Content uploaded by Mohammad Elhefny
Author content
All content in this area was uploaded by Mohammad Elhefny on Mar 28, 2018
Content may be subject to copyright.
162 Int. J. Medical Engineering and Informatics, Vol. 9, No. 2, 2017
Copyright © 2017 Inderscience Enterprises Ltd.
Developing a fuzzy OWL ontology for obesity related
cancer domain
Mohammad Abdelrahman Elhefny,
Mohammed Elmogy* and Ahmed A. Elfetouh
Information Systems Department,
Faculty of Computers and Information,
Mansoura University, Egypt
Email: mohhed20@mans.edu.eg
Email: melmogy@mans.edu.eg
Email: elfetouh@gmail.com
Farid A. Badria
Pharmacognosy Department and Liver Research Laboratory,
Faculty of Pharmacy,
Mansoura University, Egypt
Email: faridbadria@gmail.com
Abstract: Obesity is associated with various diseases, particularly
cardiovascular diseases, diabetes type 2, obstructive sleep apnea, certain types
of cancer, osteoarthritis, and asthma. The knowledge of the obesity related
cancer (ORC) domain is highly required to be represented in a structured and
formalised shape. In this paper, we develop an ontology to represent ORC
domain knowledge with its diseases, symptoms, diagnosis, and treatments.
The proposed ontology is based on the Web Ontology Language (OWL 2)
integrated with the fuzzy logic. The fuzzy developed ontology handles the
overlapping concepts, ingesting more concepts, and copes with the linguistic
domain variables, which were not possible using the regular ontologies. It
allows the users to query the fuzzy Dl reasoner for element and answer them
with the fuzzy ontology. By developing the fuzzy ORC ontology, it is expected
to be a good practice for the ontologists and knowledge engineers.
Keywords: fuzzy ontology; obesity related cancer; ORC; Web Ontology
Language 2; OWL2; knowledge representation; disease ontology.
Reference to this paper should be made as follows: Elhefny, M.A.,
Elmogy, M.E and Elfetouh, A.A. (2017) ‘Developing a fuzzy owl ontology for
obesity related cancer domain’, Int. J. Medical Engineering and Informatics,
Vol. 9, No. 2, pp.162–187.
Biographical notes: Mohammad Abdelrahman Elhefny is a Webmaster at
Communications and Information Technology Center, Mansoura University,
Egypt. He received his BSc from Information Systems Department, Faculty of
Computers and Information, Mansoura University in 2001. He is currently
working on his MSc at Information Systems Department, Faculty of
Computers and Information, Mansoura University. His research interests
involve knowledge representation, ontology engineering, Semantic Web, and
bioinformatics engineering.
Developing a fuzzy owl ontology for obesity related cancer domain 163
Mohammed Elmogy is an Associate Professor at Faculty of Computers and
Information, Mansoura University, Egypt. He received his BSc and MSc from
Faculty of Engineering, Mansoura University. He received his PhD from
Informatics Department, MIN Faculty, Hamburg University, Germany in 2010.
He has published more than 80 papers in international journals and
conferences. His current research interests are computer vision, machine
learning, pattern recognition, and biomedical engineering.
Ahmed A. Elfetouh is a Professor at Information Systems Dept., Faculty of
Computers and Information, Mansoura University, Egypt. He received his BSc,
MSc, and PhD from Accounting Information systems, Faculty of Commerce,
Mansoura University and Suez Canal Univ., Egypt. He has published more than
40 papers in international journals and conferences. His recent research
interests are intelligent information systems, decision support systems,
management information systems, and geographic information systems.
Farid A. Badria received his PhD in Pharmacognosy from the University of
Mississippi, USA. He holds a BSc in Pharmaceutical Sciences, and two MSc
from Mansoura University and the University of Minnesota. TWAS-Prize
(2013), World Intellectual Property Organization Gold Medal, Egypt (2011);
State Recognition Outstanding Award in Medicine, Egypt (2001); Outstanding
Arab Scholar, Kuwait (2000); and Khawarazmi International Award, Iran
(2000), are just some of the awards he received. He has 17 patents with granted
final certificates. With over 128 publications, he continues to lead research
projects on developing new therapy for liver disorders, immunity, skin, and
biomarkers for cancer.
1 Introduction
Nowadays, we are facing endless needs for the human’s expertise in all specialised fields,
such as medical/healthcare, education, finance, fault diagnosis, industrial applications,
and business. In addition, we need to take the actual decision at the appropriate time
based on well formalised and specialised knowledge. Therefore, it is critical to represent
the knowledge efficiently using ontologies via integrating the scattered informational
resources. Practically, there are problems while building ontologies like the linguistic
variables, overlapping concepts, and the state of uncertainty that exist in the domain.
From a medical view, the domain of obesity related cancer (ORC) is a critical topic
for research. There is a strong relationship between the obesity and the different types of
cancer. Obesity is one of the leading preventable causes of death worldwide. The fuzzy
ontology of ORC domain can be defined as the representation of knowledge and data
relating to obesity and cancer diseases, risk factors, symptoms, diagnosis, and treatment.
It is taking into account the fuzzy aspects (linguistic variables and uncertainties) that may
be present in this medical domain.
In this paper, the developed ontology will allow the users to query it for element and
get him back the fuzzy ontology for that element by using the fuzzy Dl reasoner1. The
developed ontology was encoded using Protégé 4.32 in OWL 2-DL format and then was
integrated with a pre-developed fuzzification plug-in3. The diseases hierarchy was built
upon disease ontology (DO) (Disease Ontology – Institute for Genome Sciences, 2015).
We preferred to use the DO hierarchy that represents the human disease standard
164 M.A. Elhefny et al.
ontology. It semantically incorporates ailments and therapeutic vocabularies by broad
cross mapping of its terms according to ICD, OMIM, MeSH, NCI’s thesaurus, and
SNOMED.
