Proceedings of the 20th World Congress

Proceedings
The of
of the
International
Proceedings 20th
20th World
Federation
the Worldof Congress
Automatic Control
Congress
Proceedings
The of the
International 20th World
Federation of Congress
Automatic Control
Toulouse,
The
The France,
International
International July 9-14,
Federation
Federation 2017
of
of
Available
Automatic online at www.sciencedirect.com
Automatic Control
Control
Toulouse,
Toulouse, France,
France, July
July 9-14,
9-14, 2017
2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 5861–5866
A
A New
New Approach
Approach of
of Dynamic
Dynamic Fuzzy
Fuzzy
ACognitive
New Approach of Dynamic
Knowledge Fuzzy
Networks in
Cognitive
Cognitive Knowledge
Knowledge Networks
Networks in
in
Modelling
Modelling Diagnosing
Diagnosing Process
Process of
of Meniscus
Meniscus
Modelling Diagnosing
Injury 
Process of Meniscus
Injury
Injury 
Antigoni P. Anninou ∗∗ Peter P. Groumpos ∗∗
Antigoni
Antigoni P.
P. Anninou
Poulios ∗∗ ∗ Peter P. Groumpos ∗∗ ∗
Panagiotis
Antigoni
Panagiotis P. Anninou
Anninou
Poulios
∗ Peter
∗∗ Ioannis
Peter P. Groumpos
P.Gkliatis
Groumpos
∗∗ Ioannis Gkliatis ∗∗ ∗∗
∗
Panagiotis Poulios
Panagiotis Poulios Ioannis Gkliatis ∗∗ Ioannis Gkliatis ∗∗
∗
∗ Laboratory of Automation and Robotics, Department of Electrical and
∗ Laboratory of Automation and Robotics, Department of Electrical and
∗ Laboratory of Automation
Computer
Laboratory
Computer Automation and
of Engineering,
Engineering, and Robotics,
University
Robotics,
University
of Department
Patras, Greece
of Department
Patras, Greece
Greece
of
of Electrical
(e-mails: and
Electrical
(e-mails: and
Computer
Computer Engineering,
anninou@ece.upatras.gr,
Engineering, University
University of Patras,
groumpos@ece.upatras.gr).
of Patras, Greece (e-mails:
(e-mails:
∗∗ anninou@ece.upatras.gr,
anninou@ece.upatras.gr,
Orthopaedic Clinic, Department groumpos@ece.upatras.gr).
groumpos@ece.upatras.gr).
of Medicine, University of
∗∗ anninou@ece.upatras.gr, groumpos@ece.upatras.gr).
∗∗ Orthopaedic Clinic, Department of Medicine, University of
∗∗ Orthopaedic Clinic, Department
Patras,Greece
Orthopaedic
Patras,Greece
(e-mails:
(e-mails:Clinic, Department of
panpoulios@gmail.com,
panpoulios@gmail.com, of Medicine,
Medicine, University
University of
gliatis@hotmail.com)
gliatis@hotmail.com) of
Patras,Greece (e-mails: panpoulios@gmail.com,
Patras,Greece (e-mails: panpoulios@gmail.com, gliatis@hotmail.com) gliatis@hotmail.com)
Abstract: A new approach of Dynamic Fuzzy Cognitive Knowledge Networks is presented. This
Abstract:
Abstract:
is an evolutionary A
A newnew approach
approach
type of Fuzzy of
of Dynamic
Dynamic
Cognitive Fuzzy
Fuzzy Maps Cognitive
(FCM)Knowledge
Cognitive that arose Networks
Knowledge Networks
from the need is
is presented.
presented.
for updating This
This
Abstract:
is an A
evolutionary new approach
type of of
Fuzzy Dynamic
Cognitive Fuzzy Maps Cognitive
(FCM) Knowledge
that arose Networks
from the is
needpresented.
for updating This
is
is an
classic
an evolutionary
methodology
evolutionary type
type of
in order
of Fuzzy
Fuzzy Cognitive
to overcome
Cognitive itsMaps
Maps (FCM)
drawbacks,
(FCM) that arose
concerning
that arose from the
the single
from the need for
calculation
need for updating
updating rule,
classic
stabilitymethodology
classic methodology
and real time in
in order
order
problems to
to overcome
overcome
and expand its
its drawbacks,
its use inconcerning
drawbacks, a variety the
concerning the single
single
of applications.calculation
calculation This rule,rule,
new
classic
stability methodology
and real timein order
problems to overcome
and its
expand drawbacks,
its use inconcerning
aa variety the
of single calculation
applications. This rule,
new
stability
approach
stability and
is being
andbeing real time
tested
real time problemsproblems
for its and
accuracy expand
and expand in its use
Decision in
Support
its use Support variety
in a varietySystems of applications.
in medicine,
of applications. This
trying
This new new
to
approach
model kneeis
approach being tested
isinjuries tested
by using for
for17its
itsrealaccuracy
cases ofin
accuracy in Decision
Decision
patients. The Support Systems
Systems
new proposed in
in medicine,
model is able totrying
medicine, trying
diagnose to
to
approach
model knee is being
injuries tested
by using for 17 its
real accuracy
cases of in Decision
patients. The Support
new Systems
proposed in
model medicine,
is able to trying
diagnose to
model
meniscus
model knee
knee injuries
injuries
injuries by
and
by using 17
to distinguish
using 17 real
real cases of
between
cases of patients. The
acute The
patients. new
and new proposed
degenerative
proposed model is able
injury.isSubsequently
model able to
to diagnose
diagnose we
meniscus
meniscus injuries
injuries and
and to
to distinguish
distinguish between
between acute
acute and degenerative injury. Subsequently we
observe the
meniscus
observe the
evolution
injuries
evolution and of of the
to the injury by
distinguish
injury bybetween acute and
administering
administering and
a
degenerative
a proposed
degenerative
proposed
injury.
treatment
injury. by
treatment
Subsequently
by the physician.
Subsequently
the physician.
we
we
observe
Results
observe the
of
the evolution
this new
evolution of
method,
of the
the injury
which
injury by
are
by administering
presented
administering in a proposed
detail,
a are
proposed treatment
very satisfactory
treatment by
by the
for
the physician.
both
physician. two
Results
Results of treatment
of this new
this new method, which areagreement
presented with in detail,
detail, are very
very satisfactory foroutcomes.
both two two
levels and
Results
levels and this new method,
of treatment stage, and
