Machine learning method for classification of liver disorders

Article (PDF Available)inFar East Journal of Electronics and Communications 16(4):789-800 · December 2016with 590 Reads 
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
DOI: 10.17654/EC016040789
Cite this publication
Abstract
Machine learning methods are widely used algorithms in medical field for accurate assessment of patient health data. For the last decade, the number of patients suffering from liver diseases is on increasing trend. General causes of liver disease include extreme consumption of alcohol, eating of impure food, inhale of harmful gases, and drugs. This study presents the classification of liver disorders using various machine leaning algorithms which include support vector machine (SVM), linear discriminant analysis (LDA), diagonal linear discriminant analysis (DLDA), quadratic discriminant analysis (QDA), diagonal quadratic discriminant analysis (DQDA), and Mahalanobis. The proposed model classifies liver diseases as alcoholic liver damage (ALD), primary hepatoma (PH), liver cirrhosis (LC), and cholelithiasis (C). The performance of presented algorithm is compared in terms of accuracy, sensitivity and specificity.
Advertisement
Int. J. Biomedical Engineering and Technology, Vol. 16, No. 1, 2014 27
Copyright © 2014 Inderscience Enterprises Ltd.
Intelligent techniques and applications in liver
disorders: a survey
Aman Singh*
Department of Computer Science and Engineering,
Lovely Professional University,
Jalandhar, Punjab, India
Email: amansingh.x@gmail.com
*Corresponding author
Babita Pandey
Department of Computer Applications,
Lovely Professional University,
Jalandhar, Punjab, India
Email: shukla_babita@yahoo.co.in
Abstract: Liver disease is one of the leading causes of mortality in India, as it
is in rest of the world. This paper presents a survey on intelligent techniques
applied to liver disorders between the years January 1995 and January 2013.
Individual ITs include artificial neural network (ANN), data mining (DM),
fuzzy logic (FL) etc. Integrated ITs combine methods as artificial neural
network-case-based reasoning (ANN-CBR), artificial immune system-artificial
neural network-fuzzy logic (AIS-ANN-FL) etc. The different types of liver
disorders covered in the study are hepatitis, liver fibrosis, liver cirrhosis, liver
cancer, fatty liver, liver disorders data set, hepatitis data set and hepatobiliary
disorders data set. The study identifies which ITs are applied for what types of
liver disorders and on which types of disorders maximum works has been done.
Another imperative fact emerging from this survey is that large part of the
research work on liver disorders has been done from 2007 onwards.
Keywords: ANNs; artificial neural networks; data mining; fuzzy logic; genetic
algorithm; case-based reasoning; rule-based reasoning; artificial immune system;
particle swarm optimisation; literature survey; liver disorders; intelligent
techniques; biomedical engineering.
Reference to this paper should be made as follows: Singh, A. and Pandey, B.
(2014) ‘Intelligent techniques and applications in liver disorders: a survey’, Int.
J. Biomedical Engineering and Technology, Vol. 16, No. 1, pp.27–70.
Biographical notes: Aman Singh is working as an Assistant Professor in
Department of Computer Science and Engineering at Lovely Professional
University, Punjab, India. He has about two years of teaching experience and
his areas of interest are biomedical engineering, information security, cyber
crime and computer forensics.
Babita Pandey is an Associate Professor at the School of Computer
Engineering in Lovely Professional University, Punjab, India. She has over
four years of teaching experience and has published around 30 research papers
and articles. Her main research interests are AI and multi-agent system and its
application to medicine, e-commerce and semantic web.
28 A. Singh and B. Pandey
1 Introduction
Modern era of computing has stretched its reach to the intensive and efficient usage of
Intelligent Techniques (ITs) in bioinformatics. Owing to the uncertainty in medical data
sets, deriving comprehensible information becomes a major challenge for physicians.
This challenge can lead to erroneous diagnosis of a disease, which would further lead to
improper treatment. Specifically, we can say that it would be favourable for patients if
medical experts cross-check their assessment with the help of some decision-making
systems. These systems are developed by using intelligent techniques, which
resourcefully scrutinise complex and ambiguous data sets. Implementing these ITs for
liver disorders is acting as a catalyst in overcoming the overheads and problems faced by
doctors. ITs effectively prevail over the inadequacies, help in obtaining better accuracies
and make the systems adaptable. ITs decrease the probability of occurrence of medical
errors, and reduces the cost, time and effort needed. Going into more specific discussion,
intelligent techniques that are applied to liver disorders discussed in this survey are as
follows: Artificial Neural Network (ANN), data mining (DM), fuzzy logic (FL), genetic
algorithm (GA), Artificial Neural Network-Case-Based Reasoning (ANN-CBR),
Artificial Neural Network-Data Mining (ANN-DM), Artificial Neural Network-Fuzzy
Logic (ANN-FL), Artificial Immune System-Fuzzy Logic (AIS-FL), Artificial Neural
Network-Genetic Algorithm (ANN-GA), Artificial Immune System-Genetic Algorithm
(AIS-GA), Artificial Neural Network-Particle Swarm Optimisation (ANN-PSO), Case-
Based Reasoning-Data Mining (CBR-DM), Case-Based Reasoning-Genetic Algorithm
(CBR-GA), Data Mining-Genetic Algorithm (DM-GA), Data Mining-Fuzzy Logic (DM-
FL), Fuzzy Logic-Genetic Algorithm (FL-GA), Artificial Immune System-Artificial
Neural Network-Fuzzy Logic (AIS-ANN-FL), Artificial Neural Network-Case-Based
Reasoning-Rule-Based Reasoning (ANN-CBR-RBR), Artificial Immune System-Data
Mining-Fuzzy Logic (AIS-DM-FL), Artificial Neural Network-Data Mining-Fuzzy
Logic (ANN-DM-FL), Artificial Neural Network-Data Mining-Genetic Algorithm
(ANN-DM-GA), Artificial Neural Network-Genetic Algorithm-Rule-Based Reasoning
(ANN-GA-RBR), Case-Based Reasoning-Genetic Algorithm-Particle Swarm
Optimisation (CBR-GA-PSO), Data Mining-Fuzzy Logic-Genetic Algorithm (DM-FL-
GA) and Case-Based Reasoning-Data Mining-Fuzzy Logic-Genetic Algorithm (CBR-
DM-FL-GA). Survey results confirmed the popularity and applicability of individual and
integrated ITs used for liver disorders. Though, researchers had not shown much interest
in combining two or more ITs before the year 2007.
Liver is the largest internal organ in a human body, which performs numerous
metabolic functions (Li et al., 2012). It works to filter blood, aids in digestion of fats,
makes proteins for blood clotting and most importantly detoxifies chemicals (Chuang,
2011). Liver has a vital importance to life but improper functioning of it may cause
serious health consequences. Liver disease is usually caused by inherited disorders,
contaminated food, damaged hepatocytes infected with viruses, bacteria or fungi,
excessive fat accumulation, and excessive consumption of alcohol or drugs (Chuang,
2011). Liver disease is a serious area of concern in the universal set of medicine and is
becoming the leading cause of deaths in India, as well as in other countries around the
world (http://thelivercarefoundation.org, 26th November 2013). Liver resists early
Intelligent techniques and applications in liver disorders 29
detection, as it functions normally even when partially damaged, making the disease even
more alarming because by then it might have suffered eternal damage. This indicates that
an early diagnosis of liver disorders becomes a necessity so that in time treatment can be
possible. During diagnosis, analysing complex data set of patients stretches the decision
time of doctors. To reduce this time period and effort, decision-making systems are
developed using numerous intelligent techniques.
This paper has made a contribution to medical field by presenting a study on
intelligent techniques applied to liver disorders between the years January 1995 and
January 2013. To the best of our knowledge, not a single attempt had been made to write
any survey paper on liver disorders for the last 36 years (1977–2013). Numerous authors
have written literature review sections but no complete article has found so far. This
paper would be helpful for researchers in developing efficient decision-making tools, as
one need to be well acquainted with the applicability of ITs to liver disorders and also
which method is widely applied for what types of liver disorders. The different types of
liver disorders covered in this study are: hepatitis, liver fibrosis, liver cirrhosis, liver
cancer, fatty liver, liver disorders data set, hepatitis data set and hepatobiliary disorders
data set (Table 26). This study has figured out which ITs are widely used and vice-versa,
which ITs outperformed others in comparison and what are the attributes taken for
experiments. This study also discovers the merits and demerits (if any) of medical
systems developed using individual and integrated ITs.
The rest of this paper is organised as follows: Section 2 presents the survey on
individual intelligent techniques which include ANN, DM, FL and GA. Section 3 covers
various integrated intelligent techniques such as: ANN-CBR, ANN-DM, ANN-FL, AIS-
FL, ANN-GA, AIS-GA, ANN-PSO, CBR-DM, CBR-GA, DM-GA, DM-FL, FL-GA,
AIS-ANN-FL, ANN-CBR-RBR, AIS-DM-FL, ANN-DM-FL, ANN-DM-GA, ANN-GA-
RBR, CBR-GA-PSO, DM-FL-GA and CBR-DM-FL-GA. In both Sections 2 and 3,
information is listed in tabular form. Section 4 presents the observation. Finally,
conclusions are drawn in Section 5.
2 Individual intelligent techniques
This section briefly introduces individual intelligent techniques which are applied to liver
disorders. The survey is detailed in tabular form containing following information: author
name, year of publication, attributes, intelligent techniques and other methods used, and
result and application. Based on the literature survey, merits and demerits (if any) of IT-
based proposed systems are also stated. Individual ITs discussed in this segment are
ANN, DM, FL and GA.
2.1 Artificial Neural Networks (ANNs)
Artificial neural networks are a simulated view of human brain that is composed of
artificial neurons or nodes. These neurons are made up of highly interconnected and
interacting processing units (Lin and Chuang, 2010). The first design of ANNs was given
by physiologists, McCulloch and Pitts in the year 1943. The basic structure of ANN
30 A. Singh and B. Pandey
consists of input, hidden and output layers, which collectively work as a neuron of a
human brain. Neurons in ANN communicate with each other with the help of impulses.
These impulses could be of dual nature such as electrical or chemical. It works by
acquiring raw data from the outer world for generalising the knowledge. ANNs have an
immense ability to learn and derive meaning from intricate and imprecise data. It does
not require any preceding knowledge of a problem. ANN has impactful applications in
various fields such as pattern recognition, time series prediction, data processing,
robotics and regression analysis. Apart from these fields, ANN has also spread its reach
in medicine too. It has emerged as one of the most popular tools and provides promising
results for medical data analysis. In medicine, ANN helps in diagnosis, radiology,
medical image analysis, etc. Limitations of ANN are as follow: it requires lots of
knowledge intake, ANN-based systems take lots of time to get fully trained and at time it
is difficult to find adequate solution to a problem. Apart from this, choosing an
appropriate knowledge set is yet again a major challenge.
ANNs have fascinated many researchers and have shown remarkable performance
when applied to liver disorders. ANN-based systems are reliable, robust, more accurate,
predictive, computationally simple, non -invasive and inexpensive (Ozyilmaz and
Yildirim, 2003; Azaid et al., 2006; Bucak and Baki, 2010; Hashem et al., 2010; Icer et
al., 2006; Autio et al., 2007). ANN speeds up the learning process and solves fast size-
growing problem (Lee et al., 2005). Levenberg-Marquardt training algorithm of
multilayer perceptron (MLP) network employing back propagation shows fair prediction
and obtains lower mean square errors (Icer et al., 2006). Implementation of MLP
networks trained by resilient back propagation algorithm is good in improving
classification accuracy of small classes (Autio et al., 2007). Elizondo et al. (2012)
proposed a method that detects differences in the complexity of classification problem.
Sun et al. (2005) deployed fast discrete wavelet transform for decreasing time
consumption in computation. Some limitations of ANN-based systems are: it is difficult
to explain complex classification process as rules (Lee et al., 2005), classification
accuracy for MLP is low (Ozyilmaz and Yildirim, 2003), Self-Organising Map (SOM) is
unreliable to diagnose hepatitis virus (Ansari et al., 2011), implementation of pyramid
neural network consumes a bit longer time in processing (Sun et al., 2005).