The rest of this paper is organised as follows. Section 1 introduces the basics of the
ontologies and the standard web ontology language. In Section 2, the basic information of
the medical domain is presented. Section 3 summarises the previous researchers’ work in
medical ontologies and fuzzy sets. The fuzzy owl ontology for obesity-related cancer
domain is proposed in Section 4. Section 5 discusses the experimental results. Finally,
Section 6 presents the conclusion and the future work.
1.1 Basic definitions
The Semantic Web is a web extension to enable individuals to share contents beyond the
limits of the websites and applications (Semantic Web Wiki, 2015). It means to transform
the present web (with unstructured and semi-structured documents) into a ‘web of data’,
and its stack expands on the resource description framework (RDF) (Semantic Web,
2015).
Ontology is the main method to represent, share, and reuse of knowledge on the
Semantic Web. It can be described as a domain conceptualisation into a human
intelligible, machine-clear form involving axioms, attributes, relationships, and entities
(Chen et al., 2012). W3C defined ontologies as formalised vocabularies of terms that
cover a particular domain and are shared by a users’ community. In the ontology, the
definition term is specified by its associations with the other terms (W3C, 2015a). The
domain ontology is a format of an acceptable computer representation of knowledge
about a part of an abstract or a real world (Torshizi et al., 2014).
The fuzzy ontology can be described as an extended domain ontology to overcome
the uncertainty, reasoning, and retrieval problems. The fuzzy ontologies are qualified to
deal with fuzzy knowledge (Semantic Web, 2015; Zhai et al., 2008).
1.2 Medical ontologies
Instead of reinventing the wheel and start from scratch, there are different free and open
ontologies and medical projects that can be effectively reused, such as GALEN4, MeSH5,
SNOM6, Gene Ontology7, Bio-Ontology8, OBO Foundry6, and DOID (Disease Ontology
– Institute for Genome Sciences, 2015).
1.3 OWL: the web ontology language
The OWL is used to describe ontologies. It is based on XML, and can be divided into
three language levels (OWL DL, OWL Lite, and OWL Full) (Chen et al., 2012; OWL,
2015). OWL 2 has three added profiles (OWL 2 EL, OWL 2 QL, and OWL 2 RL) (W3C,
2015b).
The built ontologies by OWL2 are stored as Semantic Web documents and support
adding properties, classes, individuals, and data values. They are mainly exchanged as
RDF documents and can be utilised alongside written information in RDF (W3C, 2015a).
Figure 1 shows an overview of the OWL2 language and the relationships among its
main building blocks. The centred ellipse can be considered as an RDF graph or an
abstract structure. Different particular syntaxes can be used for ontologies exchange, at
Developing a fuzzy owl ontology for obesity related cancer domain 165
the top part. Two semantic specifications define the OWL2 ontologies meanings, at the
bottom part. In their work with OWL 2, the majority of the users need only one syntax
and one semantics (W3C, 2015a).
Figure 1 The structure of the OWL2 (see online version for colours)
Source: W3C (2015a)
2 ORC domain
2.1 Obesity and cancer risk
Recently, the percentage of the overweighted and obese adults and children has
significantly increased. Obesity is a condition in which a human has an abnormally high
and unhealthy proportion of body fat. Obese people are more exposed to cancers as well
as coronary heart disease, stroke, high blood pressure, diabetes, and some other chronic
diseases. The rate of cases ascribed to obesity varied for different types of cancers but
recorded 40% of some cancers, specifically endometrial and esophageal cancers (Obesity
and Cancer Risk – National Cancer Institute, 2015).
BMI is defined as the subject’s weight divided by the square of their height and is
calculated as follows:
2
/BMI m h=
where m and h are the subject’s weight and height, respectively. BMI is usually expressed
in kilograms per square metre, resulting when weight is measured in kilograms and
166 M.A. Elhefny et al.
height in metres. To convert from pounds per square inch multiply by 703 (kg/m2) / (lb/sq
in).
The guidelines of NIH10 considered the 20 years old adults and older with their BMI
values into the defined categories, as shown in Table 1.
Table 1 The Guidelines of BMI
BMI categories BMI
Obese 30.0 and above
Overweight 25.0 to 29.9
Normal 18.5 to 24.9
Underweight Below 18.5
Source: Obesity and Cancer Risk – National Cancer Institute (2015)
Calle and Kaaks (2004) stated that in the USA, about two-thirds of adults were obese or
overweight by the year 2000, and 300 million adults had obesity around the world.
Table 2 Obesity-related cancer
Type of cancer
Relative risk*
with BMI of ≥
30 kg/m2
Relative risk*
with BMI of
25–30 kg/m2
PAF (%) for
US population
PAF (%) for
EU population
Colorectal (men) 2.0 1.5 35.4 27.5
Colorectal (women) 1.5 1.2 20.8 14.2
Female breast
(postmenopausal)
1.5 1.3 22.6 16.7
Endometrial 3.5 2.0 56.8 45.2
Kidney (renal-cell) 2.5 1.5 24.5 31.1
Oesophageal
(adenocarcinoma)
3.0 2.0 52.4 24.7
Pancreatic 1.7 1.3 26.9 19.3
Liver 1.5–4.0 ND ND ND
Gallbladder 2.0 1.5 35.5 27.1
Gastric cardia
(adenocarcinoma)
2.0 1.5 35.5 27.1
Note: *Relative risk estimates are summarised from the literature cited in the main text.
Source: Calle and Kaaks (2004)
Less attention was given to the strong association between the cancer types and the
causing obesity than its cardiovascular effects. In the USA, it was assessed that nearly
20% of all cancer deaths could be credited to overweight and corpulence. There is a
defined relation between the obesity and the high levels of Insulin. Table 2 shows the
relative risk of different BMI ranges with different cancer types throughout a statistical
study made for EU and US populations.
