Engineering Science and Technology, an International Journal 20 (2017) 1025–1040

Contents lists available at ScienceDirect

Engineering Science and Technology,

an International Journal
journal homepage: www.elsevier.com/locate/jestch

Full Length Article

A fuzzy ontology modeling for case base knowledge in diabetes mellitus

domain
Shaker El-Sappagh a, Mohammed Elmogy b,⇑
a
Faculty of Computers & Information, Minia University, Egypt
b
Faculty of Computers & Information, Mansoura University, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: Knowledge-Intensive Case-Based Reasoning Systems (KI-CBR) mainly depend on ontologies. Ontology
Received 29 December 2016 can play the role of case-base knowledge. The combination of ontology and fuzzy logic reasoning is crit-
Revised 22 March 2017 ical in the medical domain. Case-base representation based on fuzzy ontology is expected to enhance the
Accepted 27 March 2017
semantic and storage of CBR knowledge-base. This paper provides an advancement to the research of dia-
Available online 29 April 2017
betes diagnosis CBR by proposing a novel case-base fuzzy OWL2 ontology (CBRDiabOnto). This ontology
can be considered as the first fuzzy case-base ontology in the medical domain. It is based on a case-base
Keywords:
fuzzy Extended Entity Relation (EER) data model. It contains 63 (fuzzy) classes, 54 (fuzzy) object proper-
Case-based reasoning
Semantic retrieval
ties, 138 (fuzzy) datatype properties, and 105 fuzzy datatypes. We populated the ontology with 60 cases
Case base representation and used SPARQL-DL for its query. The evaluation of CBRDiabOnto shows that it is accurate, consistent,
Fuzzy ontology and cover terminologies and logic of diabetes mellitus diagnosis.
Diabetes diagnosis Ó 2017 Karabuk University. Publishing services by Elsevier B.V. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction
Knowledge Intensive Case-Based Reasoning (KI-CBR) success
depends on its Case Base (CB) structure and content. Ontology con-
structs domain knowledge in a machine-readable format that
humans are capable of understanding. It can be used in CBR sys-
tems as a knowledge representation formalism for specifying
domain background knowledge and CB knowledge [6]. KI-CBR
enables automatic reasoning with semantic knowledge in addition
to the syntactic properties of cases. Ontology supports the creation
of semantic retrieval algorithms to enhance the intelligence of CBR
systems [14]. It provides a conceptualization of the domain, which
consists of concepts, properties, and axioms. It has been utilized in
many medical [15] and non-medical [6] CBR systems.
For diabetes mellitus (DM) diagnosis, ontology has not been uti-
lized in CB, background knowledge, and case retrieval. A crisp CB
OWL 2 ontology for DM diagnosis is proposed by El-Sappagh
et al. [22], a background domain ontology based on SNOMED CT1
(SCT) [23], a CB preparation model from Electronic Health Record
(HER) data [24], and encoded the CB based on a proposed encoding
methodology [27]. Nevertheless, the knowledge imprecision issue
⇑ Corresponding author.
E-mail address: melmogy@mans.edu.eg (M. Elmogy).
Peer review under responsibility of Karabuk University.
1
http://www.ihtsdo.org/snomed-ct/.
has not been handled in KI-CBR paradigm yet [3]. Medical data, such
as DM diagnosis, are mostly imprecise and experience-based. The
success of CBR in this domain depends on how this issue is handled.
According to Zadeh [63], much of the human experience knowl-
edge is imprecise and inaccurate. This knowledge has to be manip-
ulated in a suitable way to prevent the poor effectiveness of
knowledge-based systems. Fuzzy sets have been integrated with
CBR to generate Fuzzy-CBR in many studies as [31]. However, they
did not utilize Fuzzy Ontology (FO) for case representation and
case retrieval causes CBR systems, which can lose many semantic
reasoning capabilities [3]. A mechanism is required to utilize ontol-
ogy as the ‘‘vehicle” for the introduction of fuzzy semantics to KI-
CBR. FO integrates fuzzy set theory into crisp ontology logic, tools,
and languages [18]. Bobillo and Straccia [9] proposed an extension
for crisp OWL 2 to generate fuzzy OWL 2 ontologies. The fuzzifica-
tion of a crisp ontology needs translation into a supported lan-
guage of an FO reasoner such as fuzzyDL [9]. We assert that FO
supports crisp aspects as well as fuzzy aspects. For example, in
DM diagnosis domain, there are crisp components (such as sex
and residence), fuzzy components (such as age and lab tests), and
semantic components (such as diseases), which are related to other
ontology, such as SCT.
Fuzzy ontologies have been used in many domains. For exam-
ple, Rodríguez et al. [57] proposed an FO to model human behavior.
Ali et al. [5] used it in opinion mining. Lee et al. [43,44] used FO and
fuzzy set for modeling diabetes application in diet and diagnosis.