Hayashi et al. (2000) stated that overall accuracy rates obtained from NeuroLinear
and NeuroRule were higher than those of Linear Discriminant Analysis (LDA) and
Fuzzy Neural Networks (FNNs). Ozyilmaz and Yildirim (2003) found the accuracy of
Conic Section Function Neural Network (CSFNN) was higher than C4.5 decision tree,
Naive Bayes classifier, Bayesian network with naive dependence and feature selection
(BNNF). Ansari et al. (2011) asserted that supervised model performs better as compared
to unsupervised one. Perez et al. (2012) proposed an Associative Memory-Based
Classifier (AMBC) method that achieves the highest classification accuracy among the
methods such as AdaBoostM1, Bagging, BayesNet and Logistic. Generalised Regression
Neural Network (GRNN) (Ansari et al., 2011) performed better than Feedforward Back
Propagation Neural Network (FFNN). Revesz and Triplet (2010) compared classification
of integration and data integration methods, both used Support Vector Machine (SVM)
linear classifier, and found that the former was more accurate than latter in case of
missing values in data. Two-Level Neural Network (2-LNN) method (Hayashi and
Intelligent techniques and applications in liver disorders 31
Setiono, 2002) achieved higher predictive accuracy than FNN and Fuzzy Multilayer
Perceptron (FMLP). ANN (Hashem et al., 2010) attained better classification accuracy
than decision tree and Multivariate Logistic Regression Analysis (MLRA). ANN
(Hashem et al., 2010) performed better than MLRA in simulating non-linear relation
between fibrosis grades and biomarkers. Radial Basis Functions (RBFs) network
(Rouhani and Haghighi, 2009) outperformed all networks including GRNN, learning
Vector Quantisation Network (LVQ), PNN and SVM, except for the one class (hepatitis
B) in which the Probabilistic Neural Networks (PNN) performed better. SVM-SA
(simulated annealing) based system (Sartakhti et al., 2012) achieved better accuracy than
methods such as C4.5, Naive Bayes (NB), LDA, LVQ, GA-SVM, MLP, GRNN, and
MLP with BP (back propagation). The survey on applicability of ANNs for liver
disorders is listed in Table 1.
Table 1 Details of ANN-based systems with their results and applications
Author, Year Attributes Intelligent techniques and
other methods Result and application
Hamamoto
et al. (1995)
Preoperative aspartate
aminotransferase, alanine
aminotransferase, alkaline
phosphatase, total
bilirubin of the serum,
hepaplastine test, ICGR15,
total liver volume,
residual liver volume and
number of platelet
Perceptron-type neural
network, linear regression
method, supervised
learning based on back
propagation method
Prediction of early
prognosis of
hepatectomised patient
with hepatocellular
carcinoma (liver cancer).
Accuracy: 100%
Hayashi et al.
(2000)
Hepatobiliary disorders
data set
Standard feedforward
network, line search
algorithm, quasi Newton
algorithm, BFGS method,
NeuroLinear and
NeuroRule rule extraction
techniques
Diagnosis of hepatobiliary
disorders.
Accuracy: NeuroRule –
88.3%
Neuro Linear – 90.2%
Hayashi and
Setiono
(2002)
Hepatobiliary disorders
data set
Standard feedforward
network with a single
hidden layer
Diagnosis of hepatobiliary
disorders.
Accuracy of 2-LNN:
83.47% (using best choice
criterion) and 91.41%
(using second best choice
criterion)
Ozyilmaz and
Yildirim
(2003)
Hepatitis data set MLP trained with standard
back propagation
algorithm, RBF trained
with OLS algorithm,
CSFNN combined MLP
and RBF, Gaussian bell
function
Diagnosis of hepatitis
disease.
Accuracy: CSFNN – 90%
RBF – 85%
MLP – 81.375%
32 A. Singh and B. Pandey
Table 1 Details of ANN-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques and
other methods Result and application
Lee et al.
(2005)
Contour of the liver cyst,
contrast between liver
tissues, grey levels of the
liver tissues
BP-CMAC neural network
(integrated with BP and
CMAC)
Classification of liver
disorders (liver cyst,
hepatoma and cavernous
haemangioma).
Accuracy: 87%
Malal and
Sadasivam
(2005)
Computerised
tomography images
Probabilistic neural
network.
Wavelet-based texture
analysis: orthogonal
wavelet transform, mean,
standard deviation,
contrast, entropy,
homogeneity and angular
second moment
Classification of diffused
liver disorders (fatty liver
and liver cirrhosis).
Accuracy: 95%
Sensitivity: 96%
Specificity: 94%
Sun et al.
(2005)
Ultrasonographicimages
of cirrhosis
Pyramid neural network
trained using
ultrasonographic images of
cirrhosis and using data
judged by clinicians, fast
discrete wavelet transform,
steepest descent method
Diagnose the type of
cirrhosis diseases.
Azaid et al.
(2006)
Ultrasound images: mean
grey level, variance of
grey levels, skewness of
grey level distribution,
kurtosis
Multi-layer back
p
ropagation neural network
trained on features (mean
grey level, variance of grey
level, skewness of grey
level and kurtosis),
quantitative tissue
characterisation technique,
square-shaped region
technique
Classify liver disorders as
fatty liver, liver cirrhosis,
liver cancer.
Accuracy: 96.125%
Revett et al.
(2006)
Case number, days since
registration, drug, age at
initial registration, sex,
days between study
enrolment and a visit,
presence of ascites,
presence of
hepatomegaly, presence
of spiders, presence of
oedema, serum bilirubin,
serum cholesterol,
albumin, alkaline
phosphatase, SGOT,
platelets, prothrombin
time, histological stages
of disease
Probabilistic neural
network, approach based
on Bayes formula, Taylor’s
polynomial approximation.
Rough Sets: Rosetta
implementation, entropy
preserving or MDL
(Minimal Description
Length) algorithm,
equivalence classes,
genetic algorithm-based
search technique
Mining a primary biliary
cirrhosis data set.
Accuracy: 87%
Intelligent techniques and applications in liver disorders 33
Table 1 Details of ANN-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques and
other methods Result and application
Icer et al.
(2006)
Power Spectral
Densities (PSD) of
Doppler signals
Feed forward multi-layer
perceptron network, sigmoid
transfer functions, training
algorithms adopted were
resilient propagation
algorithm (RP), scaled
conjugate gradient algorithm
(SCG) and Levenberg–
Marquardt algorithm (LM)
employing back propagation,
power spectral densities
(PSD) of portal venous
Doppler signals, short time
Fourier transform (STFT)
method
To determinate cirrhosis
diseases with power
spectral densities of portal
venous Doppler signals.
Accuracy: sensitivity and
specificity was 100% with
Levenberg–Marquardt
training algorithm
Autio et al.
(2007)
Liver disorders data set Multi-layer perceptron
networks trained with
resilient back propagation
algorithm, logarithmic
sigmoid function, root mean
square formula, least gradient
technique, tenfold cross-
validation
Classification of liver
disorders as sick and
healthy.
Accuracy: 71%
Dong et al.
(2008)
Liver disorders data set Support vector machines,
tenfold cross-validation
To calculate optimal value
of cost parameter in order to
minimise classification
error.
Accuracy: 68.12%
Su and Yang
(2008)
Liver disease data set
collected from
department of health
examination, Chang
gung memorial
hospital, Tao-Yuan,
Taiwan
Support vector machine
model, polynomial kernel,
Gaussian radius base function
kernel and combined kernel
functions, L-J method for
feature selection
Classification of liver
disorders.
Accuracy (kP.G): 77%
(with 100% features)
Rouhani and
Haghighi
(2009)
Sex, age, ALK, AST,
SGOT, ALT, SGPT, Bi,
T, Bi, D, G.G.T,
HBSAg, Alb, LHD, PT,
FBS, CHO and HCVAb
RBF networks: two-layer
structure, linear activation
function, Gaussian function
GRNN: radial basis layer and
a special linear layer
PNN: structure was alike
RBF networks, Gaussian
distribution, competitive
transfer function
LVQ networks: competitive
layer and a linear layer
SVM: polynomial kernel
function
Diagnosis of hepatitis
disease.
Accuracy (RBF): 96.4%
34 A. Singh and B. Pandey
Table 1 Details of ANN-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques and
other methods Result and application
Uttreshwar
and Ghatol
(2009)
Hepatitis B surface antigen
(HBsAg), hepatitis B surface
antibody (HBsAb), hepatitis
B e-antigen (HBeAg),
hepatitis B DNA
Generalised regression
neural networks, kernel-
based approximation,
logical inference, IF-Then
rules
Diagnosis of hepatitis
B.
Accuracy: 86.3237%
Bucak and
Baki (2010)
AST, ALT, AST/ALT,
albumin, protein, platelet,
and prothrombin time
CMAC neural network,
supervised learning,
quantisation, least mean
square (LMS)
Diagnosis of liver
disorders (hepatitis B,
hepatitis C, cirrhosis A,
cirrhosis B and C).
Accuracy: 100%
Hashem et al.
(2010)
Routine work tests: platelets
count, haemoglobin, WBCs,
RBCs, alanine
aminotransferase, aspartate
aminotransferase, alkaline
phosphatase, serum albumin,
total bilirubin, prothrombin
concentration, alfa-
fetoprotein, thyroid
stimulating hormone,
creatinine, urea, and blood
glucose.
Fibrotic markers: matrix
metalloproteinase-1,
metalloproteinase-2,
hyaluronic acid, tissue
inhibitor of
metalloproteinase-1, tissue
growth factor beta1, α2-
macroglobulin, haptoglobin,
apolipoprotineA1
ANN: simulate non-linear
relation between fibrosis
grades and biomarkers
Analysis of variance
(ANOVA), Tukey–Kramer
and Bonferroni multiple
comparison tests,
sequential R-square
measure, box plots
Prediction of the degree
of liver fibrosis (predict
the hepatic fibrosis
extent in patients with
HCV).
Accuracy: 93.7%
Sensitivity: 92.5%
Specificity: 94.8%
Area under ROC curve:
0.974
Revesz and
Triplet
(2010)
Case number, days between
registration and earliest of
death, transplantion or study;
age in days, gender, ascites
present, hepatomegaly
present, spiders present,
oedema, serum bilirubin,
serum cholesterol, albumin,
urine copper, alkaline
phosphatase, SGOT,
triglicerides, platelets,
prothrombin time in seconds,
status, drug, histological
stage of the disease
SVM, linear kernel, used
SVM implementation from
SVMLib
Liver disorder
classification (primary
biliary cirrhosis)
Intelligent techniques and applications in liver disorders 35
Table 1 Details of ANN-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques and
other methods Result and application
Ansari et al.
(2011)
Hepatitis data set FFNN: Levenberg–
Marquardt back
propagation algorithm,
mean square error (MSE)
formula.
GRNN: kernel function,
radial basis transfer
function, linear transfer
function, Euclidean
distance weight function.
SOM: competitive learning
approach, Euclidean
distance, link distance
function.
Diagnosis of hepatitis virus.
Accuracy: FFNN-91.3%,
GRNN-92%
Arsene and
Lisboa
(2012)
Time, triglicerides,
SGOT, serum
cholesterol, alkaline
phosphatase, ascites,
platelets, urine copper,
spiders, bilirubin,
albumin, age, gender,
presence of oedema,
prothrombin time,
hepatomegally,
histological stage of
disease, drug
Bayesian neural network,
partial logistic artificial
neural network (PLANN)
and automatic relevance
determination (ARD),
Bayesian regularisation
framework, hessian matrix
of the total error function,
local and a global
compensation mechanism
Medical survival analysis of
primary biliary cirrhosis
(PBC)
Perez et al.
(2012)
Liver disorders data set
and Hepatitis data set
Associative memory based
classifier (AMBC):
learning phase, learning
reinforcement phase and
classification phase,
Integer To Vector operator.