Developing a fuzzy owl ontology for obesity related cancer domain 167
2.2 Tumour markers and reference ranges
Tumour markers are certain types of proteins and or compounds expressed by different
cancer cells of the body. These markers can be detected in blood, urine, stool, tumour
tissue, other tissues, or bodily fluids. They are utilised to help distinguish, analyse, and
deal with a few types of cancer. The raised level of a tumour marker may be a diagnostic
factor of cancer existence, but alone it is not sufficient to diagnose cancer. Therefore,
other invasive examination such as biopsies, are usually combined with measurements of
tumour markers to diagnose cancer (Tumor Markers – National Cancer Institute, 2015).
According to National Cancer Institute (NCI) (Tumor Markers – National Cancer
Institute, 2015), Table 3 summarises the required tumour markers tests for each cancer
type. We focused on the most common cancer types in Mansoura University Hospitals,
Mansoura, Egypt.
Table 3 The cancer types and their related tumour markers
Cancer type Required tumour markers tests
Ovarian CA-125, * HE4, * 5-Protein signature (Ova1)
Colorectal * BRAF mutation V600E, Carcinoembryonic antigen (CEA), KRAS mutation
analysis
Pancreatic CA19-9
Liver AFP
Breast CA15-3/CA27.29, Estrogen receptor (ER)/progesterone receptor
(PR),Carcinoembryonic antigen (CEA), HER2/neu, * Plasminogen activator
inhibitor (PAI-1) and Urokinase plasminogen activator (uPA), * 21-Gene
signature (Oncotype DX), * 70-Gene signature (Mammaprint)
Note: *This tumour marker test is not applied in Egypt.
Association between obesity and tumour markers were under consideration at Mansoura
University Hospitals. They specified (age, gender, BMI, glucose tests [2hPG, FPG,
HbA1C], lipid profile [total cholesterol, TG (triglyceride)]), etc., Table 4 indicates the
reference ranges of tumour markers in cancer patients.
Table 4 Tumour markers reference ranges
CA-125 AFP CEA Kras
0–35 U/ml Low levels present in both
men & non-pregnant women
(0–15 IU/ml); generally results
> 400 are caused by cancer
(Half-life 4–6 days)
<2.5 ng/ml in
non-smokers <5
ng/ml in smokers
Generally, > 100
signifies metastatic
cancer
1 % is the cut-off
level between
non-mutant and
mutant Kras
CA19-9 CA15-3 ER PR HER2/neu
< 37 U/ml is
normal > 120
U/ml is generally
caused by
tumour
<31 U/ml (30% of patients
have an elevated CA 15-3 for
30–90 days after treatment, so
hold up 2–3 months after
beginning new treatment to
check)
1 % staining is the
cut-off point Above
this, it is positive
10 % staining is
the cut-off point
Source: OncoLink – The Web’s First Cancer Resource (2015)
168 M.A. Elhefny et al.
Table 5 and Table 6 list the reference ranges for glucose and cholesterol levels, which are
used by the Egyptian experts to judge the patient condition, respectively. The standard
reference values for these glucose and cholesterol levels can be found in11, 12. For
cholesterol, we found that the USA and some other countries use the same ranges while
Canada and most of the Europe use different ranges. Our Egyptian ranges are closer to
US ranges. The required tests are as follows: FPG (Fasting Plasma Glucose), 2hPG (two-
hour Plasma Glucose), HbA1C (Glycated hemoglobin), AFP (Alpha-fetoprotein), CEA
(Carcinoembryonic antigen), Kras (KRAS mutation analysis), ER (Estrogen receptor),
PR (Progesterone Receptor), T (Triglyceride), and T.Chol (Total Cholesterol).
Table 5 The diabetes reference ranges
Oral glucose tolerance test (mg/dL) FPG (mg/dL) HbA1C
Diabetic 200 or above 126 or above 6.5 or above
Prediabetic 144–199 100–125 5.7–6.4
Normal 139 or below 99 or below About 5.0
Table 6 The lipid profile reference ranges
Triglycerides (mg/dL) 60–160
Total cholesterol (mg/dL) 0–200
3 Related work
In ORC domain, there are some issues with the overlapping concepts/terms, linguistic
variables, and the uncertainty circumstances that exist and need to be addressed and
accommodated while representing its knowledge. Our work focuses on integrating the
fuzzy logic while building the ORC ontology using OWL2 and Protégé to formalise the
ORC domain. It introduces more efficient knowledge semantically representation of the
ORC domain and provides reasoning capabilities. It is useful to physicians, experts or
medical researchers, and computer scientists who are interested in this domain of
knowledge.
13The two sets of population attributable fractions (PAFs) have been computed using
these relative risks.
Parry (2004) presented a fuzzy ontology technique for medical document retrieval. To
enhance any ontology searching tool, he made a mapping between query terms and
individuals of an ontology. In any case, the relative significance of a specific mapping to
an overloaded term might be diverse for various users, and this information is essential
for the reasonable fulfilment of inquiry. For every user or a users’ group, the fuzzy
ontology was used by adding a degree membership value to every ‘overloaded’ term.
Then, from Ontology mediated search, the retrieved documents can give the probable
information request. Parry’s approach addressed the ‘overloaded’ terms (the same terms
occur more than once) not the ‘overlapping’ terms (the similar concepts in meaning that
have different degrees of usage), but it was a starting point to ensure the concept of fuzzy
use of medical ontologies.
Chen et al. (2012) introduced fuzzy rules based anti-diabetic drugs recommendation
system, fuzzy reasoning techniques, and the ontology of anti-diabetic drugs for medicine
Developing a fuzzy owl ontology for obesity related cancer domain 169
recommendation. Their experimental results showed that the drugs selection achieved a
good performance. They used the clinical practice data of the US Association of Clinical
Endocrinologists Medical Guidelines for 20 patients, and according to six attributes/tests.