method,
stage, and
which
in total
which
in total
are presented
areagreement
total presented with in Magnetic
in detail,
Magnetic
are
are very satisfactory
Resonance Imaging
satisfactory
Resonance Imaging
for both
foroutcomes.
both two
levels
The
levels and
whole treatment
methodology
and treatment stage, is
stage, theand
the in
outcome
and outcome agreement
of
in total agreement a close with Magnetic
collaboration
with Magneticbetween Resonance
between
Resonance Imaging
engineers
Imaging and outcomes.
medical
outcomes.
The
The
doctorswhole
wholeandmethodology
is significant is
methodology is the
because outcome of
of
it is a promisingaa close
close collaboration
collaboration
tool which sets between
aside the engineers
engineers and
and
main disadvantages medical
medical
The
doctorswholeand methodology
is significant is the
because outcome
it is a of a
promising close collaboration
tool which between
sets aside theengineers
main and medical
disadvantages
doctors
of Fuzzy
doctors and
and is
Cognitive
is significant
Maps
significant because
and
because allows it
it is
isus a
a promising
a wide
promising use tool
in
tool which
many
which sets
real aside
time
sets aside the
problems.
the main
main disadvantages
disadvantages
of Fuzzy
of Fuzzy Cognitive
Cognitive Maps Maps and and allows
allows us us aa wide
wide use in in many
many realreal time
time problems.
problems.
of FuzzyIFAC
© 2017, Cognitive Maps and
(International Federation a wide use
allowsofusAutomatic use in many
Control) realbytime
Hosting problems.
Elsevier Ltd. All rights reserved.
Keywords: Fuzzy Control, Dynamic Fuzzy Cognitive Knowledge Networks, Meniscus Injuries
Keywords:
Keywords: Fuzzy Fuzzy Control,
Control, Dynamic Dynamic Fuzzy Fuzzy Cognitive
Cognitive Knowledge
Knowledge Networks,
Networks, Meniscus
Meniscus InjuriesInjuries
Keywords: Fuzzy Control, Dynamic Fuzzy Cognitive Knowledge Networks, Meniscus Injuries
1. INTRODUCTION the previous value of the concept in the computation of
1.
1. INTRODUCTION
INTRODUCTION the previous
valuevalue
previous value of
of the concept
2010). in the computation of
1. INTRODUCTION the
the
the
new
previous
new value value of the
(Groumpos,
(Groumpos, the concept
concept
2010).
in
in thethe computation
computation of of
Fuzzy Cognitive Map (FCM) is a modelling method that the the new
new value
Nevertheless,
(Groumpos,
value there
(Groumpos,
are
2010).
2010).
some drawbacks that should be
Fuzzy
Fuzzy Cognitive
Cognitive even Map
Map with (FCM)
(FCM) is
is a modelling method that
works efficiently
Fuzzy Cognitive Map (FCM) missing
is aa modelling
modelling
data. Experts, method
method that Nevertheless,
for each
that Nevertheless,
overcome. Thethere
there
existingare
are
areidea
some
some of drawbacks
drawbacks
a single calculation that
that should
that should
be
be
rule for
works
works
case study,
works
efficiently
efficiently
efficiently
even
evenwith
support
even
with
with their
with
missing
missing
missing
data.
data.
Experts,
data. Experts,
knowledge Experts, for
for each Nevertheless,
each
for each
the developed overcome.
overcome. The
The
there
existing
existing
some
idea
idea of
of
drawbacks
a
a single
single calculation
calculation
should
rule
rule
be
for
for
case study, support with their knowledge the developed all the
overcome. concepts
The creates
existing problems.
idea of a Firstly
single all concepts
calculation rule are
for
case
FCMs.
case study,
Theysupport
study, support
have been with
withused their
their knowledge
in knowledge
a large range the
the ofdeveloped
ofdeveloped
applica- all all the concepts
the concepts creates
creates problems. Firstly allissue
concepts are
treated in the same way.problems.
That causes Firstly
a bigall concepts are
in which
FCMs.
FCMs.
FCMs.
They
They have
tions supporting
They have
have
been
been used
decision
been used
systems
used
in
in a
in ainlarge
a large
medicine
large
range
range of applica- all
of applica-
range(Kannappan applica-
the concepts
treated
treated in
in the
the
creates
same
same way.
way.
problems.
That
That causes
causes
Firstly
aaoutput
big
big
allissue
concepts
issue in
in
are
which
which
tions supporting decision systems in medicine (Kannappan we should
treated in define
the sameinitial
way. values
That for the
causes a big on
issue advance.
in which
tions supporting
et al.,supporting
2011), economy decision systems
(Ginis, 2015), in
in medicine
zero energy (Kannappan
buildings we we should
tions decision systems medicine (Kannappan This makesdefine
should define initial
initial
our tool values
values
initialdifficult
values to
for
for the
the
for use
output
output
output on
the because
on
weadvance.
advance.
want a
et
et
et
al.,
al., 2011),
(Vergini
al., 2011),
2011), et al., economy
2012),(Ginis,
economy
economy (Ginis,
education
(Ginis,
2015),
2015),(Cole
2015),
zero
zero energy
zero energy
energy buildings we
buildings
and Persichitte,
buildings This
This
should
makes
makes
define
our
our tool
tool difficult
difficult to
to use
use because
because
on
we
we
advance.
want
want a a
(Vergini final
This unknown
makes our diagnosis,
tool and
difficult the
to use procedure
because of
we random
want a
2000) etc.et
(Vergini
(Vergini etThe
et
al.,
al.,
2012),
al., basic
2012),
education
2012),calculation
education rule
education
(Cole
(Cole
(Cole that and
and
and
Persichitte,
Persichitte,
computes
Persichitte, the final final unknown
unknown
definition diagnosis,
diagnosis,
of thediagnosis,
output values and
and the
the
complicateprocedure
procedure of
of
the method random
random
and
2000)
2000)
value of etc.
etc.each The basic
The concept calculation
basic calculation
calculation
at every rule rule
rule that computes
that computes
simulation computes
step is the the final