Cross-validation: 50–50
training-test split, 70–30
training-test split, tenfold
cross-validation and leave-
one-out cross-validation
Diagnosis of liver disorders.
Classification accuracy using
50–50 training-test split:
BUPA – 65.40%
Hepatitis – 83.76%
Classification accuracy using
70–30 training-test split:
BUPA – 59.593%
Hepatitis – 84.86%
Classification accuracy using
10 fold cross-validation:
BUPA – 65.50%
Hepatitis – 85.16%
Classification accuracy using
leave-one-out cross-validation:
BUPA – 60.57%
Hepatitis – 85.16%
36 A. Singh and B. Pandey
Table 1 Details of ANN-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques and
other methods Result and application
Elizondo et al.
(2012)
Hepatitis data set Recursive deterministic
perceptron (RDP) neural
network, simplex method
for testing linear
separability, ANOVA
analysis
Quantifying the level of
complexity of
classification data sets
Sartakhti et al.
(2012)
Hepatitis data set SVM, RBF kernel,
simulated annealing, k-fold
cross-validation
Hepatitis diseases
diagnosis.
Accuracy: 96.25%
Babu and
Suresh (2013)
Liver disorders data set PBL-McRBFN: cognitive
component and meta-
cognitive component,
Gaussian activation
function, sample delete
strategy, neuron growth
strategy, parameter update
strategy, sample reserve
strategy
Classification performance
on liver disorders data set.
Accuracy: 72.63%
Jeon et al.
(2013)
Ultrasound images:
characteristics of lesions
including internal echo,
morphology, edge,
echogenicity and
posterior echo
enhancement
SVM, multiple ROI
selection, feature
extraction, feature-level
fusion method to combine
features, tenfold cross-
validation
Focal liver lesion
classification.
Accuracy for classification
of cysts and haemangioma:
93.77%
Accuracy for classification
of cysts and malignancies:
92.13%
Accuracy for classification
of haemangioma and
malignancies: 69.38%
Accuracy for classification
of haemangioma and
malignancies: 80% (with
feature selection algorithm)
Notes: CMAC: Cerebellar Model Articulation Controller, PBL: Projection-Based
Learning, McRBFN: Meta-cognitive Radial Basis Function Network, ROI:
Region of Interest.
Liver disorders data set attributes are: MCV (mean corpuscular volume),
Alkphos (alkaline phosphotase), SGPT (alamine aminotransferase), SGOT
(aspartate aminotransferase), gamma-GT (gamma-glutamyl transpeptidase),
drinks (number of half-pint equivalents of alcoholic beverages drunk per day).
Hepatitis data set attributes are: age, sex, steroid, antivirals, fatigue, malaise,
anorexia, liver big, liver firm, spleen palpable, spiders, ascites, varices,
bilirubin, alk phosphate, sgot, albumin, protime, histology.
Hepatobiliary disorders data set attributes are: Glutamic Oxalacetic
Transaminate (GOT, Karmen unit).
Glutamic Pyruvic Transaminase (GPT, Karmen Unit), Lactate Dehydrase
(LDH, iu/l), Gamma Glutamyl Transpeptidase (GGT, mu/ml), Blood Urea
Nitrogen (BUN, mg/dl), Mean Corpuscular Volume of red blood cell (MCV,
fl), Mean Corpuscular Haemoglobin (MCH, pg), Total Bilirubin (TBil, mg/dl)
and Creatinine (CRTNN, mg/dl).
Intelligent techniques and applications in liver disorders 37
2.2 Data mining (DM)
Data mining is a process of identifying hidden relationships and discovering new
knowledge from large data sets. The term ‘data mining’ was originated in the year 1990,
but work on this started a bit earlier. DM helps in processes such as classification,
clustering, regression and summarisation, which generate hidden facts from historical
data. Some issues in DM which needs to be taken care are: quality of data to mine, the
extent up to which data needs to be cleaned, and interoperability (data from
heterogeneous sources needs to combine and analyse). DM-based system has the
capability of replacing liver biopsy in liver disorders diagnosis (Floares, 2009). Yan et
al.’s (2008) C4.5 decision tree-based proposed method can be efficiently integrated with
algorithm like boosting to enhance prediction. Eastwood and Gabrys (2012) mentioned
advantages of a single tree classifier: simple model structure, small memory requirement
and quick calculation of predictions. Kohara et al. (2010) has done something very
interesting and out of the box by proving the feasibility to diagnose liver cirrhosis using
PCA-based statistical shape model of the liver.
In comparison, Yan et al. (2008) proposed C4.5 algorithm that has better
classification rate than other methods such as ID3 decision tree, RBF NN, BayesNet and
logistic. Based on the results achieved, Yan et al. (2008) claimed that if a patient has
gotten cirrhosis, he must has symptoms like lassitude and fatigue, chill and cold limbs,
tarnish complexion, yellow eyes or yellow body or yellow urine. Floares (2009)
developed a C5.0 decision tree and boosting-based system which has outperformed other
methods such as SVMs, Bayesian networks, neural networks of various types and
architectures, and classification and regression trees. The survey on applicability of data
mining techniques for liver disorders is listed in Table 2.
2.3 Fuzzy logic (FL)
FL-based models have been developed and utilised by numerous researchers for handling
liver disorders. The concept of fuzzy logic was introduced by Lotfi A. Zadeh in the year
1965 and was applied in medical systems approximately after 20 years. FL employs
linguistic rules in the form of IF-Then statements. FL deals with uncertainty and also
assists computer in interpreting statements which consists of intermediate constructs. For
example, if glass is half-full then pour some water. The semantic of this statement does
not correspond to any truth value, either true/false. FL-based systems are faster, liable,
cheaper (Neshat et al., 2008), robust, flexible, customisable, interpretable and easy to
train (Gadaras and Mikhailov, 2009). FL-based system also has flexible initialisation, fast
convergence and robust segmentation (Li et al., 2012). FL-based system efficiently deals
with uncertainty, ambiguous information and imprecise data (Obot and Udoh, 2011).
Ming et al.’s (2011) fuzzy-based framework uses global k-means algorithm to determine
actual number of cluster needed for different data sets and fast global k-means to
improve computation time taken by global k-means algorithm. This model was based on
enhanced supervised fuzzy clustering algorithm, which effectively handles small size
data that is noisy and atypical. Sometimes fuzzy-based systems also require more
simulation and fine tuning.
38 A. Singh and B. Pandey
Table 2 Details of DM-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Yan et al.
(2008)
Lassitude and fatigue, chill
and cold limbs, tarnish
complexion, yellow eyes or
yellow body or yellow urine
C4.5 decision tree,
tenfold cross-
validation
To analyse relationship
between child-pugh degree
and examinations of
traditional Chinese
medicine based on liver
cirrhosis.
Accuracy: 85.67 % (for
child pugh A)
Floares
(2009)
Age, aspartate
aminotransferase,
gamma-glutamyl-
transpeptidase, cholesterol,
triglycerides,
thickness of the gallbladder
wall, spleen area and
perimeter, left lobe and
caudate lobe diameter, liver
homogeneity, posterior
attenuation of the ultrasound,
liver capsule regularity, spleen
longitudinal diameter, the
maximum subcutaneous fat,
perirenal fat
C5.0 decision tree and
boosting (AdaBoost)
Liver disorders
classification (chronic
hepatitis C and B).
Accuracy: 100%
Kohara et al.
(2010)
Components of shape feature
vector
Principal component
analysis (PCA),
marching cube
algorithm, Chui
method
Diagnosis of liver cirrhosis
Luo et al.
(2011)
Jaundice, poor appetite,
fatigue, yellow urine and
hypochondriac pain
Cluster analysis:
DBScan algorithm
Association rules:
Apriori algorithm
Preventing and treating viral
hepatitis
Eastwood
and Gabrys
(2012)
Liver disorders data set Standard decision tree
induction, re-sampling
(bootstrapping), linear
discriminant analysis,
model level
combination method,
pessimistic pruning
and error-based
pruning, 10 × 10fold
cross-validation
Proposed pruning criteria
(Liver disorders UCI
database for the empirical
investigation of proposed
method).
Jen et al.
(2012)
Systolic pressure, diastolic
pressure, glutamate-pyruvate
transaminase, alpha-
fetoprotein
K-nearest neighbour,
linear discriminant
analysis with
sequential forward
selection (a bottom-up
search procedure)
Used risk factors of chronic
diseases (disease of the
liver) to build early warning
criteria.
Accuracy: 82.65%
Intelligent techniques and applications in liver disorders 39
Based on the comparisons, Badawi et al. (1999) proposed fuzzy-based classification that
attains higher sensitivity than neural network classification, and higher sensitivity and
specificity than statistical classification techniques. Obot and Udoh (2011) stated that FL
ability to work from approximate reasoning and finding precise solution makes it
superior to other methods such as ANN, RBR and CBR. Neshat et al. (2008) proposed
fuzzy system that obtains higher accuracy than other traditional diagnostic systems such
as RULES-4, C4.5, Naive Bayes, BNND, BNNF, SVM with GP, SSVM, RSVM, MLP,
PNN, GRNN, RBF, AIRS and FW-AIRS. Gadaras and Mikhailov (2009) proposed fuzzy
classification framework that achieves higher accuracy than other techniques mentioned
in the literature such as FBP-NN, BZ, GF-SVM and NF-BSP. Ming et al. (2011)
proposed a deterministic and autonomous algorithm (enhanced supervised fuzzy
clustering) which attains higher mean accuracy than supervised fuzzy clustering method.
Luukka (2011) proposed fuzzy bean-based classifier that obtains higher accuracy than
classifiers such as CN2, MLP, DIMLP and SIM. The survey on applicability of FL for
liver disorders is listed in Table 3.
Table 3 Details of FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Badawi et al.
(1999)
Mean grey level, contrast,
angular second moment,
entropy, correlation,
attenuation and speckle
separation
Fuzzy rules, MIN
compositional rule of
inference, bell
membership function
Differentiate diffuse liver
disorders.
Results of fuzzy rule-based
classification:
Specificity: 92%
Sensitivity for liver cirrhosis:
94%
Sensitivity for fatty liver:
96%
Neshat et al.
(2008)
Liver disorders data set Fuzzy rules, triangular
or trapezoidal fuzzifier,
centre of gravity
defuzzifier formula
Liver disorders diagnosis
(healthy and unhealthy
liver).
Accuracy: 91%
Gadaras and
Mikhailov
(2009)
Liver disorders data set Fuzzy rules, min-max
method, trapezoid
membership function
Classification performance
on liver disorders data set.
Accuracy: 89.9%
Luukka
(2011)
Liver disorders data set Fuzzy beans, Bocklisch
membership function,
differential evolution
algorithm
Liver disorders diagnosis.
Accuracy: 73.9%
Ming et al.
(2011)
Hepatobiliary disorders
data set
Enhanced supervised
fuzzy clustering
algorithm, k-means
algorithm, fast global k-
means, unsupervised
Gath–Geva algorithm
Liver disorders classification
(alcoholic liver damage,
primary hepatoma, liver
cirrhosis and cholelithiasis).
Accuracy: 58.78%
40 A. Singh and B. Pandey
Table 3 Details of FL-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Obot and
Udoh (2011)
Nausea, vomiting, fever,
body weakness, loss of
appetite, diarrhoea, itching,
convulsion, stupor,
headache, tremors, skin
discoloration, eye
discoloration, liver
tenderness, bile in urine,
jaundice
Fuzzy rules, max-min
method, centre of
gravity (CoG) method,
fuzzified with
membership functions
Diagnosis of hepatitis
Li et al.
