Simulation

Special Issue
Simulation: Transactions of the Society for
Modeling and Simulation International
1–16
Agent-based modeling and simulation Ó The Author(s) 2017
DOI: 10.1177/0037549717712602
of blood vessels in the cardiovascular journals.sagepub.com/home/sim

system

xebnem Bora1, Vedat Evren2, Sevcan Emek1 and Ibrahim C

S xakırlar1

Abstract
The purpose of this study is to develop a model to simulate the behavior of the human cardiovascular system for use in
medical education. The proposed model ensures that the output of the system is accurately represented in both equili-
brium conditions and imbalance conditions including in the presence of adaptive agents. In this study, field experts
develop an agent-based blood vessel model, i.e., a submodel for the stated purpose. In the proposed blood vessel model,
vessels are represented by agents whereas blood flow is represented by the interaction between agents. Adaptive beha-
vior shown by vessels in terms of resistance to the blood flow is defined by the agents’ properties, which are used as the
basis for calculating and graphically representing the physical parameters of blood flow, specifically blood pressure, blood
flow velocity, and the resistance of the vessel. The adaptation of the vessel agents is supported by a case study, which
demonstrates the adaptive behavior of the blood vessel agents through a negative feedback control mechanism. The
blood vessel model proposed is flexible in nature such that it can be adapted to account for the behavior of the vessel
sections in any vascular structure.

Keywords
Agent-based modeling and simulation, human cardiovascular system, blood vessel