They used tools like Protégé, OWL DL, Joseki server software, and SPARQL as a query
language. They used the old version of OWL, and they integrated the fuzzy logic into the
reasoning system not in building the ontology itself.
Alfonse et al. (2012) introduced a developing method to build an ontology for liver
cancer using Protégé and OWL-DL format to encode their ontology. Their ontology was
expected to benefit experts or medical researchers who need such knowledge be
semantically represented. The data was acquired from cancer.gov, medicinenet.com, and
cancer.net. They did not make an integration of fuzzy logic in their ontology, and they
used the older version of OWL (OWL 1).
Moawad et al. (2012) presented building a viral hepatitis ontology based on OBR
framework. A three stages methodology; acquisition, validation, and OWL representation
was used. In designing the ontology, the bottom-up approach was used and in
implementation, they had used Protégé.
Salem and Katoua (2012) introduced a five steps approach for developing a
web-based ontology of knowledge engineering. We favoured to apply this methodology
rather than in Moawad et al. (2012) and coming (Jusoh et al., 2013) as it was found
simple, clear, logical and more proper for developing the ORC ontology.
To develop an ontology for breast cancer domain, Jusoh et al. (2013) used a hybrid
approach. They followed a three phases methodology that included three stages:
1 preparation
2 hybrid ontology process
a build global ontology
b build local ontology
c mapping between global and local ontologies
d mapping between data sources and local ontology.
3 development of ontology.
Their knowledge resources are involved a medical officer as a domain expert, and
documentation was taken from journals, articles, and websites. They did not use fuzzy
logic in their work.
Torshizi et al. (2014) presented a savvy hybrid system based on fuzzy-ontology that
determines the severity level and recommends the treatment for benign prostatic
hyperplasia (BPH). They used ontologies for expert’s knowledge representation. They
used brainstorming procedure among experts. They used fuzzy logic to make inference
on rule bases using fuzzy variables that are in the form of if-then rules. They did not work
directly on the ontology itself; they needed to transform it into if-then rules to be used
within a fuzzy system.
Elhefny et al. (2014) introduced building a crisp ontology for representing the
obesity-related cancer domain that included diseases, diagnosis, treatment, and symptoms
classes. They had constructed their ontology using a simple methodology within Protégé
building environment; it was formatted in OWL2-DL syntax. It was helpful to reuse their
ontology as the core of the first phase of developing our fuzzy ontology for ORC domain.
170 M.A. Elhefny et al.
We extended their ontology by adding more concepts and terms, instances, and
properties. Then, we integrated fuzzy logic to handle the overlapped concepts, linguistic
variables, and uncertainty circumstances to get a more efficient representation of the
domain knowledge.
4 The proposed ORC fuzzy ontology
It is important to represent linguistic variables and overlapped concepts of Semantic Web
Languages in a standard way. It can be performed by either developing the current
Semantic Web Languages or by a procedure representing such information within current
standard languages and tools. In our suggested framework, we utilised the last approach
within OWL2 to represent the ORC ontology to meet the mentioned needs. Figure 2
shows the block diagram of the proposed framework to build the fuzzy ontology.
Figure 2 The proposed framework for building the fuzzy ontology of ORC domain (see online
version for colours)
User
Fuzzy Ontology
for that Element
Query for Element
Translated Fuzzy Ontolo gy
Into Fuzzy Dl reasoner
Language
Obesity
Related
Cancer
Domain
Build Core
Ontology
Parser
Fuzzy DL
Reasoner
Use OWL 2 with
Protégé Fuzzy
Fuzzy
Ontolog
We constructed our work in three main phases. First, we built the typical ontology by
using Protégé 4.3 to allow reasoning using standard ontology reasoners, validation, and
evaluation. The output of this phase is a validated ontology for obesity-related cancer
with no fuzzy. Second, we represented the fuzzy data types and overlapping concepts by
using OWL2 and fuzzy annotation properties through FuzzyOWL2 plug-in (Bobillo and
Straccia, 2011). Finally, we reasoned with/ queried the constructed fuzzy ontology using
fuzzyDL.
4.1 Phase 1: building ORC ontology
4.1.1 Building methodology
We began building our ontology by analysing the vocabularies of obesity-related cancer
domain. Then, we identified the most commonly used terminologies by physicians from
Mansoura University Hospitals. Several official sources were used like Obesity and
Cancer Risk – National Cancer Institute (2015), Calle and Kaaks (2004), Tumor Markers
– National Cancer Institute (2015), OncoLink – The Web’s First Cancer Resource (2015),
and Singhal et al. (2013). In ‘disease’ superclass, we made our class based on ‘DO’
(Disease Ontology – Institute for Genome Sciences, 2015) hierarchy and terms. We
followed a methodology with five processes used in Elhefny et al. (2014), as shown in
Figure 3, and pre-built ORC ontology as the core to start adding the new items to it. We
consulted the domain experts to validate the classification trees, edit terminologies, add
Developing a fuzzy owl ontology for obesity related cancer domain 171
other classes, and determine concepts synonyms. Our first layout of building ontology is
shown in Figure 4 with classes, properties, and relationship.
Figure 3 ORC ontology building methodology (see online version for colours)
Organizing and Scoping
• Determinetheobjectives.
• TheOntologyboundariesdefinition.
Data Collection & Knowledge Acquisition
Acquiring the raw data needed for ontology
develo
p
ment.
Data Analysis
• DescribetheclassesandClasshierarchy.
• Describethepropertiesoftheclasses(slots).
• Describethefacetsoftheslots(e.g.domain&
rangeofaslot,cardinality,slot‐valuetype)
• Createindividualinstancesofclasses.
Building an initial ontology from Raw Material
A preparatory Ontology is produced
Ontology Refinement
The initial develo
p
ment is iterativel
y
refined.
• Semantic Commitment
• Linguistic Study
• Approach: Top-Down
• Language: OWL 2-DL
• Tool: Protégé
With the aiding of the
domain experts.