http://dx.doi.org/10.1016/j.jestch.2017.03.009
2215-0986/Ó 2017 Karabuk University. Publishing services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1026 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
Moreover, it has been managed in some AI systems as rule-based
systems [45]. The usage of FO with rule-based systems is less
applicable than other systems because these systems require col-
lecting explicit models of domains. For DM, it is a challenge to col-
lect the set of rules that model it. Therefore, it is possible to
develop CBs that avoid the knowledge-acquisition bottleneck. FO
extends the capabilities of the crisp one and improves the accuracy
and applicability of CBR in medical domain [3].
Park et al. [52] utilized crisp ontology in a fuzzy CBR system for
the prevention ships collision. Alexopoulos et al. [3] tried to build
an FO for CBR using fuzzy algebra. Their proposed ontology has
been designed for electronic libraries. However, utilizing relational
databases and conceptual data models for building fuzzy ontolo-
gies is the most suitable form to build CB fuzzy ontology. Conse-
quently, the patient CB is often gotten from EHR database of
patient histories [10]. There is little research in FO in any medical
CBR systems. As far as we know, there are no studies in the litera-
ture that proposed a DM diagnosis fuzzy CB ontology.
Fuzzy CB ontology can be created from many sources [64]. The
translation of fuzzy EER model other than its relational schema is
better because EER is richer in semantic, and EER model is closely
related to ontology structure [64]. Moreover, many studies pro-
posed mappings from fuzzy EER model to FO [46]. Besides, chronic
diseases management requires collecting cases from the patient’s
(distributed) EHR. El-Sappagh et al. [25] proposed a case base data
model using HL7 RIM.2 A method is required to fuzzify this model
and map it to a fuzzy CB ontology. Because ontology provides a for-
mal representation of a shared conceptualization, the CB FO can be
combined with DM standard domain ontology to support the sharing
and interoperability with EHR environment [23]. Although there are
many studies in fuzzy EER models creation and translation into fuzzy
ontologies and fuzzy DLs [46,47], few studies utilized their results in
CBR especially in medical domains as DM diagnosis. Most fuzzifica-
tion strategies of EER models depend on one linguistic value to rep-
resent a feature [12]. These approaches limit the flexibility of query
problem description. They limit the retrieval algorithms efficiency.
The paper proposes a strategy to solve this issue.
In this paper, we designed a new ontology called CBRDiabOnto.
It is an OWL 2 fuzzy CB ontology to represent DM diagnosis knowl-
edge for any CBR system. The paper concentrates on the knowledge
representation formalism of CB fuzzy ontology. This ontology has
been used to build a whole CBR system [26]. Different types of fea-
tures including text, ordinal, semantic, numerical, and fuzzy types
can be modeled by the proposed ontology. As a result, the resulting
knowledge base supports many types of reasoning, such as seman-
tic, crisp, and fuzzy reasoning, in fuzzy CBR system. The rest of this
paper is organized as follows. Section 2 provides the process of
constructing CBRDiabOnto fuzzy CB ontology. Section 3 is the
querying method for the CB ontology. Section 4 evaluates the pro-
posed fuzzy ontology. Finally, Section 5 presents the conclusion
and future work.
2. Related work
Traditional CBR has been used for diabetes diagnosis in many
studies [48,40]. An evolution of this paradigm is the ontology-
based or KI-CBR [6]. Ontology plays many roles to enhance CBR
semantics [20]. This approach is generally more effective in repre-
senting, indexing, maintaining, adapting, and retrieving similar
cases than traditional ones [30,37]. Case semantic retrieval can
be enhanced by using case-base and domain background knowl-
edge in the form of ontologies [30]. Amailef and Lu [6] proposed
a domain ontology for m-government emergency response ser-
2 3
http://www.hl7.org/implement/standards/rim.cfm.
vices based on CBR. El-Sappagh et al. [23] proposed a diabetes
domain ontology for CBR system based on SNOMED CT. However,
the CBR system fundamentally depends on the case-base knowl-
edge representation [30]. As a result, Heras et al. [38] proposed a
crisp case-base ontology for case-based argumentation system
entitled ArgCBROnto. Zhukova et al. [65] proposed a crisp case-
base ontology for a human resource management system. In the
diabetes domain, ontologies have been used in many CDSS
[61,54]. However, regarding diabetes diagnosis, none of the exist-
ing ontologies is designed for CBR, and few studies have used
ontology in CBR [61,40]; these studies are very abstract and use
ontology only for domain knowledge. Jaya and Uma [39] listed
the roles of ontology in a diabetes diagnosis CBR. El-Sappagh
et al. [22] proposed a crisp OWL 2 ontology for diabetes diagnosis
case-base. This ontology can be used to store and retrieve cases
semantically. On the other hand, diabetes diagnosis depends on
the physician’s experience and the patient’s description of his case.
Vagueness in medical domains can be handled using fuzzy logic
[63], which has been used in diabetes diagnosis rule-based systems
[45]. Moreover, fuzzy logic has been integrated with CBR in hybrid
systems [1] and used for calculating the fuzzy similarity between
cases [41]. Recently, Sohn et al. [59] integrated fuzzy CBR reason-
ing with crisp ontology reasoning for personalized service in a
smart home environment. However, this hybrid system did not
benefit from fuzzy ontology reasoning capabilities.
After the success of crisp ontologies in CBR environment, fuzzy
ontologies can extend the benefits of crisp ontologies [4]. Building
a case-base fuzzy ontology is a challenge [52]. Alexopoulos et al.
[3] proposed a fuzzy case-base ontology by utilizing fuzzy algebra.
With respect to diabetes, it has utilized fuzzy ontologies in many
domains [45,44]. Lee and Wang [45] proposed a five-layer fuzzy
ontology and utilized it in a fuzzy expert system for diabetes man-
agement. Lee et al. [44] proposed a type-2 fuzzy ontology for diet
management in diabetics. Although CBR is more suitable for
managing ill-formed problems as diabetes diagnosis, there is no
fuzzy ontology-based CBR for diabetes, and there are no studies
for case-base fuzzy ontology construction. Crisp ontologies are
not suitable to address imprecise and vague knowledge, which is
inherent in real-world domains [57].
There are many techniques for developing crisp ontology [22].
However, they are not sufficient for fuzzy ontology construction
[64]. Many techniques are available to generate fuzzy ontology
including: extend an existing crisp ontology [4], map relational
database schema, map fuzzy ER model [64], map fuzzy EER model
[64], fuzzy formal concept analysis [19], and other data sources.
The difference between these approaches is mainly in what aspects
of the crisp ontology are being fuzzified, and these aspects depend
on the domain needs. Representation languages such as OWL 2
need to be able to represent fuzzy ontologies. There are two ways
either extend the language itself [60], which changes the structure
of the language or using the current language to represent fuzzy
aspects. Bobillo and Straccia [9] have used annotations to represent
fuzzy in regular OWL 2.3 We will depend on the second choice by
proposing a fuzzy case-base ontology structure (TBOX) based on a
case base EER model. This EER model is extended to fuzzy EER model
based on fuzzy set theory and then converted to fuzzy OWL 2 ontol-
ogy. Moreover, we populate the resulting ontology with the case
from EHR database (ABOX), and we provide a way for semantically
querying the resulting case base ontology. The paper has the follow-
ing contributions:
1. We proposed an EER data model for the DM diagnosis CB
according to our CB schema [26].
http://www.straccia.info/software/FuzzyOWL/.
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
2. We fuzzified this model into a fuzzy EER model using fuzzy set
theory.
3. We proposed a method to map the fuzzy EER model to a fuzzy
CB ontology and implement this ontology in OWL 2 using a
protégé 4.3 fuzzy plugin.
4. We populated the resulting ontology with instances from a
fuzzy CB database. The ontology contains 63 (fuzzy) classes,
54 (fuzzy) object properties, 138 (fuzzy) datatype properties,
105 fuzzy datatypes, and 60 real cases.
5. We proposed a method for fuzzy semantic querying of the
resulting fuzzy knowledge base. The resulting CB fuzzy ontol-
ogy can be used to build CDSS for diabetes diagnosis and can
be extended for disease treatment.
3. The proposed fuzzy ontology
Fuzzy set theory can be applied to the crisp ER model in three
levels [66]. EER model is an extension of ER model by adding some
object-oriented notations, such as generalization/specialization
and aggregation. Many studies have proposed the definition of
fuzzy EER model [64,47], and the conversion of the paper’s case
base crisp data model into a fuzzy EER model is based on this for-
mulation. We do not use fuzzy extensions to OWL for fuzzification
as used in [3]. Fuzzy aspects are modeled as annotations in regular
OWL 2 language using fuzzy OWL 2 syntax [9], and for implemen-
tation, we will use the fuzzy OWL 2 protégé’s plugin.4
The CBRDiabOnto construction process is based on mapping the
diabetes CB EER model into a fuzzy ontology. As in Fig. 1, the first
step generates the case base crisp EER model. This step is detailed
in Section 3.1. The next step is to extend this model to the fuzzy
EER model. This step is discussed in Section 3.2. After that, the gen-
erated fuzzy EER model is mapped to an FO structure (TBOX) using
a formal methodology. This step is shown in Section 3.3. Finally,
the generated ontology is populated with instances from a fuzzy
database (ABOX), which is equivalent to the fuzzy EER model. This
step is detailed in Section 3.4.
3.1. Case base EER model step
We select the DM diagnosis domain as our case study. Diabetes
mellitus (DM) encompasses a group of metabolic disorders charac-
terized by a chronic hyperglycaemic condition, resulting from
defects in insulin secretion, insulin action, or both [7,13,28]. Each
case has a description and solution.
Case structure ¼ fP; S ðPÞg;
where P is the problem description including patient symptoms and
lab test results. This part is described by 64 features.
P ¼ fLiF;LiP; GlS; Age;BMI; Residence;Gender;Occupation;KiF; Lab;UrS;HaPg;
where LiF = liver function tests, LiP = lipid profile, GlS = global symp-
toms, KiF = kidney function tests, Lab = lab tests, UrS = urination
symptoms, HaP = haematological profile. The solution S (P) is
described by five features; it is the DM diagnosis (DM, prediabetes,
gestational DM, and prediabetes gestational) and the likelihood of
suffering from other complications related to liver and kidney.
SðPÞ ¼ fD; L; N; C; Hg
where D = DM diagnosis, L = liver problem, N = nephropathy prob-
lem, C = cancer type, and H = hypercholesterolemia problem. The
full description of the data set can be found in Ref. [26]. We have
proposed a standard CB data model based on HL7 RIM and SCT
ontology in another work. A customized version of the standard
4
http://nmis.isti.cnr.it/~straccia/software/FuzzyOWL/index.html.
1027
model for our CB crisp EER has been described in [27]. Each patient
is a case represented by a row or vector in this data model. A case is
an aggregation of 10 entities. These entities have 70 features. The
patient is described by these features, and the CBR system will sug-
gest a diagnosis of patient conditions. Some features are of string
data types such as sex and residence. Others are ordinal as vision,
thirst, and fatigue. The most of the features are numerical that will
be fuzzy attributes.
3.2. Fuzzy EER model step
For modeling the fuzzy EER, we use the notations in Ref. [49].
Each crisp value is fuzzified into all of the attribute’s linguistic val-
ues. To support similarity calculation using similarity matrices, a
single fuzzy label is used to describe the feature’s crisp value. For
example, if BMI attribute represents a linguistic variable with fuzzy
values {underweight, normal, overweight, and obese}, and these
values are represented by some fuzzy functions, then BMI = 30 is
modeled as lUnderweight ð27Þ ¼ 0, lNormal ð30Þ ¼ 0:52,
lOv erweight ð30Þ ¼ 0:48, lObese ð30Þ ¼ 0, and BMI representative lin-
guistic label Normal, which has the highest membership value.
The linguistic label facilitates the fuzzy query formulation. We
use MATLAB to fuzzify all of the 70 numerical features with the
guidance of our domain experts and DM diagnosis clinical practice
guidelines (CPG), as shown in Fig. 2. In building the fuzzy EER
model, a set of attributes is used to show the idea. We select attri-
butes from all types of data used including fuzzy numerical (e.g.
2hPG, BMI, serum uric acid, Albumin, and total cholesterol), non-
fuzzy ordinal (e.g. vision, ketones), non-fuzzy semantic (e.g. kidney
disease, liver disease), and non-fuzzy textual (e.g. residence).The
fuzzy EER model for the selected attributes is shown in Fig. 3. In
this figure, all fuzzy features are denoted using FUZZY keyword
and star figure.
Primary keys are black stars with underline text. Fuzzy entities
are dashed-outline rectangles, fuzzy relationships are dashed-
outline diamonds, and l is the fuzzy membership degree. We pro-
pose to extend all attributes with fuzzy types to be modeled in sep-
arate entities because these attributes are compound and
multivalued. For example, Fuzzy Age attribute is modeled as
F_Age entity, which has an attribute to store the membership value
of each fuzzy value, an attribute to store its crisp value, and an
attribute to store its descriptive linguistic label, as shown in
Fig. 4, where Age feature is modeled as 3 fuzzy terms and 2 addi-
tional attributes. This modeling decision eases the mapping of
fuzzy EER model to its equivalent FO elements. As a result, the
fuzzy EER model in Fig. 3 is extended to Fig. 5, which is more
detailed.
3.3. CBRDiabOnto implementation (TBOX) step
The CB fuzzy EER model provides the ontology elements, such
as concepts, properties, and axioms. However, to build a high-
quality fuzzy ontology, which correctly describes a domain, a for-
mal methodology must be followed [8]. Because FO is an extension
to crisp ontology, we follow the proposed ontology engineering
methodology by El-Sappagh et al. [22]. However, for space restric-
tions, its steps are not detailed here. According to proposed case
base EER model in Fig. 5, the translation of this model into an
OWL 2 FO is accurate. There are many research studies of rules
for EER-to-fuzzy ontology transformation, and some of the complete
rule sets have been proposed in Refs. [64,46]. Table 1 provides a
small list of their correspondences.
The creation of fuzzy OWL ontology includes two steps. First,
the creation of ontology structure (TBOX), which is discussed in
this section. Second, the creation of ontology instances (ABOX),
1028 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040