(2012)
Contrast-enhanced
computed tomography
images
Unsupervised fuzzy
clustering, fuzzy c-
means
Semi-automatic liver tumour
segmentation
2.4 Genetic algorithm (GA)
The genetic algorithm was proposed in 1975 by John Holland (Ozsen and Gunes, 2009)
at University of Michigan, USA. GA is a branch of evolutionary algorithm which
imitates the process of natural evolution and survival of the fittest (Zhang and Rockett,
2011). GA finds the optimum solution from the set of candidate solutions. GA uses a
fixed-length chromosome structure and is aimed at solving optimisation or search
problems. The basic requirements of GA are a genetic representation of solution set and a
fitness function to test and evaluate the solution set. Primary genetic operators used by
GA are selection, crossover and mutation. GA has several limitations, which are:
implementation of fitness function to evaluate the solution set is quite expensive, shows
less efficiency as the complexity of the problem increases and does not operate well on
dynamic data sets. Despite this, usage of GA for optimising parameters makes the
diagnostic systems robust and invariant (Tan et al., 2003). Falco (2013) proposed a tool
that extracts knowledge in the form of IF-Then rules from databases. This is simple,
faster, robust, reliable and easy to implement. It also helps users in medical diagnosis and
gives explanation of evidences on why a patient is suffering from a specific disease.
In comparison, Tan et al. (2003) proposed a two-phase hybrid evolutionary
classification technique which performed better than methods such as C4.5 (decision tree
program), PART (rule-learning scheme) and is comparable to Naive Bayes (utilises the
Bayesian techniques). Zhang and Rockett (2011) proposed feature extraction method that
proves its superiority to competitive methods such as RBF, logistic (modified
multinomial logistic regression model), nearest-neighbour-like algorithm, Bayes network
classifier using K2 learning algorithm, instance-based learning algorithm, ADTree (the
alternating decision tree learning algorithm), Sequential Minimal Optimisation (SMO)
algorithm and C4.5 decision tree algorithm. This method also records the lowest mean
error. Wu et al. (2012) presented a GA-based feature selection algorithm, which selects a
better feature subset than serial feature combination and serial feature fusion schemes.
This algorithm performs better, in selecting feature subsets, than NMIFS (Normalised
Mutual Information Feature Selection) and GAMIFS (a hybrid filter/wrapper method
Intelligent techniques and applications in liver disorders 41
called GAMIF). Falco (2013) proposed a tool that is superior to Bayes Net, Naive Bayes,
IB 1, FLR (Fuzzy Lattice Reasoning), VFI (Voting Feature Interval), OneR, Part, and
inferior to MLP, RBF, KStar, AdaBoostM1, Bagging, Ridor (ripple down rule), J48,
NBTree. It also requires the lowest number of rules in comparison to other rule-based
classification methods (Part, OneR and Ridor). In spite of these excellent features, this
tool has a limitation of not taking uncertainty into account. The survey on applicability of
GA for liver disorders is listed in Table 4.
Table 4 Details of GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques and other
methods Result and Application
Tan et al.
(2003)
Hepatitis data set Tournament selection scheme,
genetic programming tree-based
chromosome representation, two
Boolean operators (AND and
NOT) were adopted, ramped-half-
and-half approach, fixed-length
real-coding chromosome
structure, standard tree-based
crossover and mutation operators,
standard single-point crossover,
covering algorithm, knowledge
presented as multiple IF-Then
rules, Michigan coding approach,
Pittsburgh coding approach,
Pittsburgh-like approach, paired t-
test
Predict whether a patient
with hepatitis will live or die.
Average Accuracy: 83.92%
Best Accuracy: 94.34%
Zhang and
Rockett
(2011)
Liver disorders
data set
Binary tournament selection,
depth-fair operator, roulette wheel
selection, non-destructive, depth
dependent crossover and mutation
operators, minimisation of
vectors, three-dimensional fitness
vector compromising tree
complexity, misclassification
error and Bayes error
Classification performance
on liver disorders data set
Wu et al.
(2012)
Ultrasonic liver
image data set
Two-point crossover and
mutation, roulette wheel selection
scheme, k-nearest neighbour
method, threefold cross-validation
Ultrasonic liver tissue
characterisation (cirrhosis,
hepatoma, and normal).
Accuracy: 96.62 %
Falco (2013) Liver disorders
data set
Differential evolution method,
tenfold cross-validation
mechanism
Automatic classification of
items in medical databases.
Accuracy in case of liver
disorders data set: 64.74%
Specificity: 45.08%
Sensitivity: 79.84%
ROC curve area: 62.46
42 A. Singh and B. Pandey
3 Integrated intelligent techniques
This section presents the survey results of integrated intelligent techniques applied to
liver disorders. The study is detailed in tabular form containing following information:
author name, year of publication, attributes, intelligent techniques and other methods
used, and result and application. Integrated ITs combine methods in one of the two ways:
either the techniques are applied sequentially, in which one technique is used to
accomplish a specific task that is followed by second technique and so on, or all the
techniques are applied simultaneously. For example, sometimes integration of both ANN
and CBR is used to identify the existence of liver disorders; whereas in integration ANN
is used to identify the existence of liver disorders and CBR is used to find the types of
liver disorders. It could also be possible that researchers might have used some methods
that are not considered for this survey but we have mentioned those methods, in the table,
wherever possible. This section also enlightens the benefits of integrated IT-based
systems when used for liver disorders. Integrated intelligent techniques focused in this
segment are: ANN-CBR, ANN-DM, ANN-FL, AIS-FL, ANN-GA, AIS-GA, ANN-PSO,
CBR-DM, CBR-GA, DM-GA, DM-FL, FL-GA, AIS-ANN-FL, ANN-CBR-RBR, AIS-
DM-FL, ANN-DM-FL, ANN-DM-GA, ANN-GA-RBR, CBR-GA-PSO, DM-FL-GA and
CBR-DM-FL-GA.
Table 5 Details of ANN-CBR-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Lin and
Chuang
(2010)
Hepatitis test: HBsAg,
HBeAg, Anti-HBs, Anti-
HBe, Anti-HBc Anti-HCV
Liver function test: AST
(SGOT), ALT, T-Bil, ALB,
ALP, r-GT.
Tumour marker: α-AFP
Basic information: gender,
marriage, blood type, age,
education, occupation.
Lifestyle habit: tattoo,
smoking, chewing betel-nut,
alcohol.
Lifestyle: Fatigue, sleep,
nap, exercise, breakfast
habit, vegetables, fruits,
food date mark, food
composition, low-salt, low-
sugar.
Health condition: healthy
status, weight, response to
p
hysical discomfort, healthy
examination, acupuncture,
blood donation.
ANN: BPN trained
with gradient steepest
descent algorithm.
CBR: retrieve most
similar case, vector of
features, case indexing,
case retrieval,
assigning weights to
attributes, nearest
neighbour method.
AHP: structure
decision hierarchy, pair
wise comparisons,
initiate prioritisation
evaluated consistency,
means of a consistency
ratio, compute relative
weights, geometric
mean.
Fivefold cross-
validation
To examine the existence of
liver disorders and to
determine the types of liver
disorders.
Accuracy:
ANN (diagnosis of liver
disorders): 98.04%
AHP-weighted CBR
(discovers the types of liver
disorders): 94.57%
Types of liver disorders:
90.2% for chronic hepatitis,
19.6% for liver cirrhosis,
60.2% for B hepatitis and
10% for alcohol hepatitis
Intelligent techniques and applications in liver disorders 43
Table 5 Details of ANN-CBR-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Chuang
(2011)
Hepatitis test: HBsAg,
HBeAg, Anti-HBs, Anti-
HBe, Anti-HBc Anti-HCV.
Liver function test: AST
(SGOT), ALT, T-Bil, ALB,
ALP, r-GT.
Tumour marker: α-AFP
Basic information:
Gender, marriage, blood
type, age, education,
occupation.
Lifestyle habit: Tattoo,
smoking, chewing betel-nut,
alcohol.
Lifestyle: Fatigue, sleep,
nap, exercise, breakfast
habit, vegetables, fruits,
food date mark, food
composition, low-salt, low-
sugar.
Health condition: Healthy
status, weight, response to
p
hysical discomfort, healthy
examination, acupuncture
blood donation.
ANN: BPN
implemented using
NeuroShell 2.0,
gradient steepest
descent training
algorithm.
CBR: Euclidian
distance to extract
similar cases, nearest
neighbour method.
Tenfold cross-
validation, sampling
using the Bernoli
method
Liver disorders diagnosis.
Accuracy:
BPN-CBR:
Accuracy: 95%
Sensitivity: 98%
Specificity:94%
AUC: 96%
BPN:
Accuracy: 93%,
Sensitivity: 91%
Specificity: 96%
AUC: 93%
CBR:
Accuracy: 89%
Sensitivity: 90%
Specificity: 88%,
AUC: 89%
Notes: BPN: Back propagation neural network, AHP: Analytic hierarchy process.
3.1 ANN-CBR
ANN-CBR methodology was used by Lin and Chuang (2010), where ANN is deployed
to examine the existence of liver disorders and AHP-weighted CBR is used to discover
the types of liver disorders, and by Chuang (2011), where ANN-CBR integration is
deployed to obtain enhanced accuracy in diagnosis. The integration of ANN-CBR makes
the diagnosis more accurate and comprehensive (Chuang, 2011), decreases the
occurrence of false diagnosis and avoids postponement of treatment (Lin and Chuang,
2010). Chuang (2011) made it evident that proposed ANN-CBR model achieves better
diagnostic accuracy than BPN (back propagation neural network), CART (classification
and regression tree), DA (discriminatory analysis), LR (logistic regression), CBR, LR-
CBR, DA-CBR, and CART-CBR. Lin and Chuang (2010) used AHP-weighted CBR
instead of CBR because it reduces diagnostic errors, accelerates the medical treatment
and most importantly has obtained better accuracy. AHP allocates weights to the
attributes. One appealing fact in this study (Lin and Chuang, 2010) is identifying types of
liver disorders as most of the literature work had not moved beyond diagnosis. The
survey on applicability of ANN-CBR for liver disorders is listed in Table 5.
44 A. Singh and B. Pandey
3.2 ANN-DM
ANN-DM methodology was used by Bologna (2003), where DM is used for extraction of
rules and ANN is used for classification, and by Calisir and Dogantekin (2011), where
DM is used for feature extraction and feature reduction and ANN is used for
classification. ANN-DM-based systems in medical domain have reliability, more
accuracy, small-sample problem-solving ability, correct recognition rates, simple
structure and good generalisation (Bologna, 2003; Calisir and Dogantekin, 2011; Hashem
et al., 2012). PCA-LSSVM method (Calisir and Dogantekin, 2011) achieved higher
accuracy than methods such as Weighted9NN, 18NN, ASI, MLP+BP (Tooldiag), LDA,
MLP, RBF (Tooldiag), 1NN, RBF, FS-AIRS, 15NN, FSM with rotations, FSM without
rotations, MLP with BP, QDA, Naive Bayes, Fisher discriminant analysis, LVQ, GRNN,
ASR, IncNet, CART (decision tree), PCA-AIRS, and LFC. Bologna (2003) proposed
DIMLP model that is appreciably more accurate than CN2 induction algorithm on most
of the problems. The survey on applicability of ANN-DM for liver disorders is listed in
Table 6.
Table 6 Details of ANN-DM-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Bologna (2003) Hepatitis data set
and Liver disorders
data set
ANN: discretised interpretable
multi-layer perceptron
(DIMLP) model, staircase
activation function, weighted
sum of inputs and weights, sum
squared error (SSE) function,
gradient was determined using
sigmoid functions, trained by
back propagation with default
parameters, bagging and arcing
methods based on re-sampling
techniques, relevance
hyperplane criterion.
DM: C4.5 decision trees, IF-
Then rules.
Tenfold cross-validation, t-
statistic test, two tailed test
Diagnosis of liver
disorders.
Average predictive
accuracy:
Hepatitis: 79.1%
Liver disorders: 70.15%
Calisir and
Dogantekin
(2011)
Hepatitis data set ANN: least square support
vector machine (LSSVM)
classifier, maximum Euclidean
distance, parameters includes
width of Gaussian kernels and
regularisation factor.
DM: principle component
analysis (PCA)
Diagnosis of hepatitis
diseases.
Accuracy: 96.12%
Intelligent techniques and applications in liver disorders 45
Table 6 Details of ANN-DM-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Hashem et al.
(2012)
HA, TGF-β1, α2-
macroglobulin,
MMP2, ApoA1,
urea, TIMP, MMP1
and haptoglobin
Single-stage classification
model:
ANN: tangent sigmoid transfer
function.
