Post on 04-Dec-2016
A Framework for Fuzzy Diagnosis of Hepatitis
Okure U. Obot
Department of Computer Science
University of Uyo, PMB. 1017 520003
Uyo, Akwa Ibom State, NIGERIA
e-mail: abatakure@yahoo.com
Samuel S. Udoh
Department of Computer Science
University of Uyo, PMB. 1017 520003
Uyo, Akwa Ibom State, NIGERIA email: udohss@yahoo.com
Abstract—A study of the orthodox practice of diagnosing
hepatitis revealed that inexactness in the diagnostic results has
led several patients into abusing therapies. This prompted a
further study into how this could be resolved. In this regard,
effort was made for medical doctors to specify some linguistic
labels while taking history and performing medical examinations
on the patients. The effort yielded few responses which
necessitated a study of the application of fuzzy logic technology
to medical diagnosis. The symptoms were fuzzified with some
membership functions which aided in the extraction of fuzzy rule
base. With data and rules, fuzzy inference using the maxmin
method was applied on the knowledge base, the results obtained
were defuzzified to obtain crisp outputs that represent the
diagnostic values with linguistic labels. The novelty of the result
is that the degree or extent to which a patient suffers from
hepatitis is reported to the patient and based on such revelation
therapy would be administered without an abuse.
Keywords-Fuzzy Logic; Hepatitis; Maxmin; Knowledge base;
Decision Support Filter
I. INTRODUCTION
Medical diagnosis is a very important component of medical
and health care delivery. Improper diagnosis of an ailment
often results in incorrect treatment that leads to
complications of the ailment and eventually to death. What
is involved is the elicitation of the signs and symptoms of
the disease and the determination of the extent or degree of
effects the symptoms have had on the organs. When this is
determined, appropriate therapy is administered to alleviate
the pains and heals whatever effects the organs have had.
The task of doing this effectively and efficiently at the right
time is complex and requires a deep knowledge of the
diseases, and the history of the patient. The need to arrive at
the most accurate medical diagnosis in a very timely manner
is heightened in the case of malaria and other tropical
conditions, as it is understood that a quick and accurate
diagnosis and timely initiation of treatment is a sine-qua-
non to the reduction of complications [1].
The problem of medical diagnosis is further
complicated when the decision variables (signs and
symptoms) of the disease are numerous as in the case of
hepatitis. Additionally, when the variables involved seem to
overlap with the variables of other tropical diseases.
Thirdly, the problems of uncertainty and imprecision of the
data involved is a serious threat to proper diagnosis. A
number of attempts have been made to resolve the problems
of multiple attributes decision making and conflicting
decision variables.
Researchers in medical expert systems have also
attempted to find ways to manage uncertainty and
imprecision of data in medical diagnosis. It is difficult for a
patient to tell a medical practitioner how exactly he feels
and the degree of pains he experiences using linguistics
variables. At the same time, a medical doctor cannot
determine the exact degree or extent of the sign of a disease
based on his examination. Most of the variables found in
medical diagnosis are fuzzy in form. For example, a patient
who complains of a general discomfort might not be able to
explain the degree of discomfort. A medical doctor who
examines a patient’s liver and reports of its tenderness will
only use fuzzy variables to explain the tenderness.
Hepatitis is an acute inflammation of the liver caused
by some infectious or toxic agent [2]. It can be categorized
broadly as Hepatitis A or infectious hepatitis and Hepatitis
B or Serum hepatitis. Hepatitis B is more chronic and severe
than Hepatitis A although they have conflicting signs and
symptoms. Hepatitis A is an acute virus infection that harms
the liver. The symptoms of hepatitis are fuzzy and
ambiguous. Some of the signs and symptoms of Hepatitis A
and their weights are presented in Table 1.