Fig. 1. The fuzzy ontology construction framework.

Fig. 2. The numerical attributes fuzzification.

Fig. 3. The case base fuzzy EER model.

Fig. 4. An example of fuzzy datatype.

which will be discussed in the next section. The whole ontology
has been created in Protégé 4.3 and its Fuzzy OWL plugin5 for
5
http://nmis.isti.cnr.it/~straccia/software/FuzzyOWL/index.html.
implementing the resulting fuzzy CB OWL 2 ontology. All of the fol-
lowing coding based on the syntax of Bobillo and Straccia [9].
For each fuzzy value, such as young_age, we define a fuzzy data
type in protégé. Fuzzy OWL plugin has four fuzzy datatypes (e.g.
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040 1029

Fig. 5. The extended case base fuzzy EER model.

Table 1
The correspondences between structures of fuzzy EER models and fuzzy OWL ontologies.

EER Model element Fuzzy OWL ontology element EER Model element Fuzzy OWL ontology element
Crisp entity Crisp class identifier Attribute Data property identifier
Fuzzy Entity Fuzzy class identifier Fuzzy attribute Fuzzy data property identifier
Relationship Class identifier Role Object property identifier
Cardinality The (minimum and maximum) cardinality axioms Instance of a fuzzy entity Individual identifier
Aggregation of m entities m pairs of fuzzy object property identifies Categorization Union of
...

Left (k1, k2, a, b), Right (k1, k2, a, b), Trapezoidal (k1, k2, a, b, c, d), and
Triangular (k1, k2, a, b, c)) [26]. According to the fuzzification in
Fig. 2, we select from these fuzzy data types. The plugin allows
the definition of fuzzy modifiers for classes, relationships and fuzzy
datatypes (i.e., mod(C), mod(R), and mod(d), where C is a concept, R
is a relation, and d is a fuzzy datatype) [26]. Moreover, fuzzy rela-
tionships and fuzzy axioms are implemented. All the ontology ele-
ments are extracted from its equivalent EER fuzzy model. A small
sample of the implementation of each identified element is done
as follows.
3.3.1. Fuzzy data types and fuzzy data properties
Three types of things are defined for each of the numerical fea-
tures. First, an abstract role for the numerical feature as a whole.
Second, a fuzzy data type for each linguistic term. Third, a fuzzy
concrete role (i.e., object property) for each linguistic term. Consid-
ering HbA1c lab test values, let’s its range to be [0.0, 8.0] and its
linguistic terms are lowA1c (LS (5.7, 6.05), normalA1c (Tri (5.7,
6.05, 6.4)), and highA1c (RS (6.05, 6.4)). First, we create an abstract
role named HbA1c. Second, a fuzzy data type has been created for
each of these fuzzy terms. As shown in Fig. 6, a protégé plugin

Fig. 6. An example of fuzzy data type definition.

1030 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
has been utilized to create the datatype lowA1c and then annotated
it as a fuzzy datatype. This process is repeated for all linguistic
terms in each fuzzy variable in our CB fuzzy ontology. Third, a fuzzy
concrete role has been defined for all of its linguistic values. Previ-
ously defined fuzzy datatypes are used as the ranges for these roles.
Continuing with HbA1c, we defined three fuzzy concrete roles
hasLowA1c, hasNormalA1c, and hasHighA1c. For example, the hasLo-
wA1c has been modeled as hasLowA1c (HbA1c, lowA1c) where
HbA1c is a crisp concept and lowA1c is a fuzzy data type.
3.3.2. Fuzzy modifiers, fuzzy modified data types, and fuzzy modified
roles
Modifiers improve the ontology expressiveness and its semantic
queries. Fuzzy modifiers change degrees of membership of fuzzy
data types. A fuzzy modifier is a function f mod : ½0; 1 ! ½0; 1, which
can be L (c), Tri (a, b, c) [26]. Fuzzy modifiers allow us to define an
axiom like 9Has Age:ðv eryÞyoung, and if we have an object property
Has_Disease (Patient, Disease), we can define very (Has_Disease) or
recently (Has_Disease). Domain expert helped to define modifies
values as very, slightly, somewhat, etc. In our work, these modifiers
can modify the following (according to fuzzy ontology syntax of
Bobillo and Straccia [9]:
1. data type: for instance, the data type veryLowA1c is a modified
version of lowA1c:
<fuzzyOwl2 fuzzyType=‘‘datatype”>
<Datatype type=‘‘modified” modifier=‘‘very”
base=‘‘lowA1c”/>
</fuzzyOwl2>
2. fuzzy concrete role as shown next.
3.3.3. Fuzzy modified data type properties
Two fuzzy data properties have been defined: has-Disease (to
define the severity of a disease) and lived-In (to model the length
of stay for a person in rural or urban). They are modeled as fuzzy
modified roles. For example, the has-Disease role can be modified
by very modifier in a new role very-has-Disease as:
<fuzzyOwl2 fuzzyType=‘‘role”>
<Role type=‘‘modified” modifier=‘‘very”
base=‘‘has-Disease”/>
</fuzzyOwl2>
3.3.4. CBRDiabOnto fuzzy logic
We select Zadeh fuzzy logic for CBRDiabOnto. This is imple-
mented as ontology level annotation:
<fuzzyOwl2 fuzzyType=‘‘ontology”>
<FuzzyLogic logic=‘‘zadeh”/>
</fuzzyOwl2>
Fig. 7 shows a snapshot of the created OWL 2 fuzzy CB ontology
based on CBROnto standard CB ontology of JCOLIBRI 2 CBR frame-
work [56]. The same approach has been followed by Heras et al.
[37] for creating a crisp CB ontology. The CBROnto ontology works
as the top-level organizing elements. This format facilitates the
usage of JCOLIBRI 2 API, a complete CBR API, for case retrieval,
adaptation, revision, and retention. CBROnto provides the top-
level abstraction of several disjoint concepts of CBR. These con-
cepts include the primitive concept CBRCASE, which represents
the list of cases instances in the knowledge base, CBR_INDEX, which
represent the list of case feature concepts, and CASE_COMPONENT,
which subsumes the most abstract components of a case including
the disjoint concepts CBR_DESCRIPTION and CBR_SOLUTION. More-
over, CBROnto provides a set of relationships to link the previously
mentioned concepts, such as Has_Description and Has_Solution. The
following is a very short list of the ontology axioms, where d are
user defined fuzzy data types for linguistic terms for every fuzzy
variable (cf. in El-Sappagh et al. [26], and the instance is the Con-
ceptID of an SCT concept:
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040 1031