DM: decision tree, chi-square,
entropy reduction, Gini
reduction splitting criteria.
Multivariate logistic regression
analysis (MLRA), nearest
neighbourhood (KNN).
A multistage stepwise
classification model:
ANN: linear, tangent sigmoid
and logarithmic sigmoid, back
propagation algorithms
(gradient descent, gradient
descent with momentum,
conjugate-gradient, quasi-
Newton and Levenberge–
Marquardt), mean square error).
DM: decision tree, entropy,
Gini index, chi-square test
support and confidence
measures.
Multivariate logistic regression
analysis, likelihood ratio test,
Hosmer and Lemeshow
Chisquare goodness of fit tests,
variance inflationary factor
(VIF) test
Prediction of liver fibrosis
degree in patients with
chronic hepatitis C
infection.
Accuracy:
Single staged model:
82.8% (training), 71.2%
(testing)
Multistage model: 85.6%
(testing), 81.9% (training)
3.3 ANN-FL
ANN-FL has emerged as one of the highly used integration method applied to liver
disorders. The literature work shows integration of ANN-FL as one of the best model,
which has several benefits such as enhanced accuracy (Comak et al., 2007), flexibility,
improved decision ability (Neshat and Zadeh, 2010), robustness (Dogantekin et al., 2009;
Celikyilmaz et al., 2009) and simplicity and clarity (Ceylan et al., 2011). This integration
also makes the system reliable, rapid, more accurate, easy to operate, non-invasive, more
economical and more efficient (Neshat and Zadeh, 2010; Ceylan et al., 2011; Dogantekin
et al., 2009; Celikyilmaz et al., 2009). ANN-FL methodology was used by Comak et al.
(2007), where FL is used to pre-process liver disorders data set and ANN is used to
classify, by Dogantekin et al. (2009), where fuzzy inference system-based ANN is used
for classification, by Li et al. (2010), where FL is used to compute new attribute values
and ANN is used to classify, by Ceylan et al. (2011), where FL is used to reduce the
number of segments in training pattern and ANN is used for classification, by Neshat and
46 A. Singh and B. Pandey
Zadeh (2010), where FL is used for clustering and ANN is used for classification, by Li
and Liu (2010), where FL is applied to calculate the similarity of paired data for every
class and attribute and ANN is used to classify, by Celikyilmaz et al. (2009), where ANN
is used to approximate fuzzy classification function parameters of each cluster and FL is
used to classify, and by Kulluk et al. (2013), where the proposed approach extracts brief
and accurate fuzzy classification rules (FCR) from ANNs.
Neshat and Zadeh’s (2010) fuzzy Hopfield neural network approach has fast
computational power and gains better accuracy than other neural networks such as MLP,
RBF, GRNN, PNN, LVQ and Hopfield. Li et al. (2010) proved that its method is
superior to SVM and C4.5 decision tree in terms of classification accuracy. As class
imbalance problem in medical data sets diminishes the classification performance of
traditional techniques, ANN-FL-based approach (Li et al., 2010) balances the data size
by over-sampling the minority class and under-sampling the majority class. Comak et al.
(2007) proposed medical a decision-making system that attains higher classification
accuracy than those of methods mentioned in the literature include, RULES-4, C4.5,
Naive Bayes, BNND, BNNF, SVM with GP, SSVM, RSVM, MLP, PNN, GRNN, RBF,
and AIRS. Dogantekin et al. (2009) proposed an automatic diagnosis system that has
better classification performance than other methods which includes RBF, FS-AIRS with
fuzzy, FSM with rotations, FSM without rotations, MLP with BP, QDA (quadratic
discriminant analysis), Weighted9NN, 18NN, ALI, MLP+BP, LDA, MLP, RBF, 1NN,
Na Bayes and semi-NB, Fisher discriminant analysis, LVQ, GRNN, ASR, IncNet, CART
(decision tree), PCA-AIRS and LFC. Li and Liu (2010) proposed kernel on SVM that
attains improved classification accuracy than polynomial and Gaussian kernels. Kulluk et
al. (2013) proposed a fuzzy DIFACONN-miner algorithm that yields higher accuracies
than other fuzzy rule-based classification algorithms, namely 2SLAVEsum,
FRBCS_GPsum, and GP-COACHsum. It also minimises a few complexity problem. The
survey on applicability of ANN-FL for liver disorders is listed in Table 7.
Table 7 Details of ANN-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Comak et al.
(2007)
Liver disorders data
set
ANN: least square support vector
machine (LSSVM), set of linear
equations for training.
FL: fuzzy weighting pre-
processing, triangular (input and
output) membership functions,
fuzzy IF-Then rules
Diagnosing liver disorders.
Accuracy: 94.29%
Sensitivity: 95%
Specificity: 93.33%
Celikyilmaz
et al. (2009)
Liver disorders data
set
ANN: SVM, Platt’s probability
method.
FL: classical fuzzy c-means
(FCM) clustering.
Semi-non-parametric inference
mechanism, posterior
probabilities from logistic
regression (LR), three-way data
split cross validation method
Liver disorders diagnosis.
Accuracy: 77%
Intelligent techniques and applications in liver disorders 47
Table 7 Details of ANN-FL-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Dogantekin
et al. (2009)
Hepatitis data set ANN: hybrid learning algorithm (back
propagation for non linear parameters
and least square errors for linear
parameters)
FL: fuzzy IF-Then rules, bell-shaped
membership function.
Diagnosis of hepatitis.
Accuracy: 94.16%
Sensitivity: 96.66%
Specificity: 91.66%
Neshat and
Zadeh (2010)
Liver disorders
data set
ANN: Hopfield neural network,
discrete and continuous models,
sigmoid activating function.
FL: fuzzy c-means
Liver disorders
diagnosis.
Accuracy:
FHNN –92%
HNN – 88.2%
Li et al.
(2010)
Liver disorders
data set
ANN: support vector machine
classifier.
FL: Gaussian type fuzzy membership
function and α-cut to reduce data size,
mega-trend diffusion membership
function.
Tenfold cross-validation
Deal with class
imbalance problem with
medical data sets and to
enhance classification
accuracy in BUPA liver
disorders data set.
Accuracy: 86.36%
Li and Liu
(2010)
Liver disorders
data set
ANN: support vector machine, class
probability based kernel, kernel based
on Gaussian membership function,
decomposition principle, diffusion
function technique, mega-trend
diffusion technique.
FL: triangular type membership
function
Classification
performance on liver
disorders data set.
Accuracy: 70.78%
Ceylan et al.
(2011)
Doppler signals of
90 subjects (each
subject includes
40 samples)
ANN: complex-valued artificial
neural network (CVNN), complex
back propagation (CBP) algorithm,
complex-valued activation function,
real and imaginary components.
FL: fuzzy clustering, calculation of
FFT (Fast Fourier Transform) values,
FCM clustering
Liver disorders
classification (identify
liver as healthy or
cirrhosis).
Accuracy: 100%
Sensitivity: 100%
Specificity: 100%
Kulluk et al.
(2013)
Liver disorders
data set
ANN: feedforward and recurrent
ANNs, trained using differential
evolution algorithm.
FL: triangular membership function
generates fuzzy rules by touring ant
colony optimisation (TACO)
algorithm, fixed length binary
encoding scheme to represent rules.
Tenfold cross-validation, fitness
evaluation by minimum deviation
method (MDM)
Classify liver disorders.
Accuracy: 85.60%
Notes: FHNN: Fuzzy Hopfield Neural Network, HNN: Hopfield Neural Network,
FCM: Fuzzy C-Means.
48 A. Singh and B. Pandey
3.4 AIS-FL
AIS-FL methodology was used by Polat et al. (2007), where FL is used for resource
allocation and AIS is used for classification, by Mezyk and Unold (2011), where AIS is
used for induction of fuzzy rules. Polat et al. (2007) proposed a fuzzy-artificial immune
recognition system that obtained the highest classification accuracy among classifiers
such as RULES-4, C4.5, Naïve Bayes, BNND, BNNF, SVM, SSVM, RSVM, MLP,
PNN, GRNN, AIRS and RBF. It has taken less time in computation, effectively solves
problems having large dimensioned feature space and too many classes; and required
fewer resources than traditional AIRS, which makes it more beneficial. Mezyk and
Unold (2011) proved that IFRAIS method is superior to classifiers such as C4.5, Naive
Bayes, K*, Meta END, JRip, and Hyper Pipes in terms of classification accuracy. The
survey on applicability of AIS-FL for liver disorders is listed in Table 8.
Table 8 Details of AIS-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques and other
methods Result and Application
Polat et al.
(2007)
Liver disorders
data set
AIS: artificial immune recognition
system (AIRS) supervised learning
algorithm, resource competition,
clonal selection, affinity maturation,
memory cell formation.
FL: fuzzy resource allocation, IF-
Then rules.
k-nearest neighbour algorithm, k-fold
cross-validation, tenfold cross-
validation method
Liver disorders
classification.
Accuracy: 83.36%
Mezyk and
Unold
(2011)
Hepatitis data set
and Liver disorders
data set
Induction of Fuzzy Rules with an
Artificial Immune System (IFRAIS):
sequential covering algorithm and
clonal selection algorithm, fuzzy
partition inferring based on clonal
selection algorithm, only continuous
attributes were fuzzified, IF-Then
fuzzy rules, paired t-test
To assess prediction
accuracy of liver
disorders in patients.
Accuracy:
Hepatitis: 93.87%
BUPA data set: 72.34%
3.5 ANN-GA
GA is used with ANNs to enhance classification performance of medical systems
(Gorunescu et al., 2012). GA eliminates irrelevant and noisy features, which decreases
the size of network (Dehuri and Cho, 2010). Integration of ANN-GA overcomes the non-
linearity problems and solves the complexity problems of each other (Dehuri and Cho,
2010). ANN-GA-based framework has low-computational complexity and simplicity of
architecture (Dehuri and Cho, 2010). Gorunescu et al. (2012) replaced back propagation
algorithm with GA-based learning to optimise MLP’s weights. ANN-GA methodology
was used by Dehuri and Cho (2010), where GA is used to select pertinent features and
ANN is used to classify, and by Gorunescu et al. (2012), where GA is used to optimise
the ANNs synaptic weights and ANN is used to classify. Gorunescu et al. (2012)
proposed an intelligent system that attains improved accuracy, for both complete data set
Intelligent techniques and applications in liver disorders 49
and reduced data set, than other machine learning techniques accounted in the literature
including LN, PNN, RBF, 3-MLP and 4-MLP. This intelligent system is even faster and
more effective than 3-MLP and 4-MLP. Results (Dehuri and Cho, 2010) demonstrated
that proposed method named as HFLNN outperforms other competing classification
methods such as Radial Basis Function Network (RBFN) and Functional Link Neural
Network (FLNN) with back propagation learning. The survey on applicability of ANN-
GA for liver disorders is listed in Table 9.
Table 9 Details of ANN-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Dehuri and
Cho (2010)
Liver disorders data set ANN: back propagation
learning and genetic
optimisation, trigonometric
function.
GA: single point crossover
operator, mutation
operator, selection.
Twofold cross-validation,
parametric t-test, non-
parametric Wilcoxon
signed rank test
Diagnosis of liver
disorders.
Accuracy: 77.6820%
Gorunescu
et al. (2012)
Complete data set: stiffness,
sex, body mass index,
glycaemia, triglycerides,
cholesterol,
HLD cholesterol, aspartate
aminotransferase, alanin
aminotransferase, gama
glutamyl transpeptidase,
total bilirubin alkaline,
phosphatase, prothrombin
index, quiq time, prothrombin
time ratio, prolonged activ.
partial thromboplastin time,
haematids, haemoglobin,
hematocrit, medium erytrocity
volume, avg. erytrocitary
haemoglobin, avg.
concentration of haemoglobin
in a red blood cell,
thrombocytes, sideraemia,
interquartile range.
Reduced data set: stiffness,
aspartate aminotransferase,
prothrombin index,
thrombocytes, sideraemia,
interquartile range
ANN: multi-layer
perceptron architecture.