According to Obot and Uzoka [3], the weights were
assigned with the help of some experienced physicians. The
variables are weighted according to the severity and
specificity of the signs and symptoms of hepatitis. Some of
the symptoms are discriminatory of hepatitis while others
are not. Some are specifically signs and symptoms of
hepatitis; others are not really specific but relate to it. These
form the basis of the grading or weighting factors. For an
instance, jaundice is specifically a clear sign of hepatitis and
is thus assigned the weight of 4, this is also true of skin and
eye discolorations, but the two are assigned each a weight of
3 because of their degree of severity of the cause of the
disease. Symptoms like fever, body weakness and headache
are discriminatory, that is, their presence could be because
of diseases like malaria and typhoid. Nausea, vomiting and
tender liver are specifically symptoms of hepatitis but could
also result in other diseases, so a weight of 2 is assigned to
each of them.
439978-1-4673-0126-8/11/$26.00 c©2011 IEEE
TABLE I. SIGNS AND SYMPTOMS OF HEPATITIS
S/N Name Code Weights
1 Nausea Nau 2
2 Vomiting Vom 2
3 Fever Fev 1
4 Body Weakness Wea 1
5 Loss of appetite App 1
6 Diarrhea Dia 1
7 Itching Itc 1
8 Convulsion Con 2
9 Stupor Stu 2
10 Headache Hea 1
11 Tremors Tre 1
12 Skin
discoloration
Ski 3
13 Eye discoloration Eye 3
14 Liver tenderness Liv 2
15 Bile in urine Bil 2
16 Jaundice Jau 4
The inexactness of the orthodox practice results which do
not present the degree or extent a patient suffers from
hepatitis and the ability of fuzzy logic technology in
resolving conflicts based on aggregation and collaboration
further motivated this study.
Fuzzy logic deals with imprecise, vague and ambiguous
dataset [4] as witnessed in the signs and symptoms of
hepatitis. The objective of the research is to apply the
concepts of fuzzy logic technology to determine the degree
of severity or extent of the influence of the signs and
symptoms of hepatitis to the diagnosis of hepatitis. The
diagnostic results will also resolve the fuzziness currently
exhibited in the diagnosis of hepatitis as can be observed in
Obot and Uzoka [3]. The study seeks to; develop fuzzy
membership functions for each of the symptoms of
hepatitis; develop fuzzy rule-base from which a maxmin
method of inference is applied to infer and compose the
datasets. Defuzzify the composed dataset into corresponding
output using the centre of gravity method, the results of this
form the diagnostic results which define the degree or extent
a particular patient is suffering from hepatitis. In particular,
the datasets are obtained from patients suffering from
hepatitis A. Fuzzy logic has been applied in the diagnosis of
diseases in [5-7].
In Section II, the system architecture is presented while
the research experiment is conducted and presented in
Section III. Discussion of the results of the experiment is
carried out in Section IV and Conclusion is drawn in
Section V.
II. SYSTEM ARCHITECTURE
The architecture of the framework comprises database and
rule base, which make up the knowledge base. The
inference engine employs fuzzy logic reasoning
methodology which derives its strength from the fuzzified
inputs that was also used to build the fuzzy rule base. The
maxmin method of drawing inference is employed on the
rule base, the results of this procedure are fed into the
defuzzification module to produce crisp output. This can be
filtered emotionally and cognitively by the Decision
Support Filter component of the system. The architecture is
presented in Figure 1.
Figure 1. The Architecture of Fuzzy Diagnosis of
Hepatitis
A. The Database The database is conceptualized as a relation
),...,,,( 321 naaaaR , where R is the name of the relation
and ),...,,,( 321 naaaa are the attributes of the relation. The
main relations of the system are as [3] follows:
Patient_Biodata (Ptient_number, patient_name,
date_of_birth, residential_address, telephone_number,
next_of_kin, address_of_next_of_kin, date_of_registration,
time_of_registration, date_of_history, date_of_exam)
Disease_symptom(patient_number, nausea, vomiting,
body_weakness, loss_of_appetite, diarrhea, skin_itching,
convulsion, stupor, headache, tremor).
Disease_sign(Patient_number, temperature, eye_color,
skin_color, tenderness_of_liver, bile_in_urine, jaundice).
The unstructured knowledge was obtained through
interaction with some experience medical practitioners who
assisted in assigning weights to the signs and symptoms
according to the specificity and severity of the symptoms to
hepatitis as presented in Table 1.