Fig. 7. The proposed case-base fuzzy OWL 2 ontology.

The resulting ontology contains 62 classes, 58 (fuzzy) object
properties, 138 fuzzy datatype properties, 105 fuzzy datatypes,
1316 axioms, 761 logical axioms, and 2640 concept instances for
the 60 patient cases. Each object property and datatype property
have an instance for every individual case. Moreover, the fuzzy
roles are not supported in Fuzzy OWL plugin. As a result, we cannot
implement the fuzzy object properties Has_Disease, which repre-
sents the level of criticality of a disease or its number of years,
and Lived_in, which describes the length of time a patient has lived
in rural or urban. Modeling of aggregation combines each case with
its components. We have created an object property Has_Part, and
its inverse Belongs_To, which link all parts of a case to the case
description concept DiabetesDiagnosisCase, as shown in Fig. 8. We
have created some axioms to assure that each case has one concept
from each component, as shown in Fig. 9.
Rules can be added to the resulting ontology using SWRL to
enforce and infer some logics, such as the relation between con-
cepts like Patient_Case (?x) ^ Has_Gender(?x, ‘‘Male”) ! : Has_Diag-
1032 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040

Fig. 8. The aggregation modeling in OWL 2.

Fig. 9. Case axioms for collecting one instance from each component.

noses (?x, ‘‘Gestational_Diabetic”). These rules can help to infer Table 2

missing features’ values of the new query problem and implement Transformation from instances of fuzzy EER models to fuzzy OWL ontologies.
DM diagnosis clinical practice rules. EER instance Fuzzy OWL ontology instance
The next step is to populate the resulting ontology structure
Fuzzy object instance Fuzzy individual identifier
with instances from the EER’s database. In other words, we map Attribute value Data property assertion
data object instances to fuzzy OWL ontology individuals. This step Relationship Object property assertion
will be discussed next. ...
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
Fig. 10. A fragment of the instances information in the case-base fuzzy EER model in Fig. 5.
Fig. 11. The mapping between elements of fuzzy data and fuzzy ontology.
3.4. CBRDiabOnto instantiation (ABOX) step
There are many options for ontology population. The ontology
instances can be stored in a database, and ontology-based data
access algorithm is used to semantically access the instances data-
base [32]. This option supports the scaling of an ontology because
there is no limit to the number of instances stored in a database,
but its reasoning power will suffer [33]. The other option is to store
both of the ontology structure and instances in the same structure,
and this is our choice in this paper. We have created a fuzzy data-
base for the proposed fuzzy EER model and filled it with 60 cases of
diabetic patients. These data have been collected from the EHR of
60 patients in Mansura University Hospitals, Mansura, Egypt. This
data are used to populate CBRDiabOnto. A set of transformation
rules from fuzzy database instances to FO instances is shown in
Table 2.
To illustrate the transformation process, we give an instantia-
tion of the fuzzy EER model in Fig. 5 with data shown in Fig. 10,
where each value of numerical features has been fuzzified accord-
ing to the selected membership function. Symbols of fuzzy EER
model are transformed to identifiers of fuzzy OWL ontology. All
instances for CBROnto ontology are omitted from the discussion
to simplify the process. The fuzzy values of attributes in Fig. 3 have
been modeled in separate entities as shown in Fig. 5, and they are
filled with data in the EER’s corresponding database model as
shown in Fig. 11. The EHR database (Fig. 11a) is converted to the
fuzzy database (Fig. 11b), and the resulting data are used to popu-
late the ontology (Fig. 11c). Moreover, FO has been built to support
the same structure in Fig. 5.
Fig. 12 shows the resulting ontology structure of the CB fuzzy
EER structure fragment in Fig. 5. The syntax used is OWL 2
1033
functional-style syntax.6 Fig. 13 shows the instantiation of the CB
fuzzy EER instances fragment modeled in Fig. 10, where a case is rep-
resented as a set of ontology individuals and relations (i.e., proper-
ties). Some reasoners, such as Pellet, support rule reasoning, and
inference. As a result, we may implement some rules to assert some
Pn
i¼1 li ¼ 1 for
axioms, such as for each fuzzy attribute value:
n = number of fuzzy values and li = a membership value. Although
there exist protégé plugins for FO population, such as FEER2FOnto
[19], they are not fully compatible with our work. As a result, we
can use an API for ontology programming, such as OWL API 3, Jena,
protégé API, or Pellet API. We select OWL API 37 to implement our FO
population algorithm.
Our resulting ontology combines many datatypes and many
reasoning (i.e., fuzzy, CBR, and ontology reasoning). As a result, it
supports many fuzzy semantic retrieval and query algorithms
[51]. Firstly, fuzzy labels are used to extend query case description
and to determine the fuzzy similarity between vague terms in
query evaluation. A domain expert assigns the semantic similarity
between fuzzy labels, and similarity matrices are used to measure
it. Secondly, the list of fuzzy membership values assigned to every
linguistic value enhances the fuzzy similarity algorithms accuracy.
Thirdly, the existence of the original crisp value allows the combi-
nation of fuzzy and crisp similarity measures. Fourthly, fuzzy
hedges can be defined, which enrich the query description and
similarity calculation. Finally, the existence of semantic values,
such as patient diseases, which are connected to other ontologies,
enhance the semantic retrieval of similar cases. Moreover, the
6
https://www.w3.org/TR/owl2-syntax/.
7
http://owlapi.sourceforge.net/.
1034 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040

Fig. 12. A part of the fuzzy OWL ontology structure for Fig. 5.

Fig. 13. A part of the fuzzy OWL ontology instance for Fig. 10.

resulting ontology allows dealing with abstract and incomplete
problem descriptions [53]. Storage preservation is achieved since
many cases can refer to the same feature description objects. For
example, if we have n cases C1,. . ., Cn, which have Age = 45, then
we create one Age object and make all of the n cases point to that
Age object. Cases can have relationships with other cases such as
Ci v Cj.
4. Querying the CB fuzzy ontology
The main purpose of the ontology is to use it as a knowledge
base in a CDSS. This section discusses the main step in CBR system
(i.e., case retrieval) based on querying the resulting ontology.
Fig. 14 proposes a framework for a fuzzy semantic query of the
CB fuzzy ontology. The problem is formatted as a query-case with-
out a solution as suggested in Fig. 15 where physicians are able to
use GUI to enter the patient characteristics. The query case is a vec-
tor V = <attributei = valuei>, for i is the number of features, i.e.,
<Age = 20, BMI = 40, Residence = ‘‘Rural”, Fatigue = ‘‘++”, Gen-
der = ‘‘Male”. . .>. There is no semantic at this level. This vector will
be passed to the Query Parser (Fig. 14); his module has two func-
tions namely fuzzification and encoding based on SCT [27]. Numer-
ical features in the vector are fuzzified using the same membership
functions that are used to fuzzify all cases in the knowledge base
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040 1035

Fig. 14. The fuzzy semantic query formulation and execution engine.