1. Introduction simple behaviors. As a result of the vessel agents’ beha-

In medical education, computational simulation models
have become increasingly important in developing an
understanding of the complex mechanisms and dynamics
of the human cardiovascular system (CVS). In order to
explore the functioning of the CVS, we have developed an
agent-based model, a submodel of the CVS model, that
focuses on blood flow by taking into account the real-
world behavior of heart and blood vessels.1
In this study, we present our blood vessel model in
detail. Simplicity is the hallmark of our model, and also
the feature that distinguishes it from existing models in dif-
ferent fields.1–4 That is, our system is based on simple rules
that define interactions between elements of the model,
thereby obviating the need for complicated mathematical
calculations. For this reason, the model proposed is also
easy to understand, which we consider a significant gain in
the context of medical education.
Our system is based on a bottom-up modeling approach
through agent-based modeling.5 On this basis, our model
offers a view of the complex human CVS by considering
the system as a large set of interacting vessel agents with
viors and interactions, the complex and adaptive behaviors
of the system emerge at the macro-level of the system.5,6
Thus, agent-based modeling and simulation (ABMS) pro-
vides a strong foundation for modeling complex systems
that exhibit behaviors of this kind.7, 8 Further, the model
proposed in the present study has a considerable advantage
over existing models because the vessel section can be
adapted to represent any vascular structure in the arterial
system. In other words, the vessel model we propose can
be used to account for any vascular structure by adjusting
the model parameters in accord with the desired ambient
conditions. Via our proposed model, we use the physical
parameters of a healthy adult as a basis for observing the
characteristic behaviors of the vascular structure in rela-
tion to blood flow.
1
Department of Computer Engineering, Ege University, Izmir, Turkey
2
Department of Physiology, Ege University, Izmir, Turkey
Corresponding author:
Sevcan Emek, Department of Computer Engineering, Ege University,
35100, Bornova, Izmir, Turkey.
Email: sevcan.emek@ege.edu.tr
2 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)
The rest of this study is organized as follows. Section 2
offers a brief account of the fundamentals of ABMS.
Section 3 summarizes the CVS and homeostatic control
mechanism. We do briefly refer to studies in which
researchers have offered models of the CVS in the litera-
ture review presented in Section 4, where we consider a
range of viewpoints and research angles. Section 5 pro-
vides a detailed account of our agent-based blood vessel
model, including the calculations used to create the model
and our approach to identifying agent behaviors and estab-
lishing rules regarding them. Section 6 provides a detailed
description of the simulation study, including the results
together with a comparison of the behaviors of the vessel
agents. Section 7 presents a case study in which a negative
feedback control mechanism is used to demonstrate adap-
tive behavior on the part of the blood vessel agents.
Section 8 presents a summary of our multidisciplinary
study including an account of its limitations and recom-
mendations for future research directions.
2. Agent-Based Modeling and Simulation
(ABMS)
ABMS is a powerful technique used to understand the
mechanisms of systems and/or system dynamics shown by
complex phenomena in many domains, including biomedi-
cine1, social sciences,2 ecology,3 and economics.4 A
review of the published studies that draw on ABMS shows
that the concepts of multi-agent modeling, agent-oriented
modeling, and agent-directed simulation are often used
interchangeably. Oren et al. proposed the concept of
agent-oriented simulation as a way to capture all these
concepts,9 thereby simplifying the terminology. However,
recent studies have shown that the field has moved toward
a consensus whereby ABMS has become the prevalent
and accepted term.10 ABMS refers to simulations in which
agents are used to mimic real system behaviors. It consists
of three main components: agents, a simulated environ-
ment, and a simulation environment.
Agents: These are actors operating in the real system,
influencing the simulated environment and influenced by
the simulated environment. The agents are involved in the
simulation model as model components performing actions
individually and interacting with other agents and the
simulated environment to represent behaviors in the real
system.
Simulated environment: This environment consists of
components that cannot be represented as agents in the real
system but represent important components with regard to
the representation of the system. The simulated environ-
ment is part of the simulation model. It also represents the
real environment in which agents live, operate, and inter-
act with each other. In a simulated environment, sources
and state definitions are present and are often shared in the
overall model by all the agents. However, in many simula-
tion models, an environment is defined as ‘‘an empty
world,’’ which allows the environment in which agents
perform their operations to be represented. A simulated
environment has an important role for ABMS as active
components in the real system are frequently affected by
the environment.
Simulation environment: A simulation environment
consists of agents and a simulated environment. It repre-
sents the simulation infrastructure and enables the opera-
tion of the simulation model. In conceptual terms, the
simulation environment is considered part of ABMS. The
simulation environment is conceptualized in this way prin-
cipally because if the simulation model is to represent the
current system in real terms, any needed infrastructure
must be a part of the real system. Further, the simulation
environment is the basis for the model’s operation, record-
ing the results, and a number of infrastructure services
required for the simulation model to function. For this rea-
son, the simulated environment is considered part of the
simulation model.
In the ABMS model, each agent has its own character-
istics and behaviors and acts as a component of the system.
As a result of the behavior of any given agent, other agents
in the simulated environment can be affected. During a
simulation, interactions between agents and/or interactions
between agents and the environment may lead over time to
a behavior or a behavioral pattern that is not defined in the
system. In ABMS, the concept of time is typically repre-
sented as time steps. In every time step, each agent in the
model performs an operation and interacts with other
agents.
Although ABMS is an innovative and solution-based
simulation method for many application areas, it may not
be suitable for some applications or situations. For this rea-
son, it is important to define the cases for which ABMS is
appropriate, a list of which is given next11,12:
 The agent’s decisions and behaviors can be
described separately.
 The agent’s behaviors change or adapt.
 The agent learns and exhibits changeable behaviors.
 Agents have dynamic relationships that can be
shaped and resolved.
 When agents group, the results of adaptation and
learning can be observed at the macro level.
 Based on agents’ past knowledge, the future cannot
be predicted.
 Changes in the agent’s community are observed as
a result of a process rather than as a result of inputs.
 Individualism and/or locality is important to the
system’s operation.
 In multi-level systems, the effects of emergence,
i.e., of lower-level behaviors on higher-level beha-
viors are observed.
Bora et al.
 The properties of homogenous information, linear
decision making, and rationality must not be
included, as these are inappropriate for equilibrium-
based modeling.
3. Human cardiovascular system
At the core of biology stands homeostasis,13 the most
essential concept of life and also the fundamental axiom
of physiology. The term means the maintenance of a rela-
tively stable internal environment for the cells of the
organism with regard to physical and chemical properties.
It must be understood, however, that the internal environ-
ment stands not for the intracellular space, but for the
extracellular fluid. Any cell in an organism uses various
transport mechanisms such as simple diffusion, active
transport, and endocytosis through its semi-permeable cel-
lular membranes to obtain necessary biomolecules and to
dispose of the end-products of cellular metabolism. The
internal environment is the only exchange partner for the
cells in these processes. Therefore, an efficient logistic
network with a surface-increasing branching algorithm
must be in place to effect a steady supply of these neces-
sary substrates to the internal environment of the cells and
to clear the end-products from that environment. For com-
plex multicellular animals, this network is a CVS with
blood as its carrier domain. The CVS functions primarily
to transport gases, nutrients, chemical messengers, heat,
and immunologic elements to target tissues and to remove
from those tissues the chemical and thermal products of
their metabolism.
The physiology of the CVS can be divided into two sub-
sections: the function of the heart and the function of the
vascular system. In terms of the dynamics of the vascular
system, the heart provides energy to increase the pressure
and velocity of the blood in order to perfuse organ vascular
Figure 1. Homeostatic control mechanism.
3
beds. For the heart to do this efficiently, the vascular sys-
tem plays a central role as the distributing conduit. Both
the distributing arteries and the peripheral vascular beds
present the load to the pumping heart. The theory related
to blood flow in the vascular system includes multi-faceted
aspects such as fluid mechanics, the fluid–vessel interface,
vascular tissue engineering, pulse wave propagation, and
mathematical modeling.
Mathematical formulations of blood flow through vis-
coelastic arteries are well established and documented in
various texts.14 There are also experimental measurement
methods and quantitative approaches to assessing the state
of the arterial circulation and microcirculation.
The human CVS adapts itself to asymmetrical and irre-
gular demands via feedback mechanisms in various situa-
tions. These adaptations are achieved by local and global
control mechanisms.
Almost all homeostatic control mechanisms are based
on negative feedback. These mechanisms change the vari-
able back to its original state or ‘‘ideal value’’ and have at
least three interdependent components, as shown in
Figure 1.15
The receptor/sensor is the sensing component that
monitors and responds to changes in the environment.
When it senses a stimulus, the receptor sends information
to a control center, i.e., to the component that sets the
range at which a variable is maintained. The control center
determines an appropriate response to the stimulus and
generates output accordingly. The control center then
sends signals to an effector, which can be muscles, organs,
or other structures. The effector receives output from the
control center. After the effector receives the signal, a
change occurs to correct the deviation by depressing it
with negative feedback.
A number of negative feedback loops are shown in
Figure 2.16 The local and central mechanisms are
4 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)

Figure 2. Cardiovascular control system.