B. Fuzzy System The fuzzy system is conceptualized as presented in Figure 2.
The fuzzification subcomponent of the module transforms
raw data using membership functions defined in equations
1-4. During this process, linguistic labels are assigned to the
symptoms with their corresponding degree of severity.
440 2011 World Congress on Information and Communication Technologies
Figure 2. Fuzzy Logic System
All the 16 signs and symptoms do not contribute
significantly to the cause of Hepatitis. In order to identity
which of them contributes significantly, an input
contribution measures was performed using sensitivity test
of NeuroSolutions 5.0 [8]. Six input variables namely;
Jaundice, skin discoloration, eye discoloration, nausea, urine
in Bile and liver tenderness were found to contribute
immensely to the outputs [Obot and Uzoka, 2009], while the
other 10 variables contribute very meagerly. These were
therefore used to formulate the membership functions and
consequently the fuzzy rules represented in the form of
production rules.
There are 3 groups of signs and symptoms categorized
according to their weights (2,3,4) as can be seen in Table 1.
These are classified into different membership functions
accordingly. Symptoms with a-2-point weight are presented
in equation 1, those with a-3-point weight are presented in
equation 2 while the one with a-4-point weight is presented
in equation 3. These are Nausea (2), Bile in Urine (2), liver
tenderness (2), eye discoloration (3), and Jaundice (4).
For the 2 point weight symptoms the membership function
is as presented in equation (1)
<=<=
<=<=
<=<=
=
2.0 sym(x)1.6IfHigh
1.5 sym(x)0.1IfModerate
0.5 sym(x)0.1IfLow
xSymptom )( (1)
The 3 point weight symptoms have membership function as
defined in equation (2)
<=<=
<=<=
<=<=
=
3.0x2.0IfHigh
1.9x1.1IfModerate
1.0x0.1IfLow
xSymptom )( (2)
The 4 point weight symptoms have membership function as
defined in equation (3)
<=<=
<=<=
<=<=
=
4.0x2.6IfHigh
2.5x1.6IfModerate
1.5x0.1IfLow
xSymptom )( (3)
The final diagnosis is evaluated with membership function
as presented in equation 4.
<=<=
<=<=
<=<=
=
1.0x0.80IfHepatitisHigh
0.79x0.5IfHepatitisModerate
0.49x0.1IfHepatitsLow
diagPatient )( (4)
C. The Fuzzy Rule Base
53 fuzzy rules were developed with the assistance of some
experienced doctors. The rules are represented as production
rules, some of the rules are:
R22 If Jaundice is High and Skin discoloration is
Moderate and Eye discoloration is Moderate and
Nausea is Moderate and Urine in Bile is Low and
Liver tenderness is Low Then Hepatitis is
Moderate. R39 If Jaundice is Low and Skin discoloration is Low
and Eye discoloration is Low and Nausea is Low
and Urine in Bile is Moderate and Liver tenderness
is Low Then Hepatitis is Low.
R50 If Jaundice is Low and Skin discoloration is
Moderate and Eye discoloration is Moderate and Nausea is Moderate and Urine in Bile is Low and
Liver tenderness is Low Then Hepatitis is
Moderate.
The fuzzy rules were used to develop a fuzzy rule base as
presented in Table 2 with 25 of the 53 rules.
TABLE II. FUZZY RULE BASE
Rule
No.
Jau Ski Eye Nau Bil Liv Diagnosis
1 H L M M H M M
5 H M L H M L M
8 M L L L M L L
10 L M M M L M M
15 H M M H H H H
16 H H H H H H H
17 H L L L L L L
20 M M L L L L M
21 M M M L L L M
22 H M M M L L M
24 H M M M M L H
25 H M M M M M H
26 M M M M M H M
27 M M M M H H H
30 M M M H H H H
32 M M H M M H H
34 M M H H L L M
35 M L L L L L L
36 L H H H L H M
38 M M M M H M M
39 L L L L M M L
40 L L M M L H M
42 L M M M L L L
44 L L H H L M L
50 L M M M L L M
L= Low; M= Moderate; H= High
D. The Fuzzy Inference The fuzzy inference uses the MAXMIN method of
evaluation, where the minimum of each rule precedence
2011 World Congress on Information and Communication Technologies 441
values of the firing rule is firstly obtained and the maximum
value of all the minimums in the rule base is then obtained
to represent the intermediate diagnostic value. This is used
to determine the degree of influence on the fuzzy parameter.