Fig. 15. A fuzzy query entry form.

(see Fig. 2). The structure of resulting data is shown in Fig. 11b.
Semantic features in the resulting fuzzy vector will be encoded using
the domain background-knowledge ontology to support semantic
retrieval [36]. The other features remain the same.
We utilize our previously proposed SCT OWL 2 ontology as the
background knowledge for measuring semantic similarity between
medical concepts [23]. According to CBRDiabOnto vocabulary,
Ontological Query Transformer transform the query case vector into
a semantic query containing OWL 2 individuals and property
instances, as shown in Fig. 11c. Fig. 16 shows an example using
OWL 2 Functional-Style Syntax for the previous query vector V.
the case retrieval is achieved by the Case Retrieval Module based
on ontology reasoners and SCT.
There are many options for querying the CB fuzzy ontology. The
query and the OWL ontology may be serialized into RDF form, and
SPARQL8 language is used to form SQL-like query over the ontology
[42]. However, SPARQL has an incomplete understanding of OWL
semantics because it mixes the syntax of the language with its asser-
tions [50]. Moreover, there is no canonical RDF serialization of some
OWL constructs, and different tools can serialize the same OWL con-
structs in slightly different ways.9 The f-SPARQL extends the capabil-
ities of SPARQL. However, it has no standard and requires an
extension for OWL 2 [16]. Rules reasoning and querying can be
achieved using SWRL and Semantic Query-enhanced Web Rule Lan-
8
http://www.w3.org/TR/sparql11-query/.
9
http://www.w3.org/TR/owl2-mapping-to-rdf/.
1036 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
Fig. 16. A portion of the representation of a query case.
Fig. 17. A part of a semantic query over the case base for Fig. 16.
guage (SQWRL) respectively [50]. SQWRL has built in libraries for
mathematical and logical calculations, which facilitates the imple-
mentation of the retrieval algorithm. Moreover, SQWRL free APIs
facilitates the implementation of JAVA applications for case retrieval
over ontology.10
The SPARQL-DL language can directly query the OWL 2 ontol-
ogy, and this is the optimum choice because it fully utilizes the
OWL 2 semantic capabilities [58]. It is a subset of SPARQL, which
corresponds to conjunctive queries in DLs, and it uses SPARQL syn-
tax. The SPARQL-DL query engine is settled on top of the OWL API
3. We use a combination of SPARQL-DL JAVA APIs11 and SQWRL
JAVA APIs to build our fuzzy semantic query execution engine, as
shown in Fig. 14. The Has_Part, and Belongs_To object properties,
which implemented aggregation and Has_ID data property, are used
to ensure that all selected concepts are linked to one case. We have
used JAVA and OWL API 3 to implement the Case Retrieval Module in
Fig. 14.
The formulated semantic query is passed to the Pellet rea-
soner,12 which supports SPARQL-DL, for the crisp part of the ontol-
ogy. The query is passed to fuzzyDL reasoner for fuzzy reasoning.
Reasoners allow inference of implicit knowledge over explicit one.
The format of the query is in SPARQL-DL and using SWRL built-ins,
which return cases from the fuzzy CB ontology. In its simplest form,
Fig. 17 shows a fragment of a conjunctive query for the semantic
search by suing SPARQL-DL syntax. This query finds a case with an
exact match from CB ontology for the query case in Fig. 16. In the case
10
http://protege.cim3.net/cgi-bin/wiki.pl?SQWRLQueryAPI.
11
http://www.derivo.de/en/resources/sparql-dl-api.html.
12
http://clarkparsia.com/pellet/.
of exact matching found, the querying returns the diagnosis of
matched case.
In the case of no exact match found, the whole cases are
returned to compute their similarities with the query case;
because semantic query engines do not support loops and complex
mathematical computations. This process is handled in the JAVA
code. The similarity between query case and retrieved cases is cal-
culated using the k-NN algorithm, which combines local similarity
function according to feature type (i.e., fuzzy, semantic, ordinal,
or text), and global similarity function, such as weighted Euclidean
distance function. For space restriction, we will propose a novel
fuzzy semantic similarity measure in a future work, which is suit-
able to our proposed ontology. Our CB ontology includes a datatype
property named Has_Weight to assign a weight to each feature. El-
Sappagh et al. [24] proposed the optimum weight vector in another
work. The solution of the most similar case is retrieved and used
directly or entered into the rest of CBR cycle including case adap-
tation, revision, and retention.
Following a CBR methodology, the proposed ontology allows
the CBR system to store the domain cases as concepts in the fuzzy
CB ontology, and semantically and imprecisely retrieve the most
similar cases. In DM diagnosis, patients always describe their prob-
lems by vague terms. Physicians depend on their experience,
which is vague in most cases. The patient may have some lab test
numerical results. Some feature values are concept-IDs from SCT
ontology, and some features are text. Therefore, ontology must
support all these types of data. Our proposed ontology structure
is the richest ontology structure, which maximizes the flexibility
in describing patient problem by a domain expert. In Fig. 15, the
physician can enter numerical features as numbers or as linguistic
labels. Ordinal features are selected from List Boxes. For example,
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040 1037

Fig. 18. A part of a fuzzy semantic query.

the list box of urination frequency feature has the values (Nor- some of the most common pitfalls occurring in the development of
mal = 3–5 times urination per day, + = 6–8 times urination per ontologies. We run OOPS! on the CBRDiabOnto ontology and ensure
day, ++ = 9–10 times urination per day, and +++ = More than 10 that the ontology is free of such pitfalls.
times urination per day). Semantic features are selected from our
SCT ontology, and semantic similarity is calculated [36]. In the case 5.2. Comparison against a gold standard
retrieval, linguistic labels are compared to similarity matrices pro-
posed by our domain expert. Numerical values are fuzzified, and There is no gold standard ontology to measure its similarity
the similarity is calculated using a fuzzy similarity algorithm. The with our ontology. Moreover, if a gold standard exists, then there
text features similarity is computed using exact matching. The ordi- will be no need to create other ontologies. Ontology Design Pat-
nal features similarity is calculated using similarity matrices pro- terns (ODPs) can be considered as a gold standard. However, ontol-
posed by our domain expert. ogy can follow ODPs but does not satisfy the needed functionality.
The crisp numerical features similarity can be calculated using a
distance function (i.e. Euclidean distance). If the query case is rep- 5.3. User-based or criteria-based evaluation
resented in linguistic terms, our ontology supports the direct fuzzy
retrieval of cases. For an exact match, an easy to formed query can A comparison with existing ontologies in the same domain is
be generated; Fig. 18 shows a part of the exact matching linguistic needed; however, there is no (fuzzy) CB ontology in the medical
query, which contains <. . . Age=Young, BMI=OverWeight. . .>, where domain to compare with. Alexopoulos et al. [3] proposed a fuzzy
no numerical calculations are required; it is a variation of the query CB ontology for electricity market CBR system. On the other hand,
represented in Fig. 17 where features are searched using only there Heras et al. [37] proposed a crisp CB ontology for argumentation,
fuzzy terms. The proposed fuzzy CB ontology structure, the popu- Guo et al. [35] for mould design, and Zhukova et al. [65] for
lated knowledge base, and the semantic querying capability will be resource management. Several quantification criteria for ontology
used to build knowledge intensive fuzzy-CBR system for DM diag-
nosis. We will use the most popular CBR framework (i.e., JCOLIBRI Table 3
2), which provides OntoBridge JAVA APIs13 to access ontologies in The ontology evaluation quality metrics.
JAVA. We will alter this APIs to work with fuzzy ontologies and fuzzy
Measure Ontology
reasoners. These aspects will be considered in future works.
Criteria Metrics The ArgCBROnto
proposed
5. Evaluation ontology
Complexity An average number of paths 5 3
In this section, we evaluate the resulting fuzzy CB ontology to reach a class from the
regarding its syntax, semantics, and content coverage. The evalua- root.
Average number of 1.4 1.3
tion process assesses the ontologies concerning a set of desirable
semantic relations (object
qualities or ontology requirements including conciseness, correct- properties) per class.
ness, intelligibility, and adaptability. The creation of an accepted Abstraction The average depth of the 2 2
ontology is an evolutionary problem, i.e., the ontology must be ontology.
used, criticized, and updated by researchers over a long time. There Cohesion An average number of 63 27
connected components
are no globally accepted evaluation mechanisms [8]. According to (classes).
Brewster et al. [11], precision and recall are not suitable for the Conceptualization Semantic Richness: Ratio of 58/(58 38/(38
evaluation because they depend on a comparison between evalu- the total number of + 59) + 23) = 0.62
ated ontology and a standard one. Most of the time, there are no semantic relations assigned = 0.495
to classes, divided by the
standard ontologies. Different evaluation methodologies are
total number of ontology
needed. Quality evaluation methods of ontologies are classified relations (object properties
as in Ref. [64]. We follow this method for evaluation. and subsumption relations).
Attribute Richness: Ratio of 138/ 24/
the total number of 62 = 2.26 26 = 0.92
5.1. Consistency checking
attributes (data properties
describing ontology
This ontology is serialized in the OWL 2 format with the protégé classes), divided by the total
4.3 tool. It contains 63 (fuzzy) classes, 54 (fuzzy) object properties, number of ontology classes.
Inheritance Richness: 5.0 2.875
138 (fuzzy) datatype properties, 105 fuzzy datatypes, and 60 real
Average number of sub-
cases. Consistency checking describes the syntactic-level evalua- classes per class.
tion. The Pellet, fuzzyDL 1.1, HermiT 1.3.6, and FaCT++ reasoners Completeness For Precision and Recall, Not Not
have been used with the Protégé editor to confirm that CBRDia- there are no standard Applicable Applicable
bOnto is consistent and free of errors. They do not reveal any dis- (fuzzy) case base ontologies
to compare our ontology
crepancies regarding this version of the ontology. Moreover, the
with it.
Ontology Pitfall Scanner! (OOPS!)14 online tool can help to detect Comprehension Documentation of the 2.04% 0.0%
properties
13 Documentation of the 88.71% 0.0%
http://gaia.fdi.ucm.es/research/colibri/jcolibri.
14 classes
http://oops.linkeddata.es/.
1038 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040