GA: crossover and
mutation.
Tandem feature selection
mechanism, statistical
procedures, and sensitivity
analysis, discriminant
function analysis, multiple
(linear) regression model
(both forward stepwise and
backward stepwise),
analysis of correlation
matrix, tenfold cross-
validation, binary
tournament selection, total
arithmetic recombination
Classify liver fibrosis
stadialisation in chronic
hepatitis C.
Accuracy:
61.16% (complete data
set)
65.21% (reduced data
set)
50 A. Singh and B. Pandey
3.6 AIS-GA
AIS-GA methodology was used by Ozsen and Gunes (2009), where GA is used to
determine the weights of attributes that gives minimum classification error and then these
weights are used in their own previously developed AIS-based system. The classification
accuracy of GA-AWAIS-based system is superior to both AWAIS and other traditional
classifiers mentioned in the literature such as Fuzzy-AIRS, AIRS, RSVM, MLP, SSVM,
SVM with GP, C4.5, GRNN, Naive Bayes, BNNF, BNND, RBF, RULES-4 and PNN.
The survey on applicability of AIS-GA for liver disorders is listed in Table 10.
Table 10 Details of AIS-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Ozsen and
Gunes (2009)
Liver disorders data
set
AIS: attribute
weighted artificial immune
system (AWAIS).
GA: single point crossover,
hypermutation, selection.
Tenfold cross validation
Classification performance
on liver disorders data set.
Accuracy: 85.21%
3.7 ANN-PSO
ANN-PSO methodology was used by Qasem and Shamsuddin (2011), where
TVMOPSO-based RBF networks are developed. TVMOPSO extends the algorithm to
handle multi-objective optimisation problems. TVMOPSO is simple, robust, easy to use
and easy to implement. Classification accuracy of proposed adaptive evolutionary RBF
network algorithm is superior to HMOEN_L2, HMOEN_HN and inferior to RBF
network based on MOPSO and NSGA-II algorithms. Advantages of this integration
method are: stability, consistency, simplicity, enhanced accuracy, better convergence and
small standard deviations. The survey on applicability of ANN-PSO for liver disorders is
listed in Table 11.
Table 11 Details of ANN-PSO-based systems with their results and applications
Author, Year Attributes Intelligent techniques and other
methods Result and Application
Qasem and
Shamsuddin
(2011)
Hepatitis data set ANN: RBF network, centres of
RBF network in hidden layer are
initialised from k-means
clustering algorithm, weights of
RBF network are initialised
from the least mean squared
(LMS) algorithm, crowding
distance operator.
PSO: time variant multi-
objective particle swarm
optimisation (TVMOPSO)
Diagnosis of hepatitis
diseases.
Accuracy: 82.26%
Sensitivity: 88.47%
Specificity: 41.92%
AUC: 0.652
Intelligent techniques and applications in liver disorders 51
3.8 CBR-DM
CBR-DM methodology was used by Lin (2009), where DM is used to find existence of
liver disorders and CBR is used to identify types of liver disorders. CBR participated in
problem-solving by reducing diagnostic errors and meliorating quality of treatment.
Though both CBR and DM could be used in first phase for identifying the presence of
liver disorder, but Lin (2009) choose DM technique as it obtained better results in terms
of accuracy, sensitivity and specificity. Then in second phase, CBR performed
reasonably well in identifying the types of liver disorders. The survey on applicability of
CBR-DM for liver disorders is listed in Table 12.
Table 12 Details of CBR-DM-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Lin (2009) Age, aspartate
aminotransferase, alanine
aminotransferase,
alkaline phosphatase,
total bilirubin, direct
bilirubin, total protein,
albumin, gamma-
glutamyl transpeptidase,
alpha-fetoprotein, sex,
blood type, HBsAg,
HBeAg, Anti-HBs, Anti-
HBe, Anti-HBc, Anti-
HCV
CBR: retrieve most similar
cases, reuse cases, revise
potential solution, retain new
solution, assign indices and
weights, case adaptation
(through inclusion, removal,
substitution or transformation).
DM: CART, tree-building
method, binary tree structure,
recursive binary splitting, tree
growing and tree pruning
stages, Gini diversity index
Fivefold cross-validation
methodology
Phase 1: Liver disorders
diagnosis
Phase II: Identify types
of liver disorders
(chronic hepatitis,
alcohol hepatitis, liver
cirrhosis and B
hepatitis.
Accuracy:
Phase I (CART):
Accuracy: 92.94%
Sensitivity: 96.00%
Specificity: 88.57%
AUC: 0.928
Phase II (CBR):
Accuracy: 90.00%
Sensitivity: 91.09%
Specificity: 88.41%
AUC: 0.889
3.9 CBR-GA
The CBR-GA methodology was used by Parka et al. (2011), where CBR uses GA to find
optimal cut-off distance and cut-off classification point. This integration overcomes the
limitation of conventional CBR of being deficient in reflecting asymmetric
misclassification cost. It has been found that average total misclassification cost of
proposed method (Parka et al., 2011) is considerably less than C5.0 and CART cost-
sensitive learning methods for a number of data sets. The survey on applicability of
CBR-GA for liver disorders is listed in Table 13.
52 A. Singh and B. Pandey
Table 13 Details of CBR-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Parka et al.
(2011)
Hepatitis data set CBR: cost-sensitive case-
based reasoning (CSCBR),
case retrieval.
GA: integration with nearest
neighbour method,
reproduction, crossover and
mutation operators.
Tenfold cross-validation,
paired t-test
Total misclassification cost of
CSCBR in medical data sets
(Hepatitis).
Total Cost: 7.0600
Note: CSCBR: cost-sensitive case-based reasoning.
3.10 DM-GA
DM-GA methodology was used by Sarkar et al. (2012), where DM is used for producing
rules from training data set and GA is used for handling interpretability problem, and by
Stoean et al. (2011b), where DM is used to extract features and GA is used to build rules
for establishing the diagnosis. In comparison, proposed learning system called DTGA
(decision tree and genetic algorithm) is more accurate than classifiers such as neural
network, Naive Bayes, C4.5, rough-set based rule inducer; and is less sensitive to missing
data compared to NN and C4.5 (Sarkar et al., 2012). This integration system also
enhances the performance over volumetric data and has less time complexity compared
to the majority of GA-based approaches. Cooperative Co-Evolutionary Algorithm
(CCEA) based proposed technique (Stoean et al., 2011a) has attained smallest standard
deviation and highest accuracy among classification techniques such as SVM, NN, SVM
+ PCA and NN + PCA. The survey on applicability of DM-GA for liver disorders is
listed in Table 14.
Table 14 Details of DM-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Stoean et al.
(2011)
Stiffness, sex, body mass index,
glycemia, triglycerides, cholesterol,
HDL cholesterol, aspartate
aminotransferase, alanin
aminotransferase, gama
glutamyltranspeptidase, total
bilirubin, alkaline phosphatase,
prothrombin index, TQS
(Quiq Time)
INR (prothrombin time ratio),
prolonged activated partial
thromboplastin time. haematids
(erythrocytes), haemoglobin,
hematocrit, medium erytrocity
volume, avg. erytrocitary
haemoglobin, Avg. concentration of
haemoglobin in a red blood cell,
thrombocytes, sideraemia
DM: PCA for feature
extraction, IF-Then
rules.
GA: mutation
operator, cooperative
co-evolutionary
algorithm (CCEA).
Hill climbing
algorithm
Liver fibrosis diagnosis
(differentiate between
five degrees of liver
fibrosis).
Accuracy: 62.11%
Intelligent techniques and applications in liver disorders 53
Table 14 Details of DM-GA-based systems with their results and applications (continued)
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Sarkar et al.
(2012)
Liver disorders data set DM: C4.5 decision
tree based rule
inducer algorithm, IF-
Then rules.
GA: selection, single
point crossover,
mutation
To improve prediction
accuracy for liver
disorders irrespective
to domain and size.
Accuracy: 80.02%
Note: PCA: Principal component analysis.
3.11 DM-FL
The DM-FL methodology was used by Luukka and Leppalampi (2006), where DM is
used for dimension reduction and fuzzy similarity model is used for classification, by
Luukka (2009), where fuzzy robust PCA algorithm is used for data pre-processing and
similarity classifier for classification, and by Torun and Tohumoglu (2011), where DM is
used to group the data and FL is used for classification. The DM-FL integration-based
systems are robust and effective in diagnosis. These systems also provide semantic
information about classification task (Luukka and Leppalampi, 2006) and have obtained
improved accuracies (Luukka, 2009). Luukka and Leppalampi’s (2006) fuzzy similarity
model performed fairly well as compared to other classifiers such as C4.5-1 (C4.5 with
default learning parameters) and C4.5-3 (C4.5 with parameter c equal to 95).
Classification accuracies are obtained using both dimension reduction methods: PCA and
entropy minimisation. The mean classification result is a bit higher using PCA than using
entropy. Luukka (2009) proposed FRPCA classification method which shows higher
accuracy when compared with those of conventional PCA and similarity classifier. Torun
and Tohumoglu (2011) proposed simulated annealing and subtractive clustering-based
fuzzy classifier (SASCFC) in four different versions (SASCFC Type 1, Type 2, Type 3,
and Type 4). Classifications results obtained from different versions are compared among
each other and it is found that SASCFC-Type 4 is superior. The survey on applicability
of DM-FL for liver disorders is listed in Table 15.
3.12 FL-GA
FL-GA methodology was used by Wang et al. (1998), where GA is used for generating
optimal set of fuzzy rules and membership functions, and by Chowdhury et al. (2007),
where GA is used to simultaneously integrate multiple fuzzy rule sets and their
membership function sets. The proposed genetic algorithm-based fuzzy-knowledge
integration framework needs no human intervention during integration process and is
scalable (Wang et al., 1998; Chowdhury et al., 2007). Accuracy of the framework
increases with increase in data size (Chowdhury et al., 2007). The shortcomings of the
framework are limited precision, and several unresolved issues in the field of knowledge
verification (Wang et al., 1998). In terms of classification accuracy, proposed genetic
algorithm-based fuzzy-knowledge integration framework (Wang et al., 1998) triumphs
over other learning methods mentioned in the literature such as CN2, C4, IR*, Bayes,
Assistant-86, and Diaconis and Efron. The survey on applicability of FL- GA for liver
disorders is listed in Table 16.
54 A. Singh and B. Pandey
Table 15 Details of DM-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Luukka and
Leppalampi
(2006)
Liver disorders
data set
DM: PCA and entropy minimisation
method.
FL: fuzzy similarity model,
Lukasiewicz structure for defining
memberships of objects
Detection of liver disorders.
Accuracy: 66.50% (PCA)
66.06% (entropy)
Luukka
(2009)
Liver disorders
data set
Data was pre-
p
rocessed using Fuzzy
robust PCA algorithms (FRPCA),
similarity classifier for
classification
Classification accuracy on
liver disorders data.
Accuracy: 70.25%
Torun and
Tohumoglu
(2011)
Liver disorders
data set
FL: fuzzy IF-Then rules, fuzzy
inference system (FIS).
DM: subtractive clustering, wrapper
type feature selection approach.
Simulated annealing, least square
estimation, k-fold cross-validation
Liver disorders classification.
Accuracy:
SASCFC-Type 1:73.6%
SASCFC-Type 2:73.9%
SASCFC-Type 3:73.93%
SASCFC-Type 4:74.13%
Notes: PCA: Principal Component Analysis, SA: Simulated Annealing, SC:
Subtractive Clustering.
Table 16 Details of FL-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Wang et al.
(1998)
Hepatitis data set FL: fuzzy knowledge encoding, fuzzy
knowledge integration, IF-Then rules,
isosceles-triangle functions, Parodi and
Bonelli parameters to represent each
membership function.
GA: two-substring crossover operator,
two-part mutation operator, Pittsburgh
approach
Hepatitis diagnosis.