the definition unknown and the procedure of random
2000)
value etc.
of eachThe basic
concept at every that
simulation step is the definition
affects
definitiontheofof the
the
entire
of
output
process.values
the output
output values complicate
complicate
In addition
values complicate changes the
the
the in
method
method and
and
the values
method and
value
following
value of
of each
(Kosko,
each concept
1986)
concept at
at every
every simulation
simulation step
step is
is the
the affects
affects the
the entire
entire process.
process. In
In addition
addition changes
changes in
in the
the values
values
following (Kosko, 1986) of concepts
affects the affect
entire the
process. system
In in
addition a way
changes that in is
thedifficult
values
following (Kosko, 1986) of
of concepts affect the system
following (Kosko, 1986) to
of concepts
be predicted
concepts affect
and
affect thecould
the system
systemnot in be aaaimported
in
in
way
way that
way thatinis
that is
difficult
isreal
difficult
time
difficult
N to
to be
be
intobethepredicted
predicted
model toand and
and could
could
be tested. not
not be
be
Alsobethere imported
imported is a highin
in real
real
degreetime
time
of
 to predicted could not imported in real time
Ai (n + 1) = f (k2 Ai (n) + k1  
N
N A (n)w ) (1) into
into the
the
difficulty model
model
to to
to
adjust be
be tested.
tested.
each Also
Also
different there
there
problem is
is a
a high
high
to this degree
degree of
of
existing
A (n + 1) = f (k A (n) + k N
A
j
(n)w
ji
) (1) into the
difficulty model
to to
adjust be tested.
each Also
different there
problem is a high
to this degree of
existing
A ii (n + 1) = f (k2
Ai (n + 1) = f (k2 Ai (n) + k1
2 A ii (n) + k1 1 j=1,j = i A jj (n)wji )
Aj (n)wji ) (1) difficulty
ji (1) difficulty
model. to
Another
to adjust
adjust each
important
each different
aspect
different problem
is
problem the to this
analysis
to this existing
of the
existing
j=1,j
j=1,j==ii
model.
model. Another
evolution Another important
important
of the resulting aspect
aspectthrough
networks is
is thethe analysis
analysis of
of the
time (Bourgani the
j=1,j=i model. Another important aspect is the analysis of the
Ai (n + 1) is the value of the concept Ci at the iteration evolution evolution
et al., 2014).
evolution
of
of
of
the
the
the
resulting
resulting
Inresulting
addition networks
networks through
through
FCMs through
networks have stability time
time
time
(Bourgani
(Bourgani
problem
(Bourgani
A
A (n n+1,
+ 1) is the
(n) value
is theof the concept
of the C ii at
at thetheCiteration et al., 2014).
real In addition FCMs have stability problem
Aiii (n
step
step
+
+ 1)
(n n+1, 1) is isA
A
jthe
the
(n)
value
value
is the
of
ofvaluethe
the concept
value concept
of the
C
concept
C iteration
j at the
i at theCiteration
concept at the
et
et al.,
al., 2014).
regarding 2014). In
In addition
world systems
addition FCMs have
(Carvalho
FCMs have and stability
Tomé,
stability problem
2002).
problem
step n+1, A j (n) is the value of the concept C j at the regarding
regarding real
real world
world systems
systems (Carvalho
(Carvalho and
and Tomé,
Tomé, 2002).
iteration
step n+1,
iteration Cstep
step
stepA n,
j (n) wis
ji is
the the valueweightof of
the
j n, w is the weight of interconnection
interconnection
concept C j atfrom
j the
from
These
regarding
These
reasons are
real world
reasons are
enough to examine
systemsto(Carvalho
enough examine
the
and
the Tomé, 2002).
possibility of
2002).
possibility of
iteration
concept
iteration step
j to n,
n, w
w ji
concept is the
ji is the C weight
and f of
is interconnection
the sigmoid
i weight of interconnection from from
function. These reasons
a new, reasons
These are
more comprehensive
are enough
enough to
toandexamine
flexiblethe
examine the possibility
model. In order
possibility of
of
concept C ji
concept
”k ” C
expressesjj toto concept
concept
the influence C
C ii and
and
of theff is
is the
the sigmoid
sigmoid
interconnected function.
function.
concepts a
a
to new,
new,
model more
more comprehensive
comprehensive
behavior of a and
and
nonlinear flexible
flexible
dynamic model.
model.
system, In
In aorder
order
new
concept
”k
1 C
”” expresses j to concept
the influence C i and f is the sigmoid function. a new, more comprehensive and flexible model. In order
”k
”k ” expresses
in 111the configuration
expresses the of the of
the influence
influence of the
the
new
of thevalue
interconnected
interconnected
of the concept
interconnected
concepts
concepts
concepts Ai to to
type
to
model
model behavior
behavior
of Dynamic
model behavior Fuzzy
of
of a nonlinear
Cognitivedynamic
of aa nonlinear
nonlinear dynamic
Knowledge
dynamic
system,
system,
system,
a
a new
a new
Networks
new
in
in
and the
the configuration
configuration
k2 configuration of
of
represents theof proportion the
the new
new value
value
of theof
of the
the concept
concept
contribution A
A i type
type of
of Dynamic
Dynamic Fuzzy
Fuzzy Cognitive
Cognitive Knowledge
which is Knowledge
focused on Networks Networks
in the
and k represents the the new value
proportion of of the
the concept Aof
contribution
i
i
of
(DFCKN)
type of is developed,
Dynamic Fuzzy Cognitive Knowledge dynamic
Networks
and k (DFCKN) is developed,
developed, which is’Cognitive
focused on on dynamic
and k22 represents
represents the the proportion
proportion of
 The 2present work was financially supported by the ”Andreas
of the
the contribution
contribution of of (DFCKN)
aspects of our
(DFCKN) is variables. We
is developed, which
which useis focused
is’Cognitive
focused on dynamic
Knowledge’
dynamic
aspects
aspects
because of
of
we our
our
want variables.
variables.
to emphasizeWe
We use
use
in ’Cognitive
the existing Knowledge’
Knowledge’
knowledge

