal. Toxicol. 13 (2001) 325–334.
[19] C.M. King Se, K. Inthavong, J. Tu, Inhalability of micron particles through the nose
and mouth, Inhal. Toxicol. 22 (2010) 287–300.
[20] C. Li, J. Jiang, H. Dong, K. Zhao, Computational modeling and validation of human
nasal airflow under various breathing conditions, J. Biomech. 64 (2017) 59–68.
[21] X. Li, K. Inthavong, J. Tu, Particle inhalation and deposition in a human nasal cavity
from the external surrounding environment, Build. Environ. 47 (2012) 32–39.
[22] Y. Na, K.S. Chung, S.K. Chung, S.K. Kim, Effects of single-sided inferior turbi-
nectomy on nasal function and airflow characteristics, Respir. Physiol. Neurobiol.
180 (2012) 289–297.
[23] T.S. Phillips, C.J. Roy, Richardson extrapolation-based discretization uncertainty
estimation for computational fluid dynamics, J. Fluid Eng. 136 (2014) 121401-
121401-121410.
[24] L.F. Richardson, On the approximate arithmetical solution by finite differences of
physical problems involving differential equations, with an application to the
stresses in a masonry dam, Proceedings of the Royal Society of London. Series A 83
(1910) 335–336.
[25] L.F. Richardson, The deferred approach to the limit, Philos. Trans. R. Soc. London,
Ser. A 226 (1927) 299–361.
[26] P.J. Roache, Perspective: a method for uniform reporting of grid refinement studies,
J. Fluid Eng. 116 (1994) 405–413.
50
Computers in Biology and Medicine 102 (2018) 40–50
[27] J.D. Schroeter, G.J.M. Garcia, J.S. Kimbell, Effects of surface smoothness on inertial
particle deposition in human nasal models, J. Aerosol Sci. 42 (2011) 52–63.
[28] Y. Shang, K. Inthavong, J. Tu, Detailed micro-particle deposition patterns in the
human nasal cavity influenced by the breathing zone, Comput. Fluids 114 (2015)
141–150.
[29] C.D. Sullivan, G.J.M. Garcia, D.O. Frank-Ito, J.S. Kimbell, J.S. Rhee, Perception of
Better Nasal Patency Correlates with Increased Mucosal Cooling after Surgery for
Nasal Obstruction 150 Otolaryngology - Head and Neck Surgery, United States,
2014, pp. 139–147.
[30] X. Tong, J. Dong, Y. Shang, K. Inthavong, J. Tu, Effects of nasal drug delivery device
and its orientation on sprayed particle deposition in a realistic human nasal cavity,
Comput. Biol. Med. 77 (2016) 40–48.
[31] S. Vinchurkar, L. De Backer, W. Vos, C. Van Holsbeke, J. De Backer, W. De Backer, A
case series on lung deposition analysis of inhaled medication using functional
imaging based computational fluid dynamics in asthmatic patients: effect of upper
airway morphology and comparison with in vivo data, Inhal. Toxicol. 24 (2012)
81–88.
[32] S.M. Wang, K. Inthavong, J. Wen, J.Y. Tu, C.L. Xue, Comparison of micron- and
nanoparticle deposition patterns in a realistic human nasal cavity, Respir. Physiol.
Neurobiol. 166 (2009) 142–151.
[33] Y. Wang, S. Elghobashi, On locating the obstruction in the upper airway via nu-
merical simulation, Respir. Physiol. Neurobiol. 193 (2014) 1–10.
[34] J. Wen, K. Inthavong, J.Y. Tu, S. Wang, Numerical simulations for detailed airflow
dynamics in a human nasal cavity, Respir. Physiol. Neurobiol. 161 (2008) 125–135.
[35] J. Xi, P.W. Longest, Numerical predictions of submicrometer aerosol deposition in
the nasal cavity using a novel drift flux approach, Int. J. Heat Mass Tran. 51 (2008)
5562–5577.
[36] G. Xiong, J. Zhan, K. Zuo, J. Li, L. Rong, G. Xu, Numerical flow simulation in the
post-endoscopic sinus surgery nasal cavity, Med. Biol. Eng. Comput. 46 (2008)
1161–1167.
[37] J. Zhang, Y. Liu, X. Sun, S. Yu, C. Yu, Computational fluid dynamics simulations of
respiratory airflow in human nasal cavity and its characteristic dimension study,
Acta Mechanica Sinica/Lixue Xuebao 24 (2008) 223–228.
[38] J.H. Zhu, H.P. Lee, K.M. Lim, S.J. Lee, D.Y. Wang, Evaluation and comparison of
nasal airway flow patterns among three subjects from Caucasian, Chinese and
Indian ethnic groups using computational fluid dynamics simulation, Respir.
Physiol. Neurobiol. 175 (2011) 62–69.