Accuracy: 91.61%
Chowdhury
et al. (2007)
Age, bilirubin,
alk phosphate,
SGOT, albumin
and protime
FL: fuzzy knowledge encoding and
fuzzy knowledge integration, fuzzy
rules, isosceles-triangle functions,
Parodi and Bonelli parameters to
represent each membership function.
GA: SBMAC (sub population based
max-mean arithmetical crossover),
dynamic time-variant mutation
(TVM), insertion mutation, deletion
mutation, novel evolutionary strategy
algorithm
Hepatitis diagnosis.
Accuracy: 96.33%
3.13 AIS-ANN-FL
AIS-ANN-FL methodology was used by Kahramanli and Allahverdi (2009), where AIS
algorithm is deployed to extract rules from hybrid neural network. Generated rules are
very accurate but are large in numbers. Classification accuracy of the proposed
integration approach is superior to other classification techniques mentioned in the
Intelligent techniques and applications in liver disorders 55
literatures such as C-MLP2LN, FSM and CART. The survey on applicability of AIS-
ANN-FL for liver disorders is listed in Table 17.
Table 17 Details of AIS-ANN-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Kahramanli and
Allahverdi (2009)
Hepatitis data set Hybrid neural network: artificial
neural network, fuzzy neural
network, trained with back
propagation algorithm,
fuzzification, defuzzification,
weight update method of back
propagation algorithm.
Artificial immune systems (AIS)
algorithm: extracting rules from
hybrid neural network, IF-Then
rules
Liver disorders
classification on hepatitis
data set.
Accuracy: 96.78%
Sensitivity: 97.56%
Specificity: 93.75%
3.14 ANN-CBR-RBR
ANN-CBR-RBR methodology was used by Obot and Uzoka (2009), where case-based
technique outputs constitute an input to ANN and results obtained from ANN are assisted
to form rule base. Finally, a hybrid inference engine has been built to obtain accuracy
through rule base. This hybrid system (Obot and Uzoka, 2009) provides high diagnostic
accuracy and high speed for retrieval of information. Limitations of the system are:
restricted explanation capability, ineffective in managing noisy data due to fragility of
rules, and it would not like the insertion of new knowledge. The survey on applicability
of ANN-CBR-RBR for liver disorders is listed in Table 18.
Table 18 Details of ANN-CBR-RBR-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Obot and Uzoka
(2009)
Nausea, vomiting, fever, body
weakness, loss of appetite,
diarrhoea, itching, convulsion,
stupor, headache, tremors,
skin discoloration, eye
discoloration, liver
tenderness, bile in urine,
jaundice
ANN: trained with
multi-layer perceptron
back propagation neural
networks (MLPBPNN).
CBR: retrieval using
binary search algorithm,
adaptation.
RBR: IF-Then rules
Diagnosis of hepatitis
disease
3.15 AIS-DM-FL
AIS-DM-FL methodology was used by Polat and Gunes (2007a), where DM is used for
dimensionality reduction, FL is used for data-weighted processing and AIS for
classification, and by Polat and Gunes (2007b), where DM is used for feature reduction,
FL is used for weighting the whole data set and AIS for classification. Advantage of
deploying AIRS is that it is not necessary to know the appropriate settings for the
classifier. In terms of classification accuracy, Polat and Gunes (2007a) proposed a hybrid
56 A. Singh and B. Pandey
approach that triumphs over other learning methods such as MLP, RBF (ToolDiag), MLP
+ BP (ToolDiag) and GRNN. Polat and Gunes (2007b) proposed a machine learning
approach that obtains very promising results, which are effective, accurate and superior
to weighted 9-NN, 18-NN, 15-NN, FSM with rotations, FSM without rotations, RBF
(ToolDiag), LDA, Naive Bayes, QDA, 1-NN, ASR, Fisher discriminant analysis, LVQ,
CART (decision tree), MLP with BP, ASI, LFC, IncNet, MLP, RBF and GRNN. The
survey on applicability of AIS-DM-FL for liver disorders is listed in Table 19.
Table 19 Details of AIS-DM-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Polat and Gunes
(2007a)
Hepatitis data set DM: C4.5 decision tree algorithm,
wrapper approach, filter-based
feature selection.
FL: fuzzy weighted pre-
processing, triangular membership
functions, fuzzy IF-Then rules.
AIS: AIRS supervised learning
algorithm, immune mechanisms
used are resource competition,
clonal selection, affinity
maturation and memory cell
formation,
stages of AIRS includes
initialisation, memory cell
identification and
ARB generation
Diagnosis of hepatitis
disease.
Accuracy: 81.82%
Sensitivity: 55.56%
Specificity: 83.82%
Polat and Gunes
(2007b)
Hepatitis data set DM: C4.5 decision tree algorithm.
FL: weighted with fuzzy weighted
pre-processing, triangular
membership functions, fuzzy IF-
Then rules.
AIS: AIRS, immune metaphors
used are antibody-antigen binding,
affinity maturation, clonal
selection process, resource
distribution and memory
acquisition, learning algorithm
consists of initialisation, memory
cell recognition, resource
competition and revision of
resulted memory cells.
tenfold cross-validation
Diagnosis of hepatitis
disease.
Accuracy: 94.12%
Sensitivity: 100%
Specificity: 92.85%
ADR: 96.42%
Notes: AIRS: Artificial Immune Recognition System, ARB: Artificial Recognition Balls.
3.16 ANN-DM-FL
ANN-DM-FL methodology was used by Su et al. (2006), where fuzzy ART neural
network is employed to construct information granules and DM is used to extract
knowledge rules from the granules, and by Li et al. (2011), where fuzzy-based non-linear
Intelligent techniques and applications in liver disorders 57
transformation method is applied to extend classification-related information, DM is used
for extracting the optimal subset of features and ANN is used for classification. The
hybrid model (Su et al., 2006) effectively deals with imbalanced data sets. Obtained
simulated results prove superiority of proposed fuzzy-based non-linear transformation
method (Li et al., 2011) over PCA and Kernel Principal Component Analysis (KPCA);
and superiority of proposed knowledge acquisition via information granulation model
(Su et al., 2006) over C4.5 and SVM. The survey on applicability of ANN-DM-FL for
liver disorders is listed in Table 20.
Table 20 Details of ANN-DM-FL-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Su et al. (2006) Liver disorders
data set
ANN: Fuzzy ART (Adaptive
Resonance Theory) neural network.
DM: knowledge acquisition via C4.5
decision tree
Improve classification
performance by solving
class imbalance
problems.
Accuracy (KAIG): 70%
Data type: Information
granules
Li et al. (2011) Liver disorders
data set
FL: fuzzy-based non-linear
transformation method, triangle shape
membership function, a fuzzy
membership computational approach.
DM: principal component analysis.
ANN: SVM, megatrend diffusion
(MTD) function, polynomial and
Gaussian kernel.
t-test, Friedman test, ANOVA test,
Demsar’s proposed method
To increase
classification
performance with small
medical data sets.
Accuracy:
SVM (poly): 54.21%
SVM (gaus): 54.13%
3.17 ANN-DM-GA
ANN-DM-GA methodology was used by Stoean et al. (2011b), where GA is used to
dynamically concentrate search only on the most relevant attributes, DM is used to
reduce the data dimensionality, and ANN makes the novel model flexible and good
performer. Proposed evolutionary approach also obtained better classification accuracy
than traditional SVMs. ESVM is deployed instead of SVM because it has successfully
resolved complexity of SVM. ESVM has more simplicity, stability, flexibility,
robustness, transparency, adaptability and operability, but ESVM training is a bit slow
than standard SVMs. The survey on applicability of ANN-DM-GA for liver disorders is
listed in Table 21.
3.18 ANN-GA-RBR
ANN-GA-RBR methodology was used by Ramirez et al. (2012), where RBR is used for
decision-making, GA is used for obtaining new offspring and ANN is used for
classification. Liver transplantation is the only treatment for patients having incurable
liver disorders. Availability of donors is less due to number of requirements; and
58 A. Singh and B. Pandey
transplantation is solely dependent on the availability of liver donors. This disproportion
may result in countless deaths. Ramirez et al. (2012) did a fruitful research work by
developing a donor-recipient decision system for liver transplantation, which prioritises
recipients in queue. The tool is intelligible and sensible for physicians. The survey on
applicability of ANN-GA-RBR for liver disorders is listed in Table 22.
Table 21 Details of ANN-DM-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Stoean et al.
(2011)
Stiffness, sex, BMI (body mass
index), glycemia, triglycerides,
cholesterol, HDL cholesterol,
aspartate aminotransferase, alanin
aminotransferase, gama glutamyl
transpeptidase,
total bilirubin, alkaline phosphatase,
prothrombin index, quiq time,
prothrombin time ratio, prolonged
activated partial thromboplastin
time, haematids, haemoglobin,
hematocrit, medium erytrocity
volume, avg. erytrocitary
haemoglobin, avg. concentration of
haemoglobin in a red blood cell,
thrombocytes, sideraemia
ANN: evolutionary-
powered support
vector machines
(ESVM).
DM: principal
component analysis.
GA: tournament
selection method,
reproduction
(randomly selected),
recombination
(randomly generated),
mutation (randomly
generated)
Determining the
degree of liver fibrosis
in chronic hepatitis C.
Accuracy: 77.31%
Table 22 Details of ANN-GA-RBR-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Ramirez et
al. (2012)
A liver transplant data
set composed of 1001
patterns is used for
experimentation
ANN: feedforward neural network,
linear basis function, probabilistic
function, radial basis function
neural network, probability density
function of generalised Gaussian
distribution, trained with a multi-
objective evolutionary learning
algorithm (MOEA) called
MPENSGA2.
GA: multi-objective evolutionary
algorithms, structural and
parametric mutation operators.
RBR: rule-based system designed
using two ANN models named
asMPENSGA2-E and MPENSGA2-
MS, IF-Then rules.
Liver transplantation
decision
Notes: MPENSGA2: Memetic Pareto Evolutionary NSGA2).
3.19 CBR-GA-PSO
CBR-GA-PSO methodology was used by Chang et al. (2012), where CBR is used to pre-
process data set, GA is used to evolve weights of each attribute in PSO and PSO is used
to construct the medical classification system. The proposed framework generates more
Intelligent techniques and applications in liver disorders 59
precise, effective and intelligible results. Advantage of using PSO is its capability to
overcome overlapping situation of data set. Simulated results (Chang et al., 2012)
compared with other forecasting models such as SVM, KNN, Naive Bayes, FDT,
RULES-4, C4.5, BNND, BNNF, SVM with GP, SSVM, RSVM, MLP, PNN, and GRNN
demonstrate the superiority of proposed model (CBRPSO). It is also proved that this
model has the capability to produce high compact clustering than methods such as PSO
and K-means. Different PSO-based approaches have been compared in which GA-
CBRPSO outperforms PSO and CBR-PSO. The survey on applicability of CBR-GA-PSO
for liver disorders is listed in Table 23.
Table 23 Details of CBR-GA-PSO-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Chang et al.
(2012)
Liver disorders
data set
CBR: case base weighted cluster
algorithm, weight vector, gradient
method, feature evaluation function.
GA: selection, crossover, mutation,
replacement.
PSO: PSO tool evolved by genetic
algorithm, PSO-clustering algorithm,
global searching stage, local refining
stage, k-means algorithm
Liver disorders diagnosis.
Accuracy: 76.8%
3.20 DM-FL-GA
DM-FL-GA methodology was used by Leung et al. (2011) where a DM-based
framework is introduced in which GA is used for searching and optimisation and FL is
used to increase performance. Proposed data mining framework obtains much better
results than other forecasting models mentioned in the literature such as SVM, C5.0
decision tree, neural network and Naive Bayes. The survey on applicability of DM-FL-
GA for liver disorders is listed in Table 24.
Table 24 Details of DM-FL-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Leung et al.
(2011)
Genomic sequences:
HBV DNA
sequences, either
genotype B or C
over 200 patients
DM: molecular evolution analysis,
clustering, feature selection,
classifier learning and
classification, phylogenetic tree
analysis, information gain criterion,
rule learning based on evolutionary
algorithm.
GA: generic genetic programming
(GGP).