 The present
present work was financially
financially supported by thethe ”Andreas aspects
because of our
we want variables.
to emphasize We use
in the’Cognitive Knowledge’
existing knowledge
Mentzelopoulos
 The
The present work Scholarships
was Universitysupported
of Patras”.by by the ”Andreas because
Mentzelopoulos
Mentzelopoulos Scholarships
work was financially
Scholarships University
University of
supported
of Patras”.
Patras”.
”Andreas because we we want
want to to emphasize
emphasize in in thethe existing
existing knowledge
knowledge
Mentzelopoulos Scholarships University of Patras”.
Copyright 2017 IFAC
2405-8963 © 2017, 6050Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Peer review©
Copyright © 2017
2017 IFAC
IFAC 6050
6050
Copyright ©under
2017 responsibility
IFAC of International Federation of Automatic
6050Control.
10.1016/j.ifacol.2017.08.1289
Proceedings of the 20th IFAC World Congress
5862
Toulouse, France, July 9-14, 2017 Antigoni P. Anninou et al. / IFAC PapersOnLine 50-1 (2017) 5861–5866
extracted by experts and exploited, so as to generate new
knowledge through cognitive process. As far as networks
are concerned, cognitive maps form an extensive network
(Papageorgiou and Stylios, 2008).
In this new model the new calculation rule should be
comprised by two equations. These equations will consist
not only of the inputs and outputs of the system but also
the states. The dynamic combination of the three concepts,
States, Inputs and Outputs will consist the Dynamic Fuzzy
Cognitive Knowledge Network. The combination of the
three concepts in two new, different equations, extracted
and following the basic rules of fuzzy modelling, will be
the core of the new approach, which will be applied in
modelling the diagnostic procedure of meniscus injury and
tested in 17 real patients. After the distinction between
acute and degenerative injury, we examine the influence
of daily analgesics in the evolution of the injury through
time.
2. DYNAMIC FUZZY COGNITIVE KNOWLEDGE
NETWORK
The new proposed system uses the basic theory of FCM
with the difference that the system main variables are not
treated in the same way. FCM are specialized in complex
systems and are suitable for medical decisions because
they model the human thinking, using a reasoning process
that can include uncertain descriptions. The procedure
of constructing the FCM is remaining the same. First
of all experts should define all together the basic system
variables. But the main change is that after this procedure,
variables should be divided into three categories:
• Fuzzy States
• Fuzzy Inputs
• Fuzzy Outputs
The fuzzy inputs concern signals that stimulate the sys-
tem. The fuzzy states of a dynamic system refer to a
minimum set of variables, known as state variables, which
fully describe the system and its response to any given set
of inputs. The fuzzy output variables constitute those that
we should examine their behaviour. In that way we take
into consideration what exactly each concept does. We no
more treat outputs in the same way as inputs and inputs
are separated from states. The mathematical description
of the system and the combination of initial states and
inputs are sufficient to provide information about both the
future states and outputs. Mathematically the standard
form of the new model is described by the system general
equations (Yu, 2005),(Rowell, 2002):
x(k + 1) = f [Ax(k) + Bu(k)]
y(k) = f [Cx(k) + Du(k)]
where x(k)∈ Rn is a state vector, u(k)∈ Rr is an exogenous
known input vector, y(k) ∈ Rm is the output vector in time
unit ”k”. ”k” is the discrete instant of time that input
and state vectors are defined. This is in stark contrast
with ”n” variable in equation (1). The latter expresses
the iteration steps until convergence. In the new model
the number of iteration steps depends on the samples
defined by ”k”. Each sample adds a new iteration step. The
state variables are an internal description of the system
and in combination with the system inputs are sufficient
for computing the output. Figure 1 illustrates the system
structure.
Fig. 1. System states, inputs and outputs
Matrices A, B, C and D contain constant coefficients that
describe our system. Function f is a nonlinear activation
function and is intentionally omitted because the aim of
this paper is to test the conclusions from the new separated
fuzzy equations. So we use a linear model and in order to
interpret the output we use a set of fuzzy rules since the
nature of medical problems and decisions does not require
a precise numeric interpretation of the output between
0 and 1, but an ambiguous interpretation that simply
determines whether the patient is healthy.
Arrays A, B, C and D are the connection weight matrices,
associated with the fuzzy state, input and output vectors.
More specifically, array A is a weight matrix that connects
the states between them and it corresponds to the weight
matrix of classic FCM but now it connects only the states
of the system. Array B is a new weight matrix which
represents how each input influences the states one by
one. Array C is the weight matrix of the interrelationship
between the states and the outputs and array D depicts
the influence of the inputs to the outputs respectively.
Arrays A, B, C and D must be decided by experts, de-
scribing the linguistic interrelationships between vectors.
The exact procedure is the same as in the construction of
a classic Fuzzy Cognitive Map (Groumpos, 2010). Experts
should define all these interconnections in their natural
language using linguistic variables and after a defuzzifica-
tion method of Center of Area, the arrays will be filled
with numeric values.
The algorithm that describes the new model in detail,
adjusted in medical diagnosis follows:
Step 1: Ask experts to define the basic concepts
Step 2: Experts separate them into Fuzzy States, Inputs,
and Outputs
(2)
Step 3: Experts construct weight matrices between
Outputs-States, States-States, Outputs-Inputs, States-
(3) Inputs, using fuzzy variables
Step 4: Convert weight matrices into numeric values using