integrated to ensure that all the tissues have sufficient

blood flow to function properly. In order to provide an
adequate supply of blood to the tissues, the CVS adapts to
their ever-changing metabolic and physical demands. The
blood flow is controlled locally within the tissue and sys-
temically via neural and endocrine measures. Local con-
trol of blood flow is adjusted at the level of arterioles and
small arteries and is the result of auto-regulation and
hyperemia. Systemic control takes place when there is a
perturbation in the systemic circulation. A disturbance of
this nature is sensed by various receptors and relayed to
the central nervous system for integration and then to the
effectors for correction. Figure 2 presents the overall view
of systemic regulation.
The task of the CVS is to circulate the blood pumped
from the heart around the body through the vessels.
Therefore, the parameters such as the heart-pumped blood
pressure, blood flow, and blood flow velocity are very
important in terms of vascular system dynamics. A
detailed description of these parameters is given in
Section 5.
The vascular system is a blood vessel system that con-
sists of the arterial system and the capillaries and venous
system. The arterial system comprises arteries that carry
oxygen-rich blood away from the heart through the arter-
ioles and thereby throughout the body. The aorta is both
the largest and the main artery of the human body.
Capillaries, in which diffusion occurs between blood and
cells in the tissues, are the smallest vessels. Further, capil-
laries connect the arterial and venous circulatory subsys-
tems. The venous system involves veins, such as the vena
cava and pulmonary veins. Veins are blood vessels that
carry oxygen-poor blood back to the heart. The largest
veins in the human body are the vena cava. These are two
Figure 3. Closed loop perfusion system.
large veins that enter the right chamber of the heart, from
above and below.17, 18
The CVS is a closed and balanced system. The heart
and blood vessel agents appropriate for the structure,
shown in Figure 3, were identified in our simulation
program.
The heart is defined as two agents: the left heart and the
right heart. The left heart interacts with the blood vessel
agents. The right heart agent waits to receive information
from the blood vessel agents. The lung agent provides oxy-
gen to the heart but does not interact with these agents. In
the present study, we assume that the blood pumped from
the left heart is oxygen-rich; therefore, the lung agent is
not included in the model. The blood vessel agents, i.e.,
the aortas, capillaries, and veins are defined as uniform,
and we refer to them as arteriole vessels. The value of the
blood pressure, pumped from the left heart, is the source
of the message between the blood vessel agents. The beha-
viors and roles of these agents in the system are described
in Section 5.
Bora et al. 5

4. Literature review articles and case studies associated with traditional compu-
tational approaches. Despite the related works, unmet
ABMS has been applied to many domains, including bio-
requirements in the simulation of the human CVS remain.
logical systems,19,20 several branches of the social
Other works evaluate the human CVS as a normal system
sciences,21–23 management systems,24,25 economic sys-
that consists of predefined components and functions
tems,26 physiological systems,27,28 ecological systems,29,30
accordingly. Such research does not address the adaptation
and animal societies.10 The CVS has been computationally
or self-organization needs of the CVS. Given this gap in
simulated in numerous studies, which rely on control the-
the literature, we model the human CVS as a complex sys-
ory,31 electrical circuit models,32–34 biomechanical tech-
tem consisting of autonomous interacting individuals that
niques,35 and physical and mathematical calculations.36
react to local and/or global perturbations in their environ-
The CVS is an extremely complicated homeostatic
ment by adjusting their behaviors. Thus, the simulation we
mechanism, which involves sub-systems. Almost all these
present of the human CVS is considerably more realistic
studies lack stochasticity, which is important for biological
than the models described in the literature to date. In bio-
systems. Further, the studies rely on differential equations
medical systems, building an environment is important for
to model the CVS process. These mathematical models do
agents to identify their neighbors and the state of the latter
not simplify the system or contribute to the development
and to behave accordingly. The environment includes tis-
of the system. It should also be noted that these models
sue, an extracellular matrix, or biological variables. In our
are not capable of providing accurate three-dimensional
study, the environment has the ability to utilize the con-
(3D) visualizations or interactivity. The literature includes
text, which enables the agents to take on a hierarchical
only a few studies of the CVS in which ABMS is used.
organization. Our context includes a network, a grid, and a
Among the studies that do rely on ABMS is one by Al-
continuous space in Repast Simphony. In biomedical
Jaafreh and Al-Jumaily,27 who estimated the cardiovascu-
applications, agents are usually represented as a large
lar parameters using a multiagent system (MAS) based on
number of biological cells, elements, or gene-proteins. In
combining two independent methods: pulse wave velocity
some of the applications we researched, agents traverse
(PWV) and heart rate and artery resistance. Utilizing
from one region to another region linked by paths through
multi-agent cooperation decreases the error percentage
branches of the circulatory system. The MAS concept in
and improves the accuracy of the cardiovascular para-
simulating complex biomedical processes offers flexibility
meters. Faber et al. simulated the blood flow in the capil-
and open-endedness. However, compared with mathemati-
lary vessels of the human body with the additional
cal models, biomedical applications in MAS require more
presence of foreign bodies and estimated the scope and
memory to be built into the data structures of the program-
usefulness of the data obtainable from such an environ-
ming and the experiments. In our system, we divided the
ment.37 Yazdanbod and Marcus presented a 3D, interac-
blood vessels into segments, each of which is represented
tive, agent-based model of the blood coagulation process
by an agent. We define the blood vessels as agents because
within the Lindsay Composer (LC) computational frame-
we consider it important to create a model that is flexible
work.28 Mehdizadeh et al. presented a brief survey of
in nature. As we continue to develop the model, it will be
some of the biological and biomedical applications of
a straightforward matter to add parameters and behavior
ABMS in biomedical engineering.38 These studies are
rules to the agents. Although the CVS is much more com-
related to cancer, tissue engineering, angiogenesis, lung
plicated in terms of blood vessel models, we developed a
disease, morphogenesis, and epidemiology. Mehdizadeh
basic blood vessel model in order to understand the sys-
et al. developed a multi-layered agent-based framework to
tem. In future work, we plan to develop the complex blood
model the process of sprouting angiogenesis (blood vessel
vessels model with fractals using the ABMS technique.
formation) within polymeric porous scaffolds used in
regenerative medicine.38 Mabry and Baldwin focused on
modeling cardiovascular flow with a migrating agent sys-
5. Agent-based blood vessel model
tem.39 Using the SimAgent framework, the researchers
described the basic cardiovascular flow problem followed
5.1. Modeling the vessel
by a correlation between the migrating agent paradigm In a real system, the vessels transport the nutrients needed
and physiological processes. In this study, agents repre- to cells and tissues by branching, a process that can be
senting blood flow traverse from one region to another abstracted as a parallel distribution channel. It is not feasi-
along emanating paths. The researchers drew on parallel- ble for a single vessel to reach throughout the body, thus
processing approaches and considered the ways in which supplying all the latter’s nutrient needs. Therefore, con-
the SimAgent framework provides for a unique computa- ceptualizing such a model is of little use. Instead, the more
tional approach to simulating cardiovascular processes. connections there are between the vessels, the easier it is
The CVS studies to date evince a range of modeling to access tissues and cells. We have designed our model to
and simulation approaches. We investigated scholarly reflect this reality.
6 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)
Figure 4. Schematic illustration of the need for communication
between tissue sites and upstream arterioles supplying those
regions.40
In the present study, we have developed a blood vessel
model, as shown in the dark round pattern in Figure 4.
According to Secomb, increased metabolic demand in a
tissue region (*) requires the vessels feeding that region to
dilate (+) in order to provide increased perfusion.40
In Section 7, we discuss the adaptive behavior of the
agent-based blood vessels via a negative feedback control
mechanism.
In Figure 5, the vessel is divided into vessel segments
referred to as vessel agents, each of which mimics the ves-
sel’s functionality. Information is transferred between con-
secutive agents.
Each agent has its own identity knowledge (ID), loca-
tion knowledge (x, y, z), and radius knowledge (r).
Incoming information is transmitted from the first agent to
the next agent and from that agent to the next until the last
agent is reached. The vessel agents are all the same type
and same length.
We implement a well-known pattern referred to as
‘‘publish-subscribe’’ in order to provide interaction
between the agents. Our main objective in selecting this
pattern is to mimic the vessels’ real behavior. In CVS,
each vessel knows its target vessel(s). Therefore, during
the creation of the agents, each vessel knows which ves-
sel(s) to send messages to about the blood flow. These lat-
ter vessels are referred to as listeners. The principal
advantage of this arrangement is that it ensures only the
target agent state is updated as a result of messaging. In
the CVS, blood may take 1–2 s to flow from the left heart
to the right heart. We can replicate this brief time lapse by
minimizing the messaging infrastructure delay in the simu-
lation. Each agent subscribes to the corresponding agent
and publishes messages in order to propagate messages to
its listeners. On receiving a message, the listener agent
updates itself and regulates its behavior according to the
current environment situation. Each vessel agent evaluates
its current situation according to its local knowledge and
behaves accordingly.
Figure 5. Vessel agents.
5.2. Behavior rules of the vessel agent
When a segment of a vessel is assumed to be in the form
of a straight rigid pipe, hydrodynamic equations should be
used for the continuity of the blood flow in the vessel.16,41
Using the following equations, we calculated the rules of
behavior for the vessel agents and the physical parameters
of blood flow arising from these rules.
The diameter decreases as the distance (d) from the
starting point (origin) increases by moving from the first
to the last agent, as shown in Figure 5. The cross-sectional
area (A) of each agent is calculated by:
Aðd Þ ¼ Að0Þek0 d ð1Þ
where A(d) is the cross-sectional area at distance d along
the vessel in cm2 and A(0) is the cross-sectional area of the
initial agent calculated by:
Að0Þ ¼ PIr02 ð2Þ
in accord with the exponential function of distance.16 The
tapering factor (k0) per length is given in the table in
Figure 6.
The radius value of the initial agent (r0) is entered in
the table in Figure 6 by the programmer. The radius values
of the other agents (nth agent) are calculated by:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
TAPERINGFACTOR nAGENT LENGTH
Að0Þe r0
rn ¼ ð3Þ
PI
rnt ¼ rnt1 + Dr ð4Þ
 