• Interviews
• Observations
• Document Analysis
• Questioning
• Brainstorming & Discussion
Source: Elhefny et al. (2014)
Figure 4 The ontology building initial layout (see online version for colours)
172 M.A. Elhefny et al.
4.1.2 Ontology structure
Figure 5 displays the ‘DO/DOID’ visualised hierarchy for the obesity disease class (as an
example). After applied the ‘DO’ terms and hierarchy to obesity and cancer disease
classes, we represented them using Protégé.
Figure 5 The obesity class visualisation in DO (see online version for colours)
Source: Ontology – Institute for Genome Sciences (2015)
‘DO’ stated that the ovarian cancer has synonyms such as malignant ovarian tumour,
neoplasm of ovary (disorder), ovarian neoplasm, ovary cancer, a tumour of the ovary, …
etc., To reduce the time and effort, we considered the term ‘ovarian cancer’ (that is a
subclass of Female_reproductive_organ_cancer) as a member (individual) of its
superclass cancer, without taking into account the very detailed synonyms and sub-items,
as the experts recommended, so we also did for some other diseases.
ORC ontology consists of five superclasses: disease, medical intervention, references,
patient, and country (Elhefny et al., 2014). Medical intervention class includes both
diagnosis that in turn involves cancer diagnosis and obesity diagnosis subclasses.
Treatment class includes cancer treatment and obesity treatment subclasses. References
class includes risk factors subclass that in turn involves cancer risk factors and obesity
risk factors subclasses and symptoms subclass that in turn involves cancer symptoms and
obesity symptoms subclasses. Patient class involves Male Patient and Female Patient
subclasses. Country class involves Egypt. Our used relationships are is_a, has_Disease,
IsLocatedIn, ResultsIn, hasCauses, hasSymptoms, diagnosedBy, treatedBy, see Table 7.
Table 8 displays the classes’ individuals. The hierarchy of the full ORC ontology classes
Developing a fuzzy owl ontology for obesity related cancer domain 173
is shown in Figure 6 throughout Protégé environment. Figure 8 shows an excerpt of k
classes, data and object properties in the ontology (partial).
Table 7 The object properties of the obesity-related cancers ontology
Domaina Rangeb Property
Patient Disease hasDisease
Patient Country IsLocatedIn
Obesity Obesity_Risk_Factors hasCauses
Obesity Obesity_Symptoms hasSymptoms
Obesity Obesity_Diagnosis diagnosedBy
Obesity Obesity_Treatment treatedBy
Obesity Diabetes ResultsIn
Obesity Diabetes ResultsIn
Cancer Cancer_Risk_Factors hasCauses
Cancer Cancer_Symptoms hasSymptoms
Cancer Cancer_Diagnosis diagnosedBy
Cancer Cancer_Treatment treatedBy
Notes: aDomain is a built-in property that links a property to a class description.
bThe range is a built-in property that links a property to either a class description
or a data range.
Table 8 The instances (individuals) of ORC ontology classes
Class Instances
Obesity_Risk_Factors Gender, Age, Lipids, Genes_and_family_history, Diabetes, Lifestyle,
Hormone_problems, Certain_medicines, Lack_of_sleep,
Emotional_factors, Smoking_stopping, Pregnancy,
Lack_of_energy_balance_over_time
Obesity_Symptoms Clothes_feeling_tight, Having_extra_fat_around_the_waist,
Greater_scale_measure, A_Higher_than_normal_BMI,
Having_higher_waist_circumference
Obesity_Diagnosis Gender, BMI, Blood_glucose_level_tests (Fasting_plasma_glucose,
HbA1c, Oral_gulcose_tolerance), Lipid_profile_tests (Triglyceride,
Total_cholestrol), Genetic_factors , Waist_measurement,
Retrospective_studies_in_community
Obesity_Treatment Weight_loss, Lifestyle_change (Cutting_back_on_calories,
Healthy_eating_plan, Physical_Activity), Medicines, Surgery
Cancer_Risk_Factors Age, Gender, Morbid_obesity, Inherited_gene_faults, Lifestyle,
Smoking, DNA_damage, Viruses,
Problems_with_the_immune_system
Cancer_Symptoms Feeling_ill_without_obvious_cause, Pernicious Anemia
Tumour_mutations
Cancer_Diagnosis Imaging (X-ray, CT_scan, MRI_scan, PET_scan, Ultrasound),
Tumor_markers_tests (CA-125, AFP, …), Biopsy, Endoscopy,
Physical_examination
Cancer_Treatment Radiotherapy, Surgery, Chemotherapy, Hormone_therapy,
Immunotherapy, Gene_therapy
174 M.A. Elhefny et al.
Figure 6 The ORC full ontology classes (see online version for colours)
Developing a fuzzy owl ontology for obesity related cancer domain 175
4.1.3 Some domain considerations
Initially, we thought to add cancer ‘staging’ Class, and then we found that most of the
staging 16 tests were involved in the ‘diagnosis’ class, as shown in Figure 7. In
Tumor_markers_tests class, we considered the tumour markers for our concerned cancer
types (pancreatic, colorectal, ovarian, liver, and breast cancers) according to NCI (Tumor
Markers – National Cancer Institute, 2015). Then, we refined them to fix what are done
in Egypt as in Table 3. We considered adding diabetes as a strong relationship between
obesity and diabetes mellitus exists. It is observed the higher BMI (morbid obesity) leads
to higher blood glucose levels (Type 2 diabetes). For cancer, we selected the most
common cancers that exist in Mansoura University Hospitals as mentioned in Calle and
Kaaks (2004). The Mansoura University Hospitals serve patients from all the Egyptian
cities. We found some terms were so close to each other, such as (colon cancer,
colorectal, colon and rectum, colon adenocarcinoma), (liver, hepatocellular), and
(tumour, neoplasm) that might be treated as overlapping concepts. Our information
resources included other websites like14, 15, 16 for obesity, cancer, and cancer staging,
respectively.