a defuzzification method
Step 5: Detect initial states in time unit ”k” (Patients
Initial Symptoms)
Step 6: Compute Initial Output in ”k” from equation (3)
→ Diagnosis
6051
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Antigoni P. Anninou et al. / IFAC PapersOnLine 50-1 (2017) 5861–5866 5863
Step 7: According to diagnosis, physicians propose Treat-
ment
Step 8: Treatment is a new Input in our system in time
variable ”k”
Step 9: Compute new States in time unit ”k+1” with
equation (2)
Step 10: Compute new output in ”k+1” according to
equation (3)
Step 11: Renew the new input value in ”k+2”
Step 12: Repeat steps 9-11 until desired time unit
3. MODELLING MENISCAL TEAR USING DFCKN
Immediate diagnosis and treatment is of the utmost im-
portance in medicine. In knee injuries physicians are forced
to use the solution of MRI in order to have verified results.
Is this method necessary though? A new integrated tool
that would work supportively to the physician has to be
implemented in order to make the right decision. This
paper models the diagnostic process of meniscus injury
using DFCKN and constitutes an extension of paper of
Anninou et al. (Anninou et al., 2016) that used the classic
theory of Fuzzy Cognitive Maps. Apart from these the
evolution of the disease through time with the proper
treatment is examined for the first time.
The model will be consisted of two levels. In the first level
we decide if the patient has meniscus tear and if the answer
is positive the final diagnosis in the second level concerns
the distinction between acute and degenerative injury.
In the first level the output vector is Meniscal Tear (MT)
and a high value of that corresponds to a positive result
according to equation (3).
The output is directly depended on the states. This specific
output is not affected by any input and therefore matrix
D is zero and equation (3) is being transformed to (4).
y(k) = Cx(k) (4)
The system states x(k) are comprised of the patients
symptoms at the specific time of the examination and
their initial values could be on or off, so they take two
values, 0 or 1 (Kandasamy and Smarandache, 2003). All
possible symptoms are the following 13: Clicking, Catch-
ing, Giving way / weakness, Localized pain, Episodic pain,
Pain with activity, Pain with pivoting/twisting, Change in
quality/pattern of pain, Locking, Acute swelling, Subacute
swelling, Weight bearing, Continued in athletic activity,
and they have been explained in (Anninou et al., 2016).
The state vector is a vertical vector consisting of the
following elements: C1, C2, C3, C4, C5, C6, C7, C8, C9,
C10, C11, C12, C13.
The way that the 13 states interact with the output
has been decided by experts and after the defuzzification
method of Center of Area (CoA) it is illustrated in matrix
C.
C = [0.75 0.375 0.375 0.75 0.5 0.75 0.875 0.25 0.875 0.25 0.675 0.5 0.5]
Instead of the sigmoid function f that used in FCM to
transform the output into the interval [0,1] in combi-
nation with an interpretation criterion in order for the
6052
output to be fully understandable, we now use fuzzy rules
to immediately transform the numeric value to a fuzzy
variable,because of the nature of medical problems that
demand an explicit linguistic interpretation. Fuzzy rules
resulted from the entire bandwidth spectrum of the output
value and are the following:
• IF the output is between 0-3.0125 THEN the output
fuzzy variable is ”Low”.
• IF the output is between 3.0126-6.025, THEN the
output fuzzy variable is ”Medium”
• IF the output is between 6.026-9.0375, THEN the
output fuzzy variable is ”High”
• IF the output is between 9.0376-12.05, THEN the
output fuzzy variable is ”Very High”
If the fuzzy variable is High or Very High then the patient
suffers from meniscal tear. In the first case with variable
Low, the patient is healthy and when the fuzzy variable
is Medium we need extra information to conclude to a
decision. This information concerns clinical examination
results (Tests) (Anninou et al., 2015) that influence the
output and the final decision. Possible Tests are: Mcmur-
rey’s test, Thessal’s test (20◦ knee flexion), Steinmann’s
I test, Joint Line Tenderness, Ege’s test, Childress’ test,
Appley’s test. So the second fuzzy rule is translated as
follows:
• IF Meniscal Tear is Medium AND one or more tests
from the list: T1, T2, T3, T4, T5, T6 are positive
THEN the patient suffers from meniscal tear. IF
Meniscal Tear is Medium AND only test T7 is positive
THEN the patient does not suffer from meniscal tear.
According to the diagnosis if the patient has meniscal tear
we continue to the next level FCM so as to understand
the exact injury between acute and degenerative. In that
second level we have to take into consideration the Risk
Factors, which are attributes, characteristics or exposures
that increase the likelihood of a person developing a dis-
ease, in order to check if they are susceptible to acute or
degenerative meniscal tears respectively. The total Risk
Factors are: R1: Sport activity / high Tegner score (soc-
cer), R2: Systemic laxity (> 1 score in Beighton scale), R3:
History of previous injury/ surgery, R4: Time From initial
ACL Injury > 2 − 5 years, R5: Lifting weight routine, R6:
Age > 60 years old, R7: Male gender, R8: Work related/
frequently kneeling-squatting, R9: Using stairs frequently,
R10: BMI ≥ 25 kgr/ m2 , R11: Standing / walking routine,
R12: Chronic recurring pain and swelling after exercise,
R13: Minor trauma/Insignificant traumatic injury / un-
able to recall traumatic event, R14: Quads wasting / VMO
assymetry, R15: Pain / symptoms bilaterally present, R16:
Vaguer / more subtle symptoms ( no knee block), R17:
Knee OA present (Snoeker et al., 2013).
The output vector now has two values: Acute Injury and
Degenerative Injury. In this specific level, states of the
system are a vertical vector consisting of Risk Factors: R1,
R2, R3, R4, R5, R6, R7, R8, R9, R10, R11, RF. In addition
D is a zero matrix because the output is influenced only
by the states at this level too. So we use equation (4).
 