A patient may be diagnosed to have hepatitis or severe
hepatitis, this is fuzzy in the sense that the exact degree of
severity is not determined. The fuzzy inference helps to
determine this exactness and certainty and minimize
ambiguities.
E. Defuzzification
The defuzzification interface is a mapping from a space of
fuzzy action defined over an output universe of discourse
into a space of non-fuzzy set to obtain crisp values. There
are several methods of defuzzification but the centre of
gravity (CoG) is applied in this study because it is more
accurate in representing fuzzy sets of any shape[Cochran
and Chen, 2005]. It is an averaging technique with the
(point) masses replacing the membership values. In a
discrete form CoG is defined simply as:
=
=
=n
j
ji
n
j
jji
pU
ppU
iCoG
1
1
)(
)(
)(
(6)
where n is the number of quantization used to discretise
membership function Ui (p) of the fuzzy output i.
F. The Decision Support Filter
The decision support filter is essential because some factors
such as environment, habit and existing disease might be
responsible for a case that looks like hepatitis. A patient
provides such information to aid in the final decision of the
physician in the patient’s state of health.
III. THE RESEARCH ENVIRONMENT
10 datasets collected from a hospital about hepatitis patients
are presented in Table 3. These are evaluated using
equations 1-5 and the fuzzy rules presented in Table 2. The
transcript for the first patient is presented in Table 4. The
result of the defuzzification is presented as the final
diagnostic value.
TABLE III. RAW DATA COLLECTED FROM THE HOSPITAL
Patient
No
Jau Ski Eye Nau Bil Liv Diagnosis
P001 L L M M L M M
P002 M L L H M M M
P003 H H M H H M H
P004 H M M L L M H
P005 L L M M L M M
P006 M L L L L M M
P007 M M M H M L H
P008 M H H L L L H
P009 L L H L M M L
P010 H L M M L L M
L= low, M= Moderate, H= High
Membership values were evaluated using equations 1-4 and
the fuzzy rule base of Table 2 was employed to arrive at the
following transcript. 23 rules fired for patient number P001
and the transcript of the diagnosis is presented in Table 4.
TABLE IV. TRANSCRIPT FOR PATIENT NUMBER P001
S/N Nau
(2)
Bil(2) Liv
(2)
Eye
(3)
Ski
(3)
Jau
(4)
Min
Patient
Score
Low
(0.5)
Low(0.5) Mod
(1.5)
Mod
(1.5)
Low
(0.8)
Mod
(2.5)
1 N Y Y Y Y N 0.5
2 N N N N Y Y 0.8
3 N Y Y Y N N 0.5
4 N N N Y N N 1.5
5 N Y N N Y N 0.5
6 Y Y N N N N 0.5
7 Y Y N Y N N 0.5
8 N Y N Y N N 0.5
9 N N N Y N N 1.5
10 N N Y Y N N 1.5
11 N N N Y N Y 1.5
12 N N N Y N N 1.5
13 N N N Y N Y 1.5
14 N N N N N Y 2.5
15 N Y N N N Y 0.5
16 Y Y N N Y Y 0.5
17 N Y N N N N 0.5
18 N N Y Y N Y 1.5
19 Y N Y N Y N 0.5
20 N Y N Y Y N 0.5
21 Y Y N Y N N 0.5
22 Y Y Y N Y N 0.5
23 N Y Y Y N N 0.5
704.01.21
86.14
)(
)(
1
1
==
=
=
=
n
j
ji
n
j
jji
pU
ppU
CoG
TABLE V. FUZZY DIAGNOSIS (COMPUTATION) RESULTS
Patient No Conventional
Diagnosis
Fuzzy Diagnosis
Membership Degree
P001 M M 0.704
P002 M H 0.99
P003 H H 0.97
P004 H H 0.89
P005 M H 0.95
P006 M M 0.76
P007 H H 0.91
P008 H M 0.78
P009 L L 0.45
P010 M M 0.67
442 2011 World Congress on Information and Communication Technologies
IV. RESULT OF DISCUSSION AND CONCLUSION
In this study, the diagnosis of 10 patients suffering from
Hepatitis A was evaluated. Other methodologies such as
rule base, case base and artificial neural networks have been
applied in the diagnosis of hepatitis. One advantage fuzzy
diagnosis has over the other methodologies is that it
resembles human decision making with its ability to work
from approximate reasoning and ultimately finding precise
solution. A patient who is diagnosed of Hepatitis using other
methodologies will not be told the degree or the extent to
which the disease has in the patient. With fuzzy diagnosis
this is possible as can be seen in the results presented on
Table 5. For example, the first case (Patient No. P001), the
symptoms were fuzzified so that the degree of influence of
the symptom is captured into the inference engine of the
system.