Figure 7 Diagnosis class and its members represented in ORC ontology (see online version
for colours)
176 M.A. Elhefny et al.
Figure 8 Selection of classes, data and object properties in the ontology (partial) (see online
version for colours)
4.1.4 Ontology validation and metrics
We tested the ontology consistency by using Protégé built-in reasoner(s), and its
validation by the experts’ review. Statistics and format validation were made using the
online tool provided by Manchester University, ‘ontology metrics’17, 18 for statistics and19
for format validation. Figure 9 indicates the general metrics of core (crisp) ontology using
the online tool. The metrics for both the typical (core) ontology (Elhefny et al., 2014) and
our modified one are displayed in Figure 10 using the built-in metrics tool in Protégé.
The syntax validation of our ontology to OWL 2-DL is reported in Figure 11 using
Manchester University validation tool.
4.2 Phase 2: ORC ontology extension – adding the fuzzy part using annotation
properties
We fuzzified our ORC ontology to accommodate the linguistic variables (e.g., BMI data
types; underweight, normal, overweight, and obese) and overlapping concepts (e.g.,
colorectal cancer, colon cancer, and colon adenocarcinoma). In our regular ontology, we
could not do that, as there are no sharping edges among concepts. In addition, linguistic
variables have different ranges of values.
Developing a fuzzy owl ontology for obesity related cancer domain 177
Our fuzzification process was based on OWL2 and fuzzy annotation properties that
could be done within the Protégé 4.3. The FuzzyOWL2 plug-in, made by Bobillo and
Straccia (2011) is publicly available on the web. It enables defining fuzzy elements to the
typical ontology (including fuzzy data types, weighted sum concepts, weighted concepts,
fuzzy nominals, and others), specifying the fuzzy logic wanted to be used (either Zadeh20
or Lukasiewicz21 logics). We used Zadeh fuzzy logic. The process output is a fuzzy
ontology.
Figure 9 The typical ORC ontology metrics using Manchester University validation tool
(see online version for colours)
Source: Elhefny et al. (2014)
178 M.A. Elhefny et al.
Figure 10 The typical ontology metrics (Elhefny et al., 2014) vs. our modified ORC ontology
metrics in Protégé (see online version for colours)
Developing a fuzzy owl ontology for obesity related cancer domain 179
Figure 10 The typical ontology metrics (Elhefny et al., 2014) vs. our modified ORC ontology
metrics in Protégé (continued) (see online version for colours)
180 M.A. Elhefny et al.
Figure 11 The validation report for our ORC ontology format (see online version for colours)
Eventually, the constructed fuzzy ontology uses fuzzyDL to reason with/query the
processed ontology. The plug-in is integrated with fuzzyDL reasoner (Bobillo and
Straccia, 2008), translates the annotated OWL2 ontology into fuzzyDL syntax, calls
fuzzyDL, and makes it possible to submit queries. For the moment, such queries must be
expressed using the particular syntax supported by fuzzyDL.
The fuzzy ontology can be printed on the screen or saved to a text file. The
FuzzyOWL2 plugin installation included gurobi optimisation tool22 installation to use the
query panel of the plug-in. All installation instructions were included in their plug-in
documentation.
4.2.1 Definition of fuzzy sets
As we showed in Table 1, BMI had four linguistic variables; they are underweight,
normal weight, overweight, and obese that can be fuzzified denoting the degree of a
patient being underweight, normal, overweight or obese and then represented by the
fuzzy plug-in using Protégé.
We can define the four fuzzy sets of BMI linguistic variables like:
1 underweight = FUZZY SET (18.5,1), (19.5,0)
2 normal = FUZZY SET (18.5,0), (19.5,1), (24,1), (25,0)
3 overweight = FUZZY SET (24,0), (25,1), (29,1), (30,0)
4 obese = FUZZY SET (29,0), (30,1).
The first three fuzzy sets were defined by Fehre et al. (2010) and upon them, we
described the fourth one, see Figure 12.
Developing a fuzzy owl ontology for obesity related cancer domain 181
Figure 12 The fuzzy sets for underweight, normal, and overweight BMI (see online version
for colours)
Source: Fehre et al. (2010)
Figure 13 The underweight fuzzy data type representation (as an example) (see online version
for colours)
4.2.2 Fuzzy data types representation
To represent the fuzzy atomic data types, we need to specify the parameters k1, k2, a, b,
c, d. The first four parameters are common to all of them, c and d appear in the
trapezoidal function. The parameters k1 and k2 are the minimum and maximum inclusive
values, respectively. These parameters are optional and, if omitted, then the minimum
182 M.A. Elhefny et al.
and maximum of the attributes (a, b, c, d) are assumed, respectively. We specified k1, k2
with 0, 300 as the heaviest human till now recorded more than 204 BMI (Kg/m2)23. We
represented our fuzzy data types using values of (k1, k2, a, b, c, d) as the following using
two left and right triangle functions, and two trapezoidal functions. Figure 13 shows the
underweight fuzzy data type representation in the building environment as an example:
• Underweight_datatype = Left(0, 300, 18.5, 19.5)
• Normalweight_datatype = Trapezoidal(0, 300, 18.5, 19.5, 24, 25)
• Overweight_datatype = Trapezoidal (0, 300, 24, 25, 29, 30)
• Obese_datatype = Right(0, 300, 29, 30).
The medical experts told that overlapped concepts of colorectal cancer were used
approximately as 60% for colorectal cancer, 30% for colon cancer, and 10% for Colon
Adenocarcinoma. Figure 14 shows a sample of the used fuzzy label annotation properties
for representing both fuzzy data types and overlapped concepts of colorectal cancer with
different degrees of usage (0.6, 0.3, and 0.1).