Acute
y(k) = (5)
Degenerative
Proceedings of the 20th IFAC World Congress
5864
Toulouse, France, July 9-14, 2017 Antigoni P. Anninou et al. / IFAC PapersOnLine 50-1 (2017) 5861–5866

The last value RF depends on the value of the six risk

factors between R12-R17. If one at least of the values of
R12-R17 is activated then RF=1. But if all values are
zero then ”RF”=-1. According to the following matrix
C, if RF=1 a very high positive weight is added to
degenerative injury and a very high negative weight to the
acute one. That introduces competitiveness to our system
and means that if one or more risk factors between R12-
R17 are activated then we have low possibility of the final
diagnosis to be acute injury. If RF=-1 the reverse process
is performed.
 
0.75 0.75 0.75 0.75 −1 −1 −1 −1 0 0 0 −0.75
C=
−1 −1 −1 −1 0.75 0.75 0.75 0.75 0.5 0.5 0.5 0.75
Competitiveness is very useful in medical decision support
Fig. 2. Meniscus Tear Diagnosis
systems, where we need a dominant output over all the
other possible diagnoses.
We are now ready to examine the model with 17 real case
studies from General University Hospital of Patras.
4. SIMULATION RESULTS

4.1 1st Patient

In order to fully understand the method, the diagnosis

procedure of the first patient will be presented analytically.
Patient 1 had symptoms: C1, C4, C6, C7, C11. Positive
tests were: T1, T2, T3, T4, T5, T6.
According to (4) the arising output value is 3.8 which
corresponds to the fuzzy variable ’Medium’. So we have to
examine the Test results. According to the positive tests
and the second fuzzy rule the patient suffers from meniscal
tear.
Now we examine first patient risk factors arose from his
examination. He is positive to: R1, R3, R4, R6, R7, R8,
R9, R10, R12, R14, R16, R17. First of all we observe that
we have activated values between R12 to R17. That means
that RF=1. According to equation (4) the output values
are the following:
−1.5
y(k) =
1
Consequently the accurate diagnosis for patient 1 is Degen-
erative Injury. The competitiveness reduced significantly
the value of one concept and helped us conclude to one
final decision.
4.2 Overall Patients
The overall data collected by the 17 real patients arrived
at University Hospital of Patras having knee injuries are
presented in the Table 1. According to the first level
analysis, using the second and third column of Table 1,
with symptoms as states and test results as additional
data, we compute the first level output in the last column
using a linguistic variable extracted by fuzzy rules.
Graphically the numeric values of outputs are illustrated
in Fig.2.
As far as the second level results are concerned for the
total patients, they are presented in Table 2 and Fig. 3.
Fig. 3. Final Diagnoses
5. TREATMENT STAGE-SIMULATION RESULTS
All this aforementioned methodology is used in order to
compute the output at the specific time unit k, that patient
arrived at hospital and wanted a diagnosis. But after this
diagnosis, the physician proposes an external intervention,
a treatment, so the existing states of the system are
updated. Therefore in (k+1) time unit the new states are
computed by (2) and the new output is as follows:
(6) y(k + 1) = Cx(k + 1) + Du(k + 1) (7)
The input is the proposed treatment that influences pa-
tients symptoms, and after the computation of the new
states, we compute the new output y(k+1) by using
x(k+1) instead of x(k). Treatment is steady but its influ-
ence to patient, changes through time. For knee injuries,
nonoperative treatments are often proposed, including su-
pervised sessions of physical therapy with scope for indi-
vidual adaptation thrice a week during the first 6 weeks
and then at least twice a week for further 6 weeks under
supervision of a musculoskeletal physiotherapist. It needs 6
weeks for the patient to respond. The number of time steps
we need in order to complete the treatment stage is exactly
the same as the time that patient would go to physician
for re-examination. More specifically we will check the
influence of the method proposed by the physicians that
concerns Daily analgesics (NSAIDs) that were adminis-
tered for those 6 weeks and then as required during the
follow-up. In case of contraindications to nonsteroidal anti
inflammatory drugs (NSAIDs), paracetamol 4 g/day in di-