Dr ¼ perturbationt  perturbationðt1Þ rmax ð5Þ
Each vessel segment can be affected by local and/or
global perturbations. Therefore, a vessel segment can
undergo a change in diameter as a result of a perturbation.
For this reason, we modified Equation (3) with Dr to rep-
resent changes in diameter in response to a perturbation.
Dr refers to the change in the diameter of a vessel agent at
the tth tick and is directly proportional to the difference
between the perturbations at the tth tick and at the (t2 1)th
Bora et al.
Figure 6. Model parameters.
tick. Therefore, each vessel agent can react to a perturba-
tion, should one arise, and can adjust its diameter autono-
mously, adaptively, and independently of other vessel
agents. However, in this study, our model does not respond
to changes in perturbations (i.e., Dr = 60.3 cm), as this
phenomenon is most closely related to the autonomic ner-
vous system, which is not our focus here.
As shown in Figure 6, agent length is calculated by:
VESSEL LENGTH
AGENT LENGTH ¼ ð6Þ
AGENT COUNT
i.e., the ratio of the length of the vessel (VESSEL_
LENGTH) to the number of agents (AGENT_COUNT),
both of which are entered by the programmer.
Blood flow (Q) is the volume of fluid displaced in a
vessel in unit time. Its unit is defined by cm3/sn or mL/
100 g/min. Blood flow is described by:
Q ¼ DP=R ð7Þ
which is also known as Ohm’s law.17
7
Blood flow in normal microcirculation is regulated to
meet the metabolic demands of the tissues. Blood flow is
dependent on the affecting perfusion (DP) pressure and the
radius of the blood vessels. DP refers to the pressure differ-
ence between the initial agent and the last agent. R refers to
the vessel’s resistance to the blood flow. The resistance
offered by peripheral circulation is known as systemic vas-
cular resistance. The pressure difference allows blood to
flow through the vessel. Blood pressure refers to the pres-
sure applied to the vessel wall in a unit of time. The flow
resistance (R) in the vessels occurs because of the blood
rubbing against the blood vessel walls. Blood flow resis-
tance, referred to as Poiseuille’s law,17,41 is explained by:
8Viscosity
Rn ¼ AGENT LENGTH ð8Þ
PIrn4
where Rn is the resistance calculated for each agent (nth
vessel agent segment). Viscosity refers to the blood flow
characteristics within the vessel. This value is given by the
user in Figure 6.
The blood flow velocity (VELOCITY) in the vessel is
calculated by:
VELOCITY ¼ Q=An ð9Þ
The blood flow rate refers to displacement in the vessel at
time t. Its unit is cm/s.
6. Simulation details
We developed our model using the Java language with the
Repast Simphony 2.2 tool.42
The key definitions for our agent-based blood vessel
model developed in the simulation environment are as
follows:
 Vessel length, cycle duration, agent count, tapering
factor, viscosity, and blood density constitute the
model parameters. Each model parameter has a pre-
defined value, as shown in Figure 6. The user can edit
these values in order to observe the effects of and
changes in the behaviors/outputs of the simulation.