Fig. 19. A snapshot of our ontology evaluation execution.

X
quality have been proposed [21,34,62]. We depend on the criteria D¼ wf  dðf Þ ð2Þ
of Djedidi and Aufaure [21], where each criterion can be measured f
by metrics. These criteria include complexity, cohesion, conceptu-
where wf is the weight assigned to a concept according to its repe-
alization, abstraction, completeness, and comprehension [21].
tition, and D defines how possible to map a set of user needs to
From the literature, ArgCBROnto is a complete ontology, and the
queries expressed in an ontology. Our ontology has ontology fit
other papers have not proposed ontologies. Table 3 represents a
D ¼ 100, and the ontology adheres to the most critical desiderata
comparison between ArgCBROnto and our ontology regarding
characteristic of Cimino [17] (i.e., content coverage). Fernández
these metrics. For calculating metrics, we depended on the equa-
et al. [29] suggested another data driven evaluation measure. This
tions of Zhang et al. [64]. The ontology parameters of the metrics
measure included the number of classes, properties, axioms, and
are calculated using the protégé 4.3 evaluation plugin (i.e., Ontol-
individuals. Protégé ontology evaluation plugin calculates these
ogy Evaluation.15) Fig. 19 is a snapshot of the evaluation process
metrics and others including naming conventions, class hierarchy,
of the proposed ontology.
object properties hierarchy, data type properties hierarchy, disjoint-
ness restrictions, see Fig. 19. The application of this measure on our
5.4. Lexical, vocabulary, or data level evaluation
ontology is summarized in Table 4, where CBRDiabOnto over-
weights the other ontologies. Moreover, our ontology structure is
Coverage is the completeness of terms or concepts to represent
formally derived from a fuzzy EER model of the case base, which
a domain [62]. This step can be achieved by assuring that the
makes it more accurate.
ontology contains concepts and relations equal to those in the
Finally, the domain experts have evaluated the proposed ontol-
domain, and ontology instances equal to instances in the domain
ogy content regarding the clarity and conciseness.
[11]. Our ontology has been populated with 60 cases, and the
new case can be added. As our ontology has been derived from
– Clarity: Gruber [34] stated three requirements for clarity includ-
the DM diagnosis EER model, all medical terms used in this model
ing formal definition of classes, ontology documentation, and
exist in the proposed ontology. Moreover, the domain experts col-
the using of classes according to the requirements. Since all
lect 328 medical terms from existing DM diagnosis CPGs including
terms are extracted from the EHR databases and CPGs, formal
Canadian Diabetes Guideline16 and others. The expert tests the cov-
definitions are available for all of the terms. CBRDiabOnto is
erage of CBRDiabOnto for all of these terms. The CBRDiabOnto has
documented using a natural language. In addition, the domain
100% concept coverage for all medical classes and relations required
experts define the meaning of all terms.
to describe diabetic patient cases. All needed concepts and relations
– Conciseness: the ontology should not include unnecessary con-
to describe diabetics have been verified. Some terms exist as features
cepts or redundancies. This aspect has been carefully handled
in the EER model as ‘‘HbA1c” term, and others are values, which have
during the ontology development and validation.
been populated in the 60 cases, such as ‘‘pre-diabetic” term. In addi-
tion, the CBR concepts and relations required to define the case base
5.5. Vagueness evaluation
and individual cases have been verified according to the existing CBR
API standard (i.e., JCOLIBRI 2). We used the ontology fit measure pro-
Alexopoulos and Mylonas [2] proposed a set of metrics to eval-
posed by Raimond and Sandler [55]. For a query feature f, they
uate the quality of ontology vagueness. These metrics include:
defined d as in Eq. (1):

1 f is expressible in the ontology – Vagueness Spread (VS): In a fuzzy ontology, the concepts, rela-
dðf Þ ¼ ð1Þ tions, attributes, and data types elements may be of a vague
0 Otherwise
type. VS measures the extent of vagueness representation in
moreover, the whole ontology concept coverage is calculated using the ontology, and it provides an indicator of the ontology’s
Eq. (2): potential comprehensibility and shareability, i.e., high vague-
ness spread equal less explicitness and shareability. VS is the
15
http://protegewiki.stanford.edu/wiki/Ontology_Evaluation. ratio of the number of vague ontology elements (classes,
16
http://guidelines.diabetes.ca/. relations, and data types) (VOE), divided by the total number
 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040 1039

Table 4
The ontology evaluation measures for three ontologies.

The measure Knowledge coverage measures

No. of No. of No. of Max No. of Documentation of Documentation of Properties with Properties Number of
classes properties axioms Parents the classes the properties domain with range individuals
The Proposed 62 196 1316 31 88.71% 2.04% 98.47% 98.98% 2640
ontology
ArgCBROnto 26 62 446 13 0% 0% 85.48% 77.41% 0
Alexopoulos 10 18 N/A N/A N/A N/A N/A N/A N/A
et al.