6053
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Antigoni P. Anninou et al. / IFAC PapersOnLine 50-1 (2017) 5861–5866 5865

Table 1. Data Collected By 17 Patients and Meniscal Tear Diagnosis

Patients Symptoms Positive Tests First Level Diagnosis
1 C1, C4, C6, C7, C11 T1,T2,T3,T4,T5,T6 Medium
2 C1,C2,C4,C6,C7,C11 T1,T2,T5,T6 Medium
3 C1,C3,C4,C6,C7,C11 T1,T2,T5 Medium
4 C1,C4,C6,C7,C11 T2,T3,T4,T5,T6 Medium
5 C1,C2,C4,C6,C7,C11 T2,T3,T4,T6 Medium
6 C2,C4,C7,C11,C12,C13 T1,T2,T3,T5,T6,T7 Medium
7 C1,C2,C3,C4,C7,C11 T2,T4,T5 Medium
8 C2,C4,C6,C7,C11,C12,C13 T1,T2,T5,T7 Medium
9 C1,C2,C4,C5,C6,C7,C11 T3 High
10 C1,C3,C4,C5,C11 T2,T3 Medium
11 C1,C2,C4,C7,C11 T1,T2 Medium
12 C1,C2,C3,C4,C5,C7,C11 T1,T2,T3,T4,T7 Medium
13 C1,C2,C3,C4,C5,C6,C7,C11 T1,T2,T3,T4,T7 Medium
14 C1,C2,C3,C4,C5,C6,C7,C11 T1,T2,T3,T4,T7 High
15 C1,C2,C6,C11 T1,T2,T3,T4,T5 Medium
16 C1,C2,C3,C6,C7,C11,C12,C13 T1,T2,T4,T5 Medium
17 C1,C2,C3,C7 T7 Low → HEALTHY

Table 2. Accurate Diagnosis of 17 Patients

Patients Risk Factors Acute Injury Degenerative Injury Final Diagnosis

1 R1,R3,R4,R5,R6,R7,R8,R9,R10,R12,R14,R16,R17 -4.75 3.25 Degenerative
2 R1,R3,R4,R6 1.25 -2.25 Acute
3 R1,R3,R4,R6,R8,R12,R13,R14,R16 -3.75 1.5 Degenerative
4 R1,R3,R4,R6,R7,R8,R10,R11,R12,R14,R16 -3.75 1.5 Degenerative
5 R6,R7,R8,R10,R11,R12,R13,R14,R15,R16,R17 -9 7.75 Degenerative
6 R6,R7,R8,R10 7.2 -6.75 Acute
7 R1,R3,R4,R6,R8,R9,R10,R11,R12,R13,R14,R15,R16,R17 -5.75 4.5 Degenerative
8 R6,R7,R8,R9,R12,R13,R14,R15,R16 -8 -6.5 Degenerative
9 R3,R4,R6,R8,R9,R10,R12,R13,R14,R15,R16,R17 -6.5 5 Degenerative
10 R8,R9 5 -3.25 Acute
11 R1,R6,R8 4.75 -4 Acute
12 R3,R4,R5,R8,R9,R11,R13,R14,R17 -3.5 2.75 Degenerative
13 R5,R7,R8,R9,R10,R11,R13,R14,R15,R16,R17 -8 7.5 Degenerative
14 R1,R3,R4,R5,R7,R8,R9,R10,R11 5.25 -3.75 Acute
15 R6,R8,R9,R10,R11,R13,R14,R15,R16,R17 -7 6.75 Degenerative
16 R1,R5,R7, R8,R9,R10,R11 3.75 -1.75 Acute

 0 0.375 0.375 0.375 0.375 0.375 0.375 0 0

vided doses combined with tramadol sustained release 100 0.5 0.375 0.5 0.375
mg at night was used. The goal of rehabilitation is to regain  0.5 0 0.5 0.375 0.375 0.375 0.75 0.75 0.75 0.375 0.375 0 0
 0.5 0.5 0 0 0.5 0.5 0.375 0 0.75 0 0 0 0
 
good knee control, range of motion (ROM),flexibility, and  0.5
 0.375 0.375 0 0 0 0 0 0.5 0.75 0 0.75 0

0.375 0.375 0.5 0 0 0.5 0 0 0.75 0 0 0.75 0
muscle strength and thereby to improve knee function. 
0.375 0.375 0.5 0 0.5 0 0 0 0 0

0 0.75 0
 
A = 0.375 0.75 0.75 0 0 0 0 0 0 0 0 0 0
Patient needs 4 re-examinations during these 6 weeks. 0.375
 0.75 0 0 0 0 0 0 0 0 0 0 0
These would be the 4 time steps, k+1, k+2, k+3, k+4, 0.375 0.75 0.75 0.5 0.75 0 0 0 0 0 0 0 0
 
0.675 0.675 0 0.75 0 0 0 0 0 0 0 0 0
for our algorithm. The way that this ”input” interacts 
0.375 0.375 0 0 0 0 0 0 0 0 0 0 0

to the patient in relation to time has a discrete time  
0 0 0 0.375 0.375 0.375 0 0 0 0 0 0 0
function illustration, where in ”k+1”, input is very low and 0 0 0 0 0 0 0 0 0 0 0 0 0
corresponds to u=0.1, in the next selected time variable B = [−0.1 −0.1 −0.1 −0.1 −0.3 −0.3 −0.3 −0.3 −0.3 −0.3 −0.3 0 0]
k+2 the input is u=0.5, in the third iteration of the We observe that proposed treatment will significantly re-
algorithm k+2, u has its highest value (u=1) and keep duce symptoms C5-C11 if they exist. In case study 1
it for the last iteration in the final step k+3. Matrix A patient has symptoms C1, C4, C6, C7, C11. According
expresses the interrelationships between states with each to equations (3) and (7) the symptoms and the output
other and matrix B the influence between input and states, (meniscus injury) are changing as in Table 3.
namely between treatment and symptoms. Experts defined
these relationships with linguistic variables and after the Table 3. Output Iterations
defuzzification method, the two numeric matrices are as
follow. Initial ”k+1” ”k+2” ”k+3” ”k+4”
Output 3.8 1.96 0.94 -0.31 0.1

We observe the fluctuation of output until the final appli-

cation of the 6 weeks treatment as it is illustrated in the
following figure.