The inference engine uses the fuzzy rule base extracted
from membership functions and ascertained by some
physicians to carry out inferences on the data. The
intermediate results were deffuzified to obtain crisp outputs.
The Decision support factor filters some emotional and
cognitive reasoning of some environmental and habitual
factors to produce the final output. The results presented
only show the defuzzified outputs as the patients whose data
were collected from could not be reached easily to obtain
such environmental and habitual factors.
Out of the ten (10) datasets processed as presented in
Table 5, seven (7) diagnosed on the fuzzy system returned
exactly as they were in the conventional diagnosis though
with varying degrees of influence. The other three (3)
results vary; patient number 002 and 005 move from
moderate hepatitis in the conventional system to high
(severe) in the fuzzy system, while number 008 drops from
high (severe) in the conventional system to moderate in the
fuzzy system. There is therefore a positive correlation
between the conventional diagnosis and the new fuzzy
diagnosis.
One problem commonly experienced in medical
practice is uncertainty and ambiguity of information; this
has been addressed by fuzzy diagnosis by determining the
exact degree of moderate hepatitis or high hepatitis as
evidenced in the results above. Obtaining data for this
research posed a serious threat to the success of the
research. It was difficult to persuade physicians to present
data in the form that is needed which is different from their
traditional presentation. This is evidenced in collecting only
few datasets and thus only a framework of the study could
be undertaken. Efforts are still being mounted on the
physicians to cooperate in this regard. The study has not
been implemented at a commercial level yet but the results
obtained so far are promising as they tend to what are also
obtained in the orthodox practice.
REFERENCES
[1] C. Theirfelder, C. Schill, C. Hatz and R. Nuesch, Trends in
imported malaria to Besel, Switzerland, Journal of Travel
Medicine, 15(6), 432- 436, 2008.
[2] C. R. Anderson, Your Guide to Health, Oriented Watchman,
Poona India, 1976.
[3] O.U. Obot and F.M.E. Uzoka, A Framework for application
of Neuro-Case Rule base hybridization in medical diagnosis,
Applied Soft Computing (2009) 245- 253, 2009.
[4] O.C. Akinyokun, O.U. Obot, J.J. Andy and U.A. Aletor,
Design of a Neuro-Fuzzy Expert System for the diagnosis and
therapy of Cardiovascular Diseases, Book of Abstract of the
42nd Annual Conference of Science Association of Nigeria,
held in Ijebu-Ode, Nigeria P7, 2006.
[5] O.U. Obot and F.M.E. Uzoka, Fuzzy Rule- based Framework
for the management of Tropical diseases, International
Journal Medical Engineering and Informatics 1(1): 7 – 17,
2008.
[6] F.M.E. Uzoka, J. Osuji and O. Obot, Clinical Decision
Support System (DSS) in the Diagnosis of Malaria: A Case
Comparison of two soft methodologies, Expert Systems with
applications 38(2011) 1537 – 1553, 2011.
[7] Zahan, S, A Fuzzy Approach to Computer Assisted
Myocardial diagnosis, Artificial Intelligence in medicine
21(1-3) 271 – 275, 2001. [8] Neurodimension Inc 3701, www.nd.com, 2005.
2011 World Congress on Information and Communication Technologies 443