of elements (OE), as in Eq. (3). We have C (Classes) = 62, OP will populate the ontology with more cases to enhance the case
(Object properties) = 58, FD (Fuzzy Datatypes) = 105, FDP = retrieval accuracy. We will study semantic retrieval algorithms,
(Fuzzy Datatype Properties) =138, and FOP (Fuzzy Object Prop- and we will study how to improve it. We will study the possibili-
erties) = 2. ties for improving the performance of the proposed case base
ontology such as storing the fuzzy CB ontology in a fuzzy relational
jVOEj FD þ FDP þ FOP
VS ¼ ¼ database and use fuzzy SQL to implement case retrieval.
jOEj C þ OP þ FD þ FDP
105 þ 138 þ 2 Acknowledgments
¼ ¼ 0:675 ð3Þ
62 þ 58 þ 105 þ 138
– Vagueness Explicitness (VE): It is the ratio of the number of The authors would like to thank Professor Farid Badria, Prof. of
vague ontological elements that are explicitly identified Pharmacognosy, Department and Head of Liver Research Lab,
(EVOE), divided by (VOE), as in Eq. (4). The higher the value Mansoura University, Egypt and Professor Hosam Zaghloul, Prof.
of this metric is the better is the ontology. All fuzzy elements at Clinical Pathology Department, Faculty of Medicine, Mansoura
defined in the proposed ontology are explicitly defined, and University, Egypt for their efforts in this work.
fuzzy reasoner (fuzzyDL) can infer other implicit elements at
run time. References