6054
Proceedings of the 20th IFAC World Congress
5866
Toulouse, France, July 9-14, 2017 Antigoni P. Anninou et al. / IFAC PapersOnLine 50-1 (2017) 5861–5866
Fig. 4. Evolution of the output in 6 weeks (5 iterations)
The first patient manage to reduce symptoms and knee
injury by following the treatment in a very strict way.
6. DISCUSSION OF THE RESULTS
In the examined sample 16 out of 17 patients suffered from
meniscal tear. The first level output for patient 17 was
LOW, so we did not have to proceed to the second level
as he was healthy. More specifically during second level
simulations, 10 cases were positive to degenerative injury
and the rest 6 to acute injury. The first case (patient 1)
was further analyzed in order to examine the influence of
the daily analgesics (NSAIDs) through time in the course
of patient. Considering that the first patient was following
physician’s orders, the results show full rehabilitation in
his degenerative injury. Due to space limitation the rest
patients treatment is not presented in the paper, as the
methodology is the same.
7. CONCLUSIONS-FUTURE RESEARCH
This new modelling method helps us classify all concepts
and makes it easier to influence them. Any external
intervention in our system will be considered as a new
input. States describe the system and their combination
with the inputs could inform us about the outputs for any
system. The interrelationships between these values must
be decided by experts. In this paper the whole procedure
of knee injury diagnosis was modelled and tested in 17
real cases. The results were verified by MRI results. But
we went one step further in order to examine the influence
of a proposed treatment. The choice of experts should be
very careful because it is obvious that the whole model is
based on them. The general methodology of the separation
of concepts could be the core for the development of a wide
range of applications, not only medical ones.
Future research concerns the use of a validation set to
confirm the accuracy of the model. In addition, model
could be evolved to an integrated advising software tool
useful for physicians for real time clinical use, in order
to diagnose meniscal tear and choose the appropriate
treatment, observing the outcomes of each one.
REFERENCES
Anninou, A.P., Groumpos, P.P., Gliatis, I., and Poulios, P.
(2015). A two-level competitive fuzzy cognitive map for
modelling soft tissue knee injuries. In Bioinformatics &
Bioengineering (BIBE), 2015 IEEE 15th International
Conference on. IEEE.
6055
Anninou, A.P., Poulios, P., Groumpos, P.P., and Gliatis,
I. (2016). A new meniscus injury diagnostic model
using fuzzy cognitive maps. In Control and Automation
(MED), 2016 24th Mediterranean Conference on, 7–12.
IEEE.
Bourgani, E., Stylios, C.D., Manis, G., and Georgopoulos,
V.C. (2014). Time dependent fuzzy cognitive maps for
medical diagnosis. In Hellenic Conference on Artificial
Intelligence, 544–554. Springer.
Carvalho, J.P. and Tomé, J. (2002). Issues on the stability
of fuzzy cognitive maps and rule-based fuzzy cogni-
tive maps. In Proceedings of the 21st International
Conference of the North American Fuzzy Information
Processing Society, NAFIPS2002, New Orleans.
Cole, J.R. and Persichitte, K.A. (2000). Fuzzy cognitive
mapping: Applications in education. International Jour-
nal of Intelligent Systems, 15(1), 1–25.
Ginis, L.A. (2015). The use of fuzzy cognitive maps
for the analysis of structure of social and economic
system for the purpose of its sustainable development.
Mediterranean Journal of Social Sciences, 6(3 S5), 113.
Groumpos, P.P. (2010). Fuzzy cognitive maps: Basic
theories and their application to complex systems. In
Fuzzy cognitive maps, 1–22. Springer.
Kandasamy, W.V. and Smarandache, F. (2003). Fuzzy
cognitive maps and neutrosophic cognitive maps. Infinite
Study.
Kannappan, A., Tamilarasi, A., and Papageorgiou, E.I.
(2011). Analyzing the performance of fuzzy cognitive
maps with non-linear hebbian learning algorithm in
predicting autistic disorder. Expert Systems with Ap-
plications, 38(3), 1282–1292.
Kosko, B. (1986). Fuzzy cognitive maps. International
Journal of man-machine studies, 24(1), 65–75.
Papageorgiou, E. and Stylios, C. (2008). Fuzzy cognitive
maps. Handbook of Granular Computing, 755–774.
Rowell, D. (2002). State-space representa-
tion of lti systems. URL: http://web. mit.
edu/2.14/www/Handouts/StateSpace. pdf.
Snoeker, B.A., Bakker, E.W., Kegel, C.A., and Lucas, C.
(2013). Risk factors for meniscal tears: a systematic
review including meta-analysis. journal of orthopaedic
& sports physical therapy, 43(6), 352–367.
Vergini, E., Kostoula, T.E.C., and Groumpos, P. (2012).
Modeling zero energy building with a three–level fuzzy
cognitive map. Recent Advances in Environmental and
Earth Sciences and Economics, 275–280.
Yu, W. (2005). State-space recurrent fuzzy neural net-
works for nonlinear system identification. Neural pro-
cessing letters, 22(3), 391–404.