Figure 14 The used fuzzy label annotation properties for both fuzzy data types & overlapped
concepts of colorectal cancer (see online version for colours)
4.2.3 Phase 3: query the fuzzy ontology
Using the installed plugin and Gurobi software, we can send queries in specified syntax
and predefined tags to our constructed fuzzy ontology and get fuzzy answers. To check
our ontology response and consistency, we made some queries like:
Developing a fuzzy owl ontology for obesity related cancer domain 183
(max-subs?Colorectal_cancerLarge_intestine_cancer), (min-subs?Colorectal_cancer
Large_intestine_cancer) to get the maximum and minimum values of concept implication
Colorectal_cancer ->Large_intestine_cancer, as shown in Figure 15.The fuzzy ontology
responded with the expected answers for the given queries.
Figure 15 Results: get the minimum and maximum values of concept implication
Colorectal_cancer ->Large_intestine_cancer (see online version for colours)
5 Results
The fuzzy ontology validation was made in two stages. First, validating the phase 1 of the
regular ontology using the regular validation tools as shown in Section 4.1.4. Then,
validating the second and third phases of getting answers from the fuzzy Dl reasoner that
reflect user’s queries using the fuzzy annotations approach.
The fuzzy ontologyreplied with the expected answers for the given queries, if the
fuzzy ontology had something wrong with fuzzy representation, the reasoner would
provide an only ‘ERROR’ response with no result. In addition, Bobillo and Straccia
(2011) made an experimental evaluation of using fuzzy annotation properties. The final
evaluation decision was there is no additional overhead for the annotations, and good
performance was acquired.
184 M.A. Elhefny et al.
Figure 16 The results of the experimental evaluation, (a) the impact of the percentage of
annotations in the parsing time (b) the impact of the percentage of annotations in the
translation time (c) influence of the percentage of annotations in the parsing time and
the translation time into FuzzyDL syntax (see online version for colours)
(a)
(b)
(c)
Source: Bobillo and Straccia (2011)
Developing a fuzzy owl ontology for obesity related cancer domain 185
Figure 16 displays their experimental evaluation results using Galen ontology.
Figure 16(c) shows the influence of the percentage of annotations (%) in both PT (the
parsing time) and TT (the translation time) into fuzzyDL syntax. The parsing time and
the translation time are shown for both weighted sums (WSs) and weighted concepts
(WCs).
The numbers of annotated elements influence in the PT is shown in Figure 16(a). It is
noticeable there is a semi-linear growing of the PT concerning the number of annotated
elements. A fuzzy ontology with a 40% of annotated elements would take one more
second to be parsed than the original Galen ontology. In addition, it is obvious that there
are no considerable differences between WCs and WSs, in general, which means the
types of the fuzzy concepts are not significative.
The numbers of annotated elements influence in the TT is shown in Figure 16(b).
Again, there is a semi-linear growing of the running time concerning the number of
annotated elements, and there are no significant differences because of the type of the
fuzzy concepts (Bobillo and Straccia, 2011).
6 Conclusions
The ORC(s) is a rich and significant medical domain. From our experiment, the proposed
fuzzy ontology was better to represent this domain than the typical one for several
reasons. One of them was the ability to represent overlapping concepts and linguistic
variables that had not sharp edges to be represented in regular ontologies, and this was
done via the fuzzy annotation properties (like fuzzy data types, weighted sum concepts,
… etc.). It led us to accommodate more concepts and make a wider range of
vocabularies. Second, enabling the user(s) to send queries to the fuzzyDLreasoner that in
turn replies with fuzzy ontologies. Third, it best fits for rich domains having fuzzy
knowledge that need to be represented within ontologies. Finally, it leads to good
performance, in general.
We introduced a simple three phases methodology to build the fuzzy ontology that is
expected to be good practice for ontologists and knowledge engineers in medical field
aiding them to solve the overlapping concepts, linguistic variables, and reasoning
problems. Both physicians and intelligent systems can exploit obesity-related cancer
fuzzy ontology in knowledge sharing, reusability, and reasoning.
In future, we may extend this work to include all the cancer types with their elements
in the fuzzification process or may work on a particular cancer type with the study of
patients group’s real data. The fuzzy plug-in may need more development to facilitate the
users to submit more queries than the predefined ones.
Acknowledgements
The authors want to express gratitude toward Dr. Ashraf Khater, Prof. of Surgical
Oncology for his efforts, Dr. Moahamed Awad Ibrahem, Assoc. Prof. of Medical
Oncology (Mansoura University Oncology Center), Dr. Noureddin Sadawi (Department
of Computer Science, Brunel University, London, UK), and Mr. Wolfram Bartussek
186 M.A. Elhefny et al.
(Lecturer in University of Applied Sciences, Darmstadt, Germany) for their support
during this work.
References
Alfonse, M., Aref, M.M. and Salem, A-B.M. (2012) ‘Ontology-based knowledge representation for
liver cancer’, Electronic Proceedings of the International eHealth, Telemedicine and Health
ICT Forum for Educational, Networking and Business, pp.840–844.
Bobillo, F. and Straccia, U. (2008) ‘FuzzyDL: an expressive fuzzy description logic reasoner’,
IEEE International Conference on Fuzzy Systems, Vols. 1–5, pp.923–930.
Bobillo, F. and Straccia, U. (2011) ‘Fuzzy ontology representation using OWL 2’, International
Journal of Approximate Reasoning Archive, October, Vol. 52, No. 7, pp.1073–1094.
Calle, E.E. and Kaaks, R. (2004) ‘Overweight, obesity and cancer: epidemiological evidence and
proposed mechanisms’, Nature Reviews Cancer, August, Vol. 4, No. 8, pp.579–591, doi: 10.
1038/nrc1408.
Chen, S-M., Huang, Y-H., Chen, R-C., Yang, S-W. and Sheu, T-W. (2012) ‘Using fuzzy reasoning
techniques and the domain ontology for anti-diabetic drugs recommendation’, in Pan, J-S.,
Chen, S-M. and Nguyen, N.T. (Eds.): ACIIDS 2012, Part I, LNAI 7196,
Springer-VerlagBerlin, Heidelberg, pp.125–135.