The mean arterial pressure (MAP) of the vessel, the
minimum blood pressure (PRESSURE_MIN) of a
healthy adult is defined as [60, 89], and the maxi-
mum blood pressure (PRESSURE_MAX) is
defined as [100, 140].
 A cardiac cycle (tðK Þ) is represented in 1 period,
which, in turn, is represented as a 100 simulation
cycle (tick count).
We simulated two types of vessel models: one in the
form of a single vessel without branching (Figure 7), and
the other in the form of a branching vessel (Figure 8).
8 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)
Figure 7. Vessel model without branching interface.
On comparing the findings for the model without
branching vessels to the findings for the model with
branching vessels, we found that agent behaviors change
from the point at which branching begins. This difference
is shown in Table 1.
Cross-sectional models of main vessel A and branching
vessel B, which branches from A, are shown in Table 1.
Branching, which starts at a point of main vessel A, is
stated as a percentage. Our principal reason for stating the
value as a percentage is that the length of the vessel varies
according to the physical size of the individual. The length
of main vessel A per 80% corresponds to the 40th agent
(An agent) in the simulation environment. The 40th agent
continues to transmit the message from this point to the
next agents. The 40th agent also transmits the message to
the branching vessel B agents. The connection, created
from the first agent of the branching vessel B (B0 agent)
changes the global behavior of the main vessel A from that
point. When a new branching vessel C (or more branching
vessels), which branches from B, is included in the model,
the behavior parameters of branching vessels remain the
same as in Table 1.
An examination of the behavior parameters shows that
the diameter of the vessels to tissues and cells gradually
decreased. Even though the effective cross-sectional area
increases as a result of branching, the vessels narrow adap-
tively so that the perfusion pressure does not drop below
the critical value when the terminal point is reached.
Therefore, the radius of the vessel agents decreases along
the vessel length in both models. The radii of the vessel
agents are shown in Figure 9.
The MAP of the vessel agents depends on the pressure
values pumped by the heart periodically during a cardiac
cycle, as shown in Figure 10. MAP is determined by the
cardiac output, the systemic vascular resistance, and the
perfusion pressure.
The graphical results obtained from comparisons of
applying behavior parameters in Table 1 to both models
are presented in Figures 11, 12, 13, and 14. According to
these results, we offer the following observations:
 In Figure 11(a), the cross-sectional areas of main
vessel A’s agents decrease. However, as the arterial
vascular bed expands from the point of branching,
the total cross-sectional area of main vessel A’s
and branching vessel B’s and C’s respective agents
leads to an increase in the total effective cross-
sectional area, as shown in Figure 11(b).
Bora et al. 9

Figure 8. Branching vessel agent model interface.

Table 1. Comparison of vessel agents’ behaviors in a blood vessel branching pattern.

Behavior parameters Vessel model without branching Vessel model with branching

Main vessel (A)

Branching vessel (B)

Radius (r) rAn < rAðn + 1Þ Decrease rAn > rAðn + 1Þ + rB0 Increase
Cross-sectional area (A) AAn < AAðn + 1Þ Decrease AAn < AAðn + 1Þ + AB0 Increase
(depending on radius)
Resistance (R) 1 Increase RA ∼ 1 Decrease
RA ∼ rð4A + BÞ
rA4
Blood flow (Q) QAin ffi QAout Stable QAn ffi QAðn + 1Þ + QB0 Decrease
QAn ffi QAðn + 1Þ
Velocity (V) VA n > VA ð n + 1 Þ Increase VA n < V A Decrease
ðn + 1 Þ
VA n < VB
Pressure (P) PA n > PA ð n + 1 Þ Increase PAn < PAðn + 1Þ Decrease
PA n < PB
10 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)

 Blood flow velocity increases constantly depending

on the narrowing of the vessel, as shown in Figure
14(a). Blood flow velocity decreases in accord with
an increasing effective cross-sectional area from the
branching point, as shown in Figure 14(b). After the
point at which branching begins, blood flow velo-
city starts increasing as the radius of the vessel
agents decreases.

These results in respect of the parameters were obtained

in reference to the normal conditions of a healthy adult.
The following studies prove the medical validity of our
experimental results:

 Deviha at al. demonstrated numerical simulation

Figure 9. Graphic of vessel agent radius.
 The vessel agents’ resistance to blood flow
increases as the radius decreases, as shown in
Figure 12(a). However, due to branching, as shown
in Figure 12(b), this resistance decreases depending
on the radius of main vessel A’s and branching ves-
sel B’s and C’s respective agents.
 The blood flow entering the vessel without branch-
ing is approximately equal to the blood flow leav-
ing through the vessel without branching, in Figure
13(a). However, due to branching, as shown in
Figure 13(b), the blood flow is spread depending
on the radius of the vessel agents. The total blood
flow in the branches equals the blood flow entering
the vessel, i.e., the point at which branching begins.
results for the normal human being in regard to ves-
sel, radius, total cross-section, wall thickness, aver-
age velocity, vessel length, and calculation of flow
rate.43
 Jayalalitha et al. numerically calculated CVS para-
meters for a normal patient and an abnormal
patient.44 In their model, the flow analysis is evalu-
ated in reference to the area of the cross-section,
pressure drop, permeability, the pressure difference,
Poiseuille’s equation, Darcy’s law, and Reynold’s
number.
 Betts e al. demonstrated relationships among blood
vessels that can be compared in terms of vessel dia-
meter, total cross-sectional area, average blood
pressure, and the velocity of blood flow in the
table.45

Figure 10. Graphic of mean arterial pressure.