jEVOEj FD þ FDP þ FOP 105 þ 138 þ 2
VE ¼ ¼ ¼ ¼ 1:0
jVOEj FD þ FDP þ FOP 105 þ 138 þ 2
– Vagueness Intensity (VI): It is the level of disagreeing of ontol-
ogy users on the validity of the (potential) instances of the
FO elements. For example, Age concept may be fuzzified as
Young, Middle-Aged, and Old, as discussed in Fig. 2. This fuzzi-
fication may be not acceptable by all ontology users or
domain experts. Alexopoulos and Mylonas [2] argued that
the exact formula for this metric depend on the way one
decides to measure user agreement. Because the proposed
ontology is derived from a domain fuzzy EER model, which
is prepared and approved previously by both domain expert
and CPG, the result is highly acceptable by our domain
expert.
6. Conclusion
In this paper, we proposed a novel fuzzy CB ontology (CBRDia-
bOnto) for DM diagnosis fuzzy KI-CBR system. The resulting ontol-
ogy is enriched with multiple types of data, such as fuzzy, crisp,
text, ordinal, and semantic data. These different types of data facil-
itate the development of fuzzy semantic-case retrieval algorithms
and support queries expression by physicians. This way, we
improved the most critical steps of CBR system (i.e., case represen-
tation and case retrieval). The CBRDiabOnto is the unique CBR
fuzzy ontology in the medical domain and especially in the DM
diagnosis domain. It has been populated with 60 cases instances
from the EHRs data of diabetics. It contains 63 (fuzzy) classes, 54
(fuzzy) object properties, 138 (fuzzy) datatype properties, 105
fuzzy datatypes, and 60 real cases. We implemented a user inter-
face in JAVA to collect the description of the patient query case.
We implemented a framework to fuzzify and code the case
description. The resulting query case is formatted into a semantic
conjunctive query. This query is executed using some Java APIs
(i.e. FuzzyDL APIs and OWL API 3) on the fuzzy CB ontology. The
similarity of the retrieved cases from the ontology can be calcu-
lated based on local-global similarity algorithm. In the future, we
[1] M. Abdul, A. Muhammad, N. Mustapha, S. Muhammad, N. Ahmad, Database
ð4Þ
workload management through CBR and fuzzy based characterization, Appl.
Soft Comput. 22 (2014) 605–621.
[2] P. Alexopoulos, P. Mylonas, Towards vagueness-oriented quality assessment of
ontologies, Artif. Intell. Methods Appl. 8445 (2014) 448–453.
[3] P. Alexopoulos, M. Wallace, K. Kafentzis, D. Askounis, Utilizing imprecise
knowledge in ontology-based CBR systems by means of fuzzy algebra, Int. J.
Fuzzy Syst. 12 (1) (2010).
[4] P. Alexopoulos, M. Wallace, K. Kafentzis, D. Askounis, IKARUS-Onto: a
methodology to develop fuzzy ontologies from crisp ones, Knowl. Inf. Syst.
32 (3) (2012) 667–695.
[5] F. Ali, K. Kwak, Y. Kim, Opinion mining based on fuzzy domain ontology and
Support Vector Machine: a proposal to automate online review classification,
Appl. Soft Comput. 47 (2016) 235–250.
[6] K. Amailef, J. Lu, Ontology-supported case-based reasoning approach for
intelligent m-Government emergency response services, Decis. Support Syst.
55 (1) (2013) 79–97.
[7] American Diabetes Association (ADA), 2017. http://www.diabetes.org/
(Accessed 14 Mars 2017).
[8] T. Bright, E. Furuya, G. Kuperman, J. Cimino, S. Bakken, Development and
evaluation of an ontology for guiding appropriate antibiotic prescribing, J.
Biomed. Inform. 45 (1) (2012) 120–128.
[9] F. Bobillo, U. Straccia, The fuzzy ontology reasoner fuzzyDL, Knowl.-Based Syst.
95 (2016) 12–34.
[10] M. Branden, N. Wiratunga, D. Burton, S. Craw, Integrating case-based
reasoning with an electronic patient record system, Artif. Intell. Med. 51 (2)
(2011) 117–123.
[11] C. Brewster, H. Alani, S. Dasmahapatra, Y. Wilks, Data driven ontology
evaluation, in: proceedings of the International Conference on Language
Resources and Evaluation, Lisbon, Portugal, 2004, pp. 164–168.
[12] P. Cadena, L. Garrido, Fuzzy case-based reasoning for managing strategic and
tactical reasoning in StarCraft, Adv. Artif. Intell. 7094 (2011) 113–124.
[13] Canadian Diabetes Association (CDA), 2017. https://www.diabetes.ca/
(Accessed 14 Mars 2017).
[14] J. Chang, M. Lee, T. Wang, Integrating a semantic-based retrieval agent into
case-based reasoning systems: a case study of an online bookstore, Comput.
Ind. 78 (2016) 29–42.
[15] J. Chen, S. Su, C. Chang, Diabetes care decision support system, in: IEEE 2nd
International Conference on Industrial and Information Systems (IIS), 1, 2010,
pp. 323–326.
[16] J. Cheng, Z. Ma, L. Yan, F-SPARQL: a flexible extension of SPARQL, Database
Expert Syst. Appl. 6261 (2010) 487–494.
[17] J. Cimino, Desiderata for controlled medical vocabularies in the twenty-first
century, Methods Inf. Med. 37 (4–5) (1998) 394–403.
[18] V. Cross, Fuzzy ontologies: the state of the art, in: IEEE Conference on Norbert
Wiener in the 21st Century (21CW), 2014, pp. 1–8.
[19] V. Cross, M. Kandasamy, Fuzzy concept lattice construction: a basis for
building fuzzy ontologies, in: IEEE International Conference on Fuzzy Systems
(FUZZ), 2011, pp. 1743–1750.
1040 S. El-Sappagh, M. Elmogy / Engineering Science and Technology, an International Journal 20 (2017) 1025–1040
[20] B. Diaz-Agudo, P. Gonzalez-Calero, An architecture for knowledge intensive
CBR systems, Adv. Case-Based Reasoning 1898 (2000) 37–48.
[21] R. Djedidi, M. Aufaure, ONTO-EVOAL an Ontology Evolution Approach Guided
by Pattern Modeling and Quality Evaluation, in: Foundations of Information
and Knowledge Systems, Springer, Berlin Heidelberg, 2010, pp. 286–305.
[22] Shaker El-Sappagh, S. El-Masri, M. Elmogy, R. Riad, B. Saddik, An ontological
case base engineering methodology for diabetes management, J. Med. Syst. 38
(8) (2014) 1–14.
[23] Shaker El-Sappagh, M. Elmogy, S. El-Masri, A. Riad, A diabetes diagnostic
domain ontology for CBR system from the conceptual model of SNOMED CT,
in: Proceedings of the Second International Conference on Engineering and
Technology (ICET 2014), Cairo, Egypt, 2014, pp. 1–7.
[24] Shaker El-Sappagh, M. Elmogy, A. Riad, H. Zaghlol, F. Badria, EHR data
preparation for case based reasoning construction, in: Proceedings of the 2nd
International Conference on Advanced Machine Learning Technologies and
Applications (AMLTA14), 488, 2014, pp. 483–497.
[25] Shaker El-Sappagh, M. Elmogy, A. Riad, A CBR system for diabetes mellitus
diagnosis: case-base standard data model, Int. J. Med. Eng. Inf. 7 (3) (2015)
191–208.
[26] Shaker El-Sappagh, M. Elmogy, A. Riad, A fuzzy-ontology-oriented case-based
reasoning framework for semantic diabetes diagnosis, Artif. Intell. Med. 65 (3)
(2015) 179–208.
[27] Shaker El-Sappagh, M. Elmogy, An encoding methodology for medical
knowledge using SNOMED CT ontology, J. King Saud Univ. Comput. Inf. Sci.
28 (3) (2016) 311–329.
[28] Shaker El-Sappagh, Farman Ali, DDO: a diabetes mellitus diagnosis ontology,
Appl. Inf., Springer, Berlin Heidelberg 3 (1) (2016) 5–33.
[29] M. Fernández, C. Overbeeke, M. Sabou, E. Motta, What makes a good ontology?
A case study in fine-grained knowledge reuse, Semantic Web 5926 (2009) 61–
75.
[30] M. Gan, X. Dou, R. Jiang, From ontology to semantic similarity: calculation of
ontology-based semantic similarity, Sci. World J. 2013 (2013) 1–11.
[31] C. García-Diéguez, M. Herva, E. Roca, A decision support system based on fuzzy
reasoning and AHP–FPP for the ecodesign of products: application to footwear
as a case study, Appl. Soft Comput. 26 (2015) 224–234.
[32] G. Gottlob, S. Kikot, R. Kontchakov, et al., The price of query rewriting in
ontology-based data access, Artif. Intell. 213 (2014) 42–59.
[33] B. Grau, et al., Towards query formulation and query-driven ontology
extensions in OBDA systems, in: OWL Experiences and Directions Workshop
(OWLED 2013), vol. 1080 of CEUR Workshop Proceedings, 2013.
[34] T. Gruber, Toward principles for the design of ontologies used for knowledge
sharing, Int. J. Hum Comput Stud. 43 (1995) 907–928.
[35] Y. Guo, J. Hu, Y. Peng, A CBR system for injection mould design based on
ontology: a case study, Comput. Aided Des. 44 (6) (2012) 496–508.
[36] S. Harispe, D. Sánchez, et al., A framework for unifying ontology-based
semantic similarity measures: a study in the biomedical domain, J. Biomed.
Inform. 48 (2014) 38–53.
[37] S. Heras, V. Botti, V. Julian, An ontological-based knowledge-representation
formalism for case-based argumentation, Agreement Technol. 8068 (2013)
105–119.
[38] S. Heras, V. Botti, V. Julian, ArgCBROnto: A knowledge representation
formalism for case-based argumentation, Agreement Technol. 8068 (2013)
105–119.
[39] A. Jaya, G. Uma, Role of ontology in case-based reasoning (CBR) for diagnosing
diabetes, J. Inf. Technol. 5 (3) (2009) 17–23.
[40] M. Jha, D. Pakhira, B. Chakraborty, Diabetes detection and care applying CBR
techniques, Int. J. Soft Comput. Eng. 2 (6) (2013) 2231–2307.
[41] A. Khanum, M. Mufti, M. Javed, M. Shafiq, Fuzzy case-based reasoning for facial
expression recognition, Fuzzy Sets Syst. 160 (2009) 231–250.
[42] I. Kollia, B. Glimm, I. Horrocks, SPARQL query answering over OWL ontologies,
in: The Semantic Web: Research and Applications, Springer, Berlin Heidelberg,
2011, pp. 382–396.
[43] C. Lee, M. Wang, G. Acampora, V. Loia, C. Hsu, Ontology-based intelligent fuzzy
agent for diabetes application, IA’09 IEEE Symposium on Intelligent Agents,
IEEE, 2009, pp. 16–22.
[44] C. Lee, M. Wang, C. Hsu, Z. Chen, Type-2 fuzzy set and fuzzy ontology for diet
application, Advances in Type-2 Fuzzy Sets and Systems, vol. 301, 2013, pp.
237–256.
[45] C. Lee, M. Wang, A fuzzy expert system for diabetes decision support
application, IEEE Trans. Syst. Man Cybern. Part B Cybern. 41 (1) (2011) 139–
153.
[46] Y. Lv, Constructing fuzzy ontology based on fuzzy EER model, in: IEEE Seventh
International Conference on Fuzzy Systems and Knowledge Discovery (FSKD
2010), 2010, pp. 1109–1113.
[47] Z. Ma, Mapping fuzzy EER model into fuzzy relational database model, in:
Proceedings of the International Conference on Computational Intelligence
and Software Engineering (CiSE 2009), 2009, pp. 1–4.
[48] C. Marlinga, S. Montanib, I. Bichindaritzc, P. Funkd, Synergistic case-based
reasoning in medical domains, Expert Syst. Appl. 41 (2) (2014) 249–259.
[49] J. Medina, O. Pons, M. Vila, GEFRED: a generalized model of fuzzy relational
databases, Inf. Sci. 76 (1) (1994) 87–109.
[50] M. O’Connor, A. Dos, SQWRL: a query language for OWL, in: Proceedings of
OWL: Experiences and Directions (OWLED), 2009, p. 529.
[51] S. Pal, S. Shiu, Foundations of Soft Case-Based Reasoning, first ed., Wiley-
Interscience, 2004.
[52] G. Park, J. Benedictos, C. Lee, M. Wang, Ontology-based fuzzy-CBR support
system for ship’s collision avoidance, in: Proceedings of the 6th International
Conference on Machine Learning and Cybernetics, 4, 2007, pp. 1845–1850.
[53] I. Péreza, R. Wikstrmb, et al., Linguistic consensus models based on a fuzzy
ontology, Procedia Comput. Sci. 17 (2013) 498–505.
[54] A. Rahimi, S. Liaw, J. Taggart, P. Ray, H. Yu, Validating an ontology-based
algorithm to identify patients with Type 2 diabetes mellitus in electronic
health records, Int. J. Med. Informatics 83 (10) (2014) 768–778.
[55] Y. Raimond, M. Sandler, Evaluation of the music ontology framework, in:
Semantic Web: Research and Applications, 2012, pp. 255–269.
[56] J. Recio-García, B. Díaz-Agudo, P. González-Calero, The COLIBRI platform: tools,
features and working examples, Successful Case-based Reasoning
Applications-2, Vol. 494, 2014, pp. 55–85.
[57] N. Rodríguez, M. Cuéllar, J. Lilius, M. Calvo-Flores, A fuzzy ontology for
semantic modelling and recognition of human behavior, Knowl. Based Syst. 66
(2014) 46–60.
[58] E. Sirin, B. Parsia, SPARQL-DL: SPARQL Query for OWL-DL, in: Proceedings of
OWL: Experiences and Directions (OWLED), 2007, p. 258.
[59] M. Sohn, S. Jeong, H. Lee, Case-based context ontology construction using
fuzzy set theory for personalized service in a smart home environment, Soft.
Comput. 18 (2014) 1715–1728.
[60] G. Stoilos, G. Stamou, J. Pan, Fuzzy extensions of OWL: logical properties and
reduction to fuzzy description logics, Int. J. Approximate Reasoning 51 (2010)
656–679.
[61] H. Yu, S. Liaw, J. Taggart, A. Khorzoughi, Using ontologies to identify patients
with diabetes in electronic health records, in: International Semantic Web
Conference, 2013. pp. 77–80.
[62] J. Yu, J. Thom, A. Tam, Evaluating ontology criteria for requirements in a
geographic travel domain, in: On the Move to Meaningful Internet Systems
2005: Coopis, DOA, and ODBASE, Springer, Berlin Heidelberg, 2005, pp. 1517–
1534.
[63] L. Zadeh, From search engines to question-answering systems the need for
new tools, in: Advances in Web Intelligence, Springer, Berlin Heidelberg, 2003,
pp. 15–17.
[64] F. Zhang, Z. Ma, L. Yan, J. Cheng, Construction of fuzzy OWL ontologies from
fuzzy EER models: a semantics-preserving approach, Fuzzy Sets Syst. 229
(2013) 1–32.
[65] I. Zhukova, M. Kultsova, M. Navrotsky, A. Dvoryankin, Intelligent support of
decision making in human resource management using case-based reasoning
and ontology, Knowl.-Based Software Eng. 466 (2014) 172–184.
[66] A. Zvieli, P. Chen, Entity-relationship modeling and fuzzy databases, in:
Proceedings of IEEE International Conference on Data Engineering, 1986, pp.
320–327.