Elhefny, M.A., Elmogy, M. and Elfetouh, A.A. (2014) ‘Building OWL ontology for obesity related
cancer’, 9th IEEE International Conference on Computer Engineering & Systems (ICCES),
Cairo, Egypt, pp.177–183, ISBN: 978-1-4799-6593-9, DOI: 10.1109/ICCES.2014.7030953.
Fehre, K., Mandl, H. and Adlassnig, K-P. (2010) ‘A fuzzy arden syntax compiler’, Schreier, G.,
Hayn, D. and Ammenwerth, E. (Eds.): Tagungsband der eHealth2010: Health Informatics
meets eHealth, 6–7 Mai 2010, Wien, OCG.
Jusoh, F., Ibrahim, R., Othman, M.S. and Omar, N. (2013) ‘Development of breast cancer ontology
based on hybrid approach’, International Journal of Innovative Computing, Vol. 1, No. 1,
pp.15–22.
Moawad, I.F., Al Marzoqi, G. and Salem, A-B.M. (2012) ‘Building OBR-based owl ontology for
viral hepatitis’, Electronic Proceedings of the International eHealth, Telemedicine and Health
ICT Forum for Educational, Networking and Business, pp.821–825.
Obesity and Cancer Risk – National Cancer Institute (2015) Fact Sheet [online] http://www.cancer.
gov/cancertopics/factsheet/Risk/obesity (accessed 30 December 2015).
OncoLink – The Web’s First Cancer Resource (2015) Patient Guide to Tumor Markers [online]
http://www.oncolink.org/treatment/article.cfm?id=296 (accessed 11 December 2015).
Web Ontology Language (OWL) (2015) [online] http://www.w3.org/TR/owl-features/ (accessed 1
March 2014).
Parry, D. (2004) ‘A fuzzy ontology for medical document retrieval’, ACSW Frontiers ‘04
Proceedings of the Second Workshop on Australasian Information Security, Data Mining and
Web Intelligence, and Software Internationalisation – Volume 32, ACM, pp.121–126.
Salem, A-B.M. and Katoua, H.S. (2012) ‘Web-based ontology of knowledge engineering’, Journal
of Communication and Computer, Vol. 12, No. 9, pp.1336–1342.
Semantic Web (2015) [online] http://en.wikipedia.org/wiki/Semantic_Web (accessed 20 March
2015)
Semantic Web Wiki (2015) [online] http://semanticweb.org/wiki/Main_Page (accessed 20 March
2015).
Singhal, M., Manchireddy Vishnu, V.R., Rama Raju, S.V. and Upadhyay, Y. (2013)
‘Interrelationship between obesity and cancer (a review)’, Academic Journal of Cancer
Research, Vol. 6, No. 1, pp.13–20, ISSN: 1995-8943, DOI: 10.5829/idosi.ajcr.2013.6.1.739.
Developing a fuzzy owl ontology for obesity related cancer domain 187
Torshizi, A.D., Torshizi, G.D. and Eghbali, K. (2014) ‘A hybrid fuzzy-ontology based intelligent
system to determine level of severity and treatment recommendation for benign prostatic
hyperplasia’, Computer Methods and Programs in Biomedicine, Vol. 113, No. 1, pp.301–313.
Tumor Markers – National Cancer Institute (2015) Fact Sheet [online] http://www.cancer.gov/
cancertopics/factsheet/detection/tumor-markers (accessed 21 December 2015).
World Wide Web Consortium (W3C) (2015a) OWL2 Web Ontology Language Document
Overview (Second Edition) [online] http://www.w3.org/TR/owl2-overview (accessed 23
December 2013).
World Wide Web Consortium (W3C) (2015b) OWL2 New Features [online] http://www.w3.org/
TR/2012/REC-owl2-new-features-20121211/ (accessed 1 December 2013).
Zhai, J., Shen, L., Zhou, Z. and Liang, Y. (2008) ‘Fuzzy ontology model for knowledge
management’, Intelligent Information Processing IV, IFIP – The International Federation for
Information Processing Volume 288, pp.58–67.
Notes
1 http://nemis.isti.cnr.it/~straccia/software/fuzzyDL/fuzzyDL.html.
2 http://protege.stanford.edu/.
3 http://nemis.isti.cnr.it/~straccia/software/FuzzyOWL.
4 http://www.opengalen.org.
5 http://www.nlm.nih.gov/mesh/meshhome.html.
6 http://www.snomed.org.
7 http://www.geneontology.org.
8 http://www.bioontology.org.
9 http://www.obofoundry.org.
10 http://www.nih.gov/- The National Institutes of Health.
11 http://labtestsonline.org/understanding/analytes/glucose/tab/test/.
12 http://www.mayoclinic.org/diseases-conditions/high-blood-cholesterol/in-depth/cholesterol-
levels/art-20048245.
13 http://www.medscape.com/viewarticle/487381_4.
14 http://www.nhlbi.nih.gov/health/health-topics/topics/obe/causes.html.
15 http://www.cancerresearchuk.org/cancer-info/cancerandresearch/all-about-cancer/what-is-
cancer/what-causes-cancer.
16 http://www.cancer.gov/cancertopics/factsheet/detection/staging.
17 http://mowl-power.cs.man.ac.uk:8080/metrics.
18 http://www.w3.org/2001/sw/wiki/Ontology_Metrics.
19 http://mowl-power.cs.man.ac.uk:8080/validator/.
20 http://en.wikipedia.org/wiki/Fuzzy_logic.
21 http://en.wikipedia.org/wiki/%C5%81ukasiewicz_logic.
22 http://www.gurobi.com.
23 http://en.wikipedia.org/wiki/List_of_the_heaviest_people.