Bora et al. 11

Figure 11. (a) Graphic of effective cross-sectional area of main vessel A. (b) Graphic of effective cross-sectional area of branching
vessel A.

Figure 12. (a) Graphic of resistance of main vessel A. (b) Graphic of resistance of branching vessel A.

Figure 13. (a) Graphic of blood flow of main vessel A. (b) Graphic of blood flow of main vessel A, branching vessel B and C.

7. Case study: local autoregulation of
blood flow
In this case study, we observed the adaptive behavior of
our agent-based blood vessel model during metabolic
activity, which may increase or decrease depending on
external factors. Using a negative feedback control
mechanism, we simulated the metabolic response of
agent-based blood vessels (Figure 15).
If the metabolic activity increases, the oxygen demand
of the tissue increases. In a 100 tick run time under normal
conditions, we increased the metabolic activity in the 30th
tick count. In Figure 15, the arterioles are represented by
our branching vessel model, as described in the previous
section.
The receptor agent is stimulated by the chemical factors
(oxygen (O2) and carbon dioxide (CO2)). The receptor
agent relays the O2 and CO2 levels to the control center. In
the control center, O2 and CO2 levels are controlled by set
point levels. If the CO2 level increases (Figure 16) and the
O2 level decreases (Figure 17), then the arterioles dilate in
12 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)
Figure 14. (a) Graphic of blood flow velocity of main vessel A. (b) Graphic of blood flow velocity of main vessel A, branching vessel
B and C.
Figure 15. Negative feedback blood flow control system.
response to a higher level of CO2. In the simulation, the
radius of the blood vessel agents is dilated to 0.3 cm. The
widening of the blood vessels, i.e., vasodilation, increases
blood flow to the tissue, bringing additional O2 to meet the
increased metabolic demand and removing waste CO2.
In real life situation, especially for the local feedback
regulation, the same anatomical structure may assume all
the roles for the receptor, the control center and the effec-
tor. However, in order for our abstraction approach to be
comparable, we chose to apply the same data flow logic
for all levels of simulation.
In the simulation, metabolic activity is increased in the
30th tick count. Until this tick count, the partial pressure of
the carbon dioxide (PaCO2) is normal range. The PaCO2
value range is 38–42 mmHg at sea level.46 When meta-
bolic activity starts to increase, the PaCO2 level gradually
increases, as shown in Figure 16.
In Figure 17, tissue O2 saturation (SaO2) decreases
while CO2 production increases. Both low O2 and high
CO2 dilate the blood vessel agents. The normal value range
of SaO2 is 94–100% at sea level.46 However, the critical
value of SaO2 is 89%.
In our simulation, when the threshold value falls below
89%, the negative feedback control system is activated and
blood flow increases (Figure 18).
In Figure 18, the normal value range of the blood flow
is approximately 50 mL/100 g/min at sea level.46–48 In the
presence of both low SaO2 and high PaCO2, the blood ves-
sel agents dilate and so the blood flow increases and regu-
lates SaO2 and PaCO2. When metabolic activity starts to
increase, the blood vessels dilate and blood flow increases
and regulates the metabolic response.
Blood vessel resistance is influenced by both systemic
and local control mechanisms. Local control of blood ves-
sel resistance matches tissue blood flow to the metabolic
needs of the tissue. In Figure 19, when the resistance falls
as the blood vessel agents dilate, the blood flow increases
(active red line).
Bora et al.
Figure 16. Partial pressure of the carbon dioxide (PaCO2) of blood vessel agents.
Figure 17. Oxygen saturation (SaO2) of blood vessel agents.
We modeled the agent-based blood vessels as the main
effectors of blood flow regulation, as a crucial aspect of
the metabolic response is that arterioles need to receive
information about the metabolic status of the tissue they
supply.
The flow diagram of the closed negative feedback
mechanism is shown in Figure 20.49
13
An increase in oxygen demand by the cells in a tissue
leads to increased blood flow. High tissue metabolism
releases a large amount of vasodilator substance compris-
ing CO2, lactic acid, adenosine, histamine, and hydrogen
ion. These substances cause vasodilation by relaxing of
the smooth muscle on the vessels which in turn reduces
resistance and causes increased blood flow. When blood
14 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)

Figure 18. Relationship between blood flow, PaCO2, and SaO2.

Figure 19. Resistance of blood vessel agents in normal

conditions (passive blue line) and resistance of dilating the blood
vessel agents (active red line).

flow increases, the CO2– the end-product of ATP produc-

tion – can be cleared from the tissue, which in turn reduces
hydrogen ion.17 It is explained by:

CO2 + H2 O!H2 CO3 !H+ + HCO

3 ð10Þ
After the metabolic activity of the tissue returns to nor-
mal levels, vasodilator substance concentrations and the
arteriole diameter return to their initial states.
8. Conclusions
In this study, we have developed an agent-based blood
vessel model of the CVS. In a real system, the vascular
Figure 20. Local autoregulation of blood flow.
structure is governed by multiple physical and chemical
parameters. The vessel-related functions of only some of
these parameters were investigated in our simulation study.
The blood vessel model sets a good example for modeling
other vessels in the vascular system as well. However, this
case shows how difficult it is to take a complex real
Bora et al.
system and simulate it in a virtual environment. For this
reason, the present study is interdisciplinary in nature,
which is a strength, although some difficulties can arise in
creating a system’s engineering scenario. Members of the
interdisciplinary research group use different problem-
solving methods to solve problems. Moreover, members
are influenced by how others perform their tasks.
Interdisciplinary studies require far-reaching conceptual
integration and require a mature foundation in the respec-
tive disciplines.50,51 This study emerged in accordance
with the principles of interdisciplinary research. Therefore,
we anticipate that the results reported herein will be used
as a tool for both educational exploration and research.
Following this study, we plan to create an arterial vas-
cular bed map in which compared to our model here will
account for considerably more branching vessels.
Our goal is to provide a model of the distribution of
blood flow to tissues and organs via this map and to com-
plete the CVS closed loop model together with the venous
system. We anticipate that in the not-too-distant future,
our model developed will be used to educate medical stu-
dents and as a basis for engaging in research focused on
both healthy individuals and on those requiring cardiovas-
cular intervention.
Funding
This research received no specific grant from any funding agency
in the public, commercial, or not-for-profit sectors.
References
1. Shi Z, Wu C and Ben-Arieh D. Agent-based model: a surging
tool to Simulate infectious diseases in the immune system.
Open J Modell Simul 2014; 2: 12–22.
2. Epstein J and Axtell R. Growing artificial societies: aocial
science from the bottom up. Washington, DC: Brookings
Institution Press, 1996.
3. Grimm V, Revilla E, Berger U, et al. Pattern-oriented model-
ing of agent-based complex systems: lessons from ecology.
Science 2005; 310: 987–991.
4. Phan D and Varenne F. Agent-based models and simulations
in economics and social sciences: from conceptual explora-
tion to distinct ways of experimenting. J Artif Societies Social
Simul 2010; 13: 5.
5. Crespi V, Galstyan A and Lerman K. Top-down vs bottom-up
methodologies in multi-agent system design. Auton Robots
2008; 24: 303–313.
6. Di Marzo Serugendo G, Gleizes M-P and Karageorgos A.
Self-organising software from natural to artificial adaptation.
Springer Science & Business Media, 2011.
7. North MJ and Macal CM. Managing business complexity: dis-
covering strategic solutions with agent-based modeling and
simulation. Oxford: Oxford University Press, 2007.
8. Wolf T and Holvoet T. Emergence versus self-organisation:
different concepts but promising when combined. Lect Notes
Comput Sci 2005; 3464: 1–15.
15
9. Oren T, Numrich SK, Uhrmacher AM, et al. Agent-directed
simulation: challenges to meet defense and civilian require-
ments. In: Proceedings of the 2000 winter simulation confer-
ence (WSC) (ed JA Joines, et al.), Orlando, FL, 10–13
December 2000, pp.1757–1762. IEEE.
10. Macal CM and North MJ. Agent-based modeling and simu-
lation. In: Proceedings of the 2009 winter simulation confer-
ence (WSC), Austin, TX, 13–16 December 2009, pp.86–98.
IEEE.
11. Macal CM and North MJ. Tutorial on agent-based modeling
and simulation. J Simul 2010; 4: 151–162.
12. Hare M and Deadman P. Further towards a taxonomy of
agent-based simulation models in environmental manage-
ment. Math Comput Simul 2004; 64: 25–40.
13. Marieb EN and Hoehn K. Human anatomy and physiology:
cardiovascular system. 8th ed. San Francisco, CA: Benjamin
Cummings, 2010.
14. Aaf AN. Circulatory system dynamics. Biophysics and
Bioengineering Series. Vol. 1. Academic Press, 1978.
15. Koc H. Homeostasis and physiological control systems.
Lecture Notes, 2013.
16. Li JK. Dynamics of the vascular system. World Scientific, 2004.
17. Guyton AC and Hall JE. Textbook of medical physiology.
Elsevier Health Sciences, 2015.
18. Moini J. Anatomy and physiology for health professionals
with companion web site. Burlington, MA: Jones and Bartlett
Learning, 2012.
19. Railsback S, Lytinen S and Jackson S. Agent-based simula-
tion platforms: review and development recommendations.
Simulation 2006; 82: 609–623.
20. Tobias R and Hofmann C. Evaluation of free Java-libraries
for social-scientific agent based simulation. J Artif Societies
Social Simul 2004; 7: 6.
21. Griffin AF and Stanish C. An agent-based model of pre-
historic settlement patterns and political consolidation in the
Lake Titicaca Basin of Peru and Bolivia. Struct Dyn 2007;
2(2).
22. Malleson N. Using simulation to predict prospective burglary
rates in Leeds and Vancouver. In: 7th national crime map-
ping conference, Manchester, UK, 7–8 May 2009.
23. Carley KM, Fridsma DB, Casman E, et al. BioWar: scalable
agent-based model of bioattacks. IEEE Trans Syst Man
Cybern Part A Syst Humans 2006; 36: 252–265.
24. Conway SR. An agent-based model for analyzing control poli-
cies and the dynamic service-time performance of a capacity-
constrained air traffic management facility. In: 25th congress
of the international council of the aeronautical sciences (ICAS
2006), Hamburg, Germany, 3–8 September 2006.
25. Van Dam KH, Lukszo Z, Ferreira L, et al. Planning the loca-
tion of intermodal freight hubs: an agent based approach. In:
Proceedings of the 2007 IEEE international conference on
networking, sensing and control, London, 15–17 April 2007,
pp.187–192. IEEE.
26. Charania AC, Olds JR and DePasquale D. Sub-orbital space
tourism market: predictions of the future marketplace using
agent-based modeling. In: 57th international astronautical
congress (IAC), 2006 http://www.sei.aero/eng/papers/uploads/
archive/IAC-06-E3.4_present.pdf, accessed: 01.06.2017 (2006,
accessed 1 June 2017).
16 Simulation: Transactions of the Society for Modeling and Simulation International 00(0)
27. Al-Jaafreh M and Al-Jumily A. Multi agent system for esti-
mation of cardiovascular parameters. In: Computers, com-
munications, & signal processing with special track on
biomedical engineering, Kuala Lumpur, Malaysia, 15
November 2005, pp.269–299. IEEE.
28. Yazdanbod I and Marcus S. An agent-based simulation of
blood coagulation processes. J Undergraduate Res Alberta
2011; 1: 13–18.
29. Mock KJ and Testa JW. An agent-based model of predator-
prey relationships between transient killer whales and other
marine mammals. Technical report, University of Alaska
Anchorage, USA, May 2007.
30. Mast EH, van Kuik GA and Van Bussel GJ. Agent-based
modelling for scenario development of offshore wind energy.
Report, Delft University of Technology, The Netherlands.
2007.
31. Wu Y, Allaire P, Tao G, et al. Modeling, estimation and con-
trol of cardiovascular systems with a left ventricular assist
device. In: American control conference, Portland, OR, 8 Jun
2005, pp.3841–3846. IEEE.
32. Naik KB and Bhathawala PH. Mathematical modelling and
simulation of human systemic arterial system. Int J Eng
Innovative Technol 2014; 4(1).
33. Takahashi S, Sakawa K, Shiraishi Y, et al. Modeling, simu-
lation and parameter estimation of the cardiovascular system
by using model based approach. In: InSICE annual confer-
ence (SICE), Nagoya, Japan, 14 September 2013, pp.493–
500. IEEE.
34. Ghasemalizadeh O, Mirzae MR, Firoozabadi B, et al. Exact
modeling of cardiovascular system using lumped method.
In: International conference on bioinformatics & computa-
tional biology (BIOCOMP), Las Vegas, NV, 14–17 July
2008, pp.408–417.
35. Arıbasx E, PisxkinS
x and C
xelebi MS. 3D blood flow simulations
in human arterial tree bifurcations. In: 14th national biome-
dical engineering meeting, Izmir, Turkey, 20–22 May 2009,
pp.1–4. IEEE.
36. Fajdek B and Golnik A. Modeling and simulation of human
circulatory system. In: 15th International conference on
methods and models in automation and robotics,
Miedzyzdroje, Poland, 23 August 2010, pp.399–404. IEEE.
37. Faber L, Boryczko K and Kisiel-Dorohinicki M. Hybrid
architecture for simulation of blood flow with foreign bodies.
In: 28th European conference on modelling and simulation
(ECMS 2014), Brescia, Italy, 27–30 May 2014, pp.523–529.
38. Mehdizadeh H, Artel A, Brey EM, et al. Multi-agent systems
for biomedical simulation: modeling vascularization of por-
ous scaffolds. In: International conference on principles and
practice of multi-agent systems, Wollongong, NSW, 16–18
November 2011, pp.113–128. Berlin: Springer.
39. Mabry SL and Baldwin K. Modeling cardiovascular flow
with a migrating agent system. In: Proceedings of interna-
tional conference on health sciences simulation: 2000 west-
ern multiconference, San Diego, CA, 23–27 January 2000.
40. Secomb TW. Theoretical models for regulation of blood
flow. Microcirculation 2008; 15: 765–775.
41. Gabrys E, Rybaczuk M and Kedzia A. Blood flow simula-
tion through fractal models of circulatory System. Chaos
Solutions Fractals 2006; 27: 1–7.
42. Nikolai C and Madey G. Tools of the trade: a survey of vari-
ous agent based modeling platforms. J Artif Societies Social
Simul 2009; 12(2): 2.
43. Deviha VS, Rengarajan P and Hussain RJ. Modeling blood
flow in the blood vessels of the cardiovascular system using
fractals. Appl Math Sci 2013; 7: 527–537.
44. Jayalalitha G, Deviha VS and Uthayakumar R. Fractal model
of the blood vessel in cardiovascular system. In: 15th interna-
tional conference on advanced computing and communications,
Guwahati, India, 18–21 December 2007, pp.175–182. IEEE.
45. Betts JG, College TJ, DeSaix P, et al. Anatomy and physiol-
ogy, Chapter 20.2.BC Open Textbooks, https://open
textbc.ca/anatomyandphysiology/chapter/20–2-blood-flow-
blood-pressure-and-resistance/ (2013, accessed 1 June 2017).
46. Hadjiliadis D, Zieve D and Ogilvie I. Blood gases,
www.nlm.nih.gov/medlineplus/ency/article/003855.htm (2015,
accessed 1 June 2017).
47. Anaesthesia UK. Cerebral blood flow (CBF), www.frca.
co.uk/article.aspx?articleid=100754 (2007, accessed 1 June
2017).
48. Shardlow E and Jackson A. Cerebral blood flow and intracra-
nial pressure. Anaesth Intensive Care Med 2008; 9: 222–225.
49. Silverthorn DU. Human physiology: an integrated approach.
4th ed. Upper Saddle River, NJ: Pearson Education, 2007.
50. Newell WH. The case for interdisciplinary studies: response
to Professor Benson’s five arguments. Issue Interdiscip Stud
1983; 2(1): 19.
51. Jones C. Interdisciplinary approach: advantages, disadvan-
tages, and the future benefits of interdisciplinary studies.
ESSAI 2010; 7(1): 26.
Author biographies
Sxebnem Bora received her PhD degree from Ege
University, Turkey in 2006. She is currently an assistant
professor in the Computer Engineering Department at Ege
University. Her research interests include dependable
computing, multiagent systems, self-adaptive systems,
complex-adaptive systems, and agent-based modeling and
simulation.
Vedat Evren is an MD-PhD at the Ege University School
_
of Medicine, Department of Physiology, Izmir, Turkey.
Sevcan Emek is a PhD candidate and a research assistant
in the Computer Engineering Department at Ege
University, Turkey. She received the BS and MS degrees
in Computer Engineering from Dumlupınar University in
2009 and 2011, respectively. Her research interests
include artificial intelligence, control systems and agent-
based modeling and simulation.
Ibrahim Cxakırlar received the BS, MS, and PhD degrees
in computer engineering from Computer Engineering
Department of Ege University in 2007, 2009, and 2015,
respectively. He is currently a software engineer in
Concur France SAS.