FL: fuzzy measure for good
performance of classification
method.
10 K-fold method
To classify the HBV
DNA data into liver
cancer (HCC) and
normal (CON, control)
classes
60 A. Singh and B. Pandey
3.21 CBR-DM-FL-GA
CBR-DM-FL-GA methodology was used by Chang et al. (2010), where CBR is used for
decomposing the database into a set of smaller database, fuzzy decision tree is used to
classify data and GA is used for selecting optimal fuzzy terms which in long-term
improves accuracy, and by Fan et al. (2011), where CBR is used for clustering the data
set, fuzzy decision tree is used to classify data and GA is used for further improving the
classified result by evolving the number of fuzzy terms. These integrated systems are
more precise and effective and also productively assisting doctors in diagnosis (Chang et
al., 2010; Fan et al., 2011). It has been noticed that CBR-FDT model (Chang et al., 2010)
reaches the highest classification accuracy among other benchmark classifiers such as
improved particle swarm optimisation model (KNMPSO), SVM, K-nearest neighbour
(KNN), SVMS, RMSVM and back propagation neural networks (BPN). Average hit rate
of this model is highest also among all mentioned approaches on different database
classification applications. Proposed CBFDT model (Fan et al., 2011) also shows
promising performance and obtains higher classification accuracy when compared with
RULES-4, C4.5, BNND, BNNF, SVM with GP, SSVM, RSVM, KNN, Naive Bayes,
MLP, PNN GRNN, HNFB and SVM. The survey on applicability of CBR-DM-FL-GA
for liver disorders is listed in Table 25.
Table 25 Details of CBR-DM-FL-GA-based systems with their results and applications
Author, Year Attributes Intelligent techniques
and other methods Result and Application
Chang et al.
(2010)
Liver disorders
data set
CBR: case-based weighted cluster
algorithm.
DM and FL: fuzzy decision tree (FDT)
generated from ID 3 algorithm based on
recursive binary partitioning algorithm,
data fuzzification, triangle membership
functions, fuzzy rules.
GA: reproduction/selection, replacement.
Stepwise regression analysis (SRA) model
Liver disorders
diagnosis.
Accuracy: 81.6%
Fan et al.
(2011)
Liver disorders
data set
CBR: case-based weighted cluster
algorithm, clustering method used was k-
means algorithm.
DM and FL: fuzzy decision tree (FDT)
generated from ID 3 algorithm based on
recursive binary partitioning algorithm,
triangle membership functions, fuzzy
rules
GA: representation and
selection(tournament method), two-point
crossover method, two-point mutation
method, binary code was adopted, de-
coding, tournament method, replace,
terminate
Stepwise regression analysis (SRA) model
Liver disorders
diagnosis.
Accuracy: 81.6%
(Average)
90.40% (Best)
Note: ID: Iterative Dichotomiser.
Intelligent techniques and applications in liver disorders 61
4 Observation
We have already mentioned that the important goals of this paper are to survey and
classify intelligent techniques applied to liver disorders and identify which intelligent
techniques are applied to what types of liver disorders. The database search covers the
papers that had been published till January 2013. All the articles are carefully studied,
sorted and classified based on implementation of intelligent techniques. Individual ITs
include ANN, DM, FL and GA. Integrated ITs combine methods such as ANN-CBR,
ANN-DM, ANN-FL, AIS-FL, ANN-GA, AIS-GA, ANN-PSO, CBR-DM, CBR-GA,
DM-GA, DM-FL, FL-GA, AIS-ANN-FL, ANN-CBR-RBR, AIS-DM-FL, ANN-DM-FL,
ANN-DM-GA, ANN-GA-RBR, CBR-GA-PSO, DM-FL-GA, and CBR-DM-FL-GA.
Based on all the search results Table 26 have been prepared. This table detailed the
applicability of intelligent techniques to different types of liver disorders (hepatitis, liver
fibrosis, liver cirrhosis, liver cancer, fatty liver, liver disorders data set, hepatitis data set,
hepatobiliary disorders data set and others). Liver disorders data set is used to diagnose
liver disorders that might arise from excessive alcohol consumption. Hepatobiliary
disorders data set is used to diagnose Alcoholic Liver Damage (ALD), primary hepatoma
(PH), liver cirrhosis (LC) and cholelithiasis (C). Table 26 presents the five articles in
total out of which three are ANN-based (Lee et al., 2005; Jeon et al., 2013; Su and Yang,
2008), one is DM-based (Jen et al., 2012) and one is ANN-GA-RBR-based (Ramirez et
al., 2012). Either of the two conditions have been checked before classifying an article in
‘others’ column: first, if it has neither used liver disorders data set, hepatitis data set and
hepatobiliary disorders data set nor it has covered the types of liver disorders which are
considered for this survey. Second, if the work is related to liver other than diagnosis.
Table 26 presents which individual and integrated ITs are appreciably used for what
types of liver disorders. For hepatitis liver disorders ANN, DM, FL, ANN-CBR, CBR-
DM, FL-GA and ANN-CBR-RBR are used; for liver fibrosis ANN, ANN-DM, ANN-
GA, DM-GA and ANN-DM-GA are applied; for liver cirrhosis ANN, DM, FL, GA,
ANN-CBR, ANN-FL and CBR-DM are used; for liver cancer ANN, FL, GA and DM-
FL-GA are applied; for fatty liver ANN and FL are used; for liver disorders data set
ANN, DM, FL, GA, ANN-DM, ANN-FL, AIS-FL, ANN-GA, AIS-GA, DM-GA, DM-
FL, ANN-DM-FL, CBR- GA-PSO and CBR-DM-FL-GA are applied; for hepatitis data
set ANN, GA, ANN-DM, ANN-FL, AIS-FL, ANN-PSO, CBR-GA, FL-GA, AIS-ANN-
FL and AIS-DM-FL are used; for hepatobiliary disorders data set ANN and FL are
applied.
For comparison among individual ITs, Figure 1 is presented which shows the number
of ANN, DM, FL and GA based studies. Figure 2 illustrates the preference of integrated
ITs compared to others. It is observed from Figures 1 and 2 that most of the researchers
have preferred to use ANN and ANN-FL methodologies compared to other techniques. It
has also found that ANN is mostly integrated with DM and FL, and vice versa. Figure 3
gives graphical representation about the usage of single and integrated ITs after every
few years. The most surprising fact is the negligible applicability of integrated ITs to
liver disorders before the year 2007. If we talk in percentage than 80% usage of
integrated techniques is after 2007 with just 20% usage till 2006 (Figure 4). Another
message emerging from the study is that not even a single intelligent technique had been
applied between 1999 and 2002.
62 A. Singh and B. Pandey
Table 26 Applicability of intelligent techniques to different types of liver disorders
Liver Disorders
Hepatitis Fibrosis Cirrhosis Cancer Fatty Liver disorders
data set
Hepatitis
data set
Hepatobiliary
disorders data set Others Total
ANN 3 1 8 4 2 4 5 2 3 32
DM 2 2 1 1 6
FL 1 1 1 1 3 1 8
GA 1 1 2 1 5
ANN-CBR 2 1 3
ANN-DM 1 1 2 4
ANN-FL 1 6 1 8
AIS-FL 2 1 3
ANN-GA 1 1 2
AIS-GA 1 1
ANN-PSO 1 1
CBR-DM 1 1 2
CBR-GA 1 1
DM-GA 1 1 2
DM-FL 3 3
FL-GA 1 1 2
AIS-ANN-FL 1 1
ANN-CBR-RBR 1 1
AIS-DM-FL 2 2
ANN-DM-FL 2 2
ANN-DM-GA 1 1
ANN-GA-RBR 1 1
CBR-GA-PSO 1 1
DM-FL-GA 1 1
CBR-DM-FL-GA 2 2
Intelligent techniques and applications in liver disorders 63
Figure 1 Comparison of the numbers of individual intelligent techniques based published papers
between the years 1995 and 2013
Figure 2 Comparison of the numbers of integrated intelligent technique-based published papers
between the years 1995–2013
64 A. Singh and B. Pandey
Figure 3 Comparison of the numbers of individual and integrated intelligent technique-based
published papers between the years 1995 and 2013
Figure 4 Comparative view of percentage use between the intervals 1995–2006 and 2007–2013
On computed relative comparison, it has been found that ANN method is most
significantly used with its 60% rate, FL method is in the second rank with its 17% rate,
DM method is in the third rank with its 14% rate, and GA method is the last one with
only 9% rate. In integrated intelligent techniques, mostly used methodology is ANN-FL
with 20% rate, then ANN-DM, DM-FL with 8% rate, then AIS-FL, ANN-GA, ANN-
CBR, DM-GA, FL-GA, AIS-DM-FL, ANN-DM-FL, CBR-DM-FL-GA with 5% rate and
the last ones are AIS-GA, ANN-PSO, CBR-DM, CBR-GA, ANN-CBR-RBR, AIS-ANN-
FL, ANN-DM-GA, ANN-GA-RBR, CBR-GA-PSO, DM-FL-GA with 2% rate.
Intelligent techniques and applications in liver disorders 65
During this survey, we have found that the term intelligent techniques is not used
well enough as a keyword in articles. Terms like ANN, AIS, CBR, DM, FL, GA, PSO,
RBR, hybrid systems and integrated systems are mostly chosen instead of ITs. So we
have introduced a new keyword ‘intelligent techniques’ which refers to all methodologies
mentioned in this study. Along with number of benefits, writing a survey paper also has
several limitations like author’s limited knowledge as one needs to have extensive
background knowledge for accumulating, studying and classifying articles; number of
papers may have used intelligent techniques but still left out due to some indexing
problems; publications using languages other than English cannot be included; it is
difficult to cite all academic articles that are listed in Science Citation Index (SCI) as the
amount of available text is increasing rapidly; and we have limited access to online
database and also bounded by time constraint.
Hopefully, this study would be productive for neophyte researchers, about what to do
and what not to, in developing medical decision-making systems which assist physicians
in handling liver disorders. Novice researchers can either use methodologies like ANN,
ANN-FL which has wide acceptance and has obtained higher accuracy results or can
choose techniques such as AIS and PSO which have not been explored enough hitherto
and can give improved results either alone or when integrated with some other intelligent
techniques. For hepatitis diagnosis, ANN and ANN-CBR would be more suitable; for
liver cirrhosis diagnosis, ANN and DM would be more suitable; for liver cancer and fatty
liver diagnosis, ANN would be more suitable; for liver disorders data set, ANN-FL
would be more suitable; for hepatitis data set and hepatobiliary disorders data set, again
ANN would be more suitable.
5 Conclusion
This paper presents a survey of literature concerned with intelligent techniques applied to
liver disorders between the January 1995 and January 2013. Articles were searched using
different keyword indices such as ‘liver disorders’, ‘liver disorders diagnosis’, ‘hepatitis’,
‘liver fibrosis’, ‘liver cirrhosis’, ‘liver cancer’, ‘fatty liver’, ‘ANN used for liver
disorders’, ‘AIS used for liver disorders’, ‘CBR used for liver disorders’, ‘DM used for
liver disorders’, ‘FL used for liver disorders’, ‘GA used for liver disorders’, ‘PSO used
for liver disorders’ and ‘RBR used for liver disorders’; and then classified based on
intelligent techniques applied and which techniques is used for what types of liver
disorders. The trends indicating from the survey tables are that all intelligent techniques
were being progressively applied, from 2007 to 2013, to liver disorders. Systems
developed using intelligent techniques to handle imprecise, unstructured and dynamic
data of patients, and also assist physicians by acting as a second opinion tool in decision-
making process for liver disorders diagnosis.
Optimistically, this study has attained the objective of a survey in the following
manner: it provides detail about the articles published for liver disorders from 1995 to
2013; it presents the information about which individual and integrated techniques are
used for what types of liver disorders; it specifies the articles published with their results
and applications; it portrays accurately the characteristics of ITs and compares their
usage among each others; it narrows the researcher’s work as they become aware with
66 A. Singh and B. Pandey
pros and cons of the intelligent techniques; it also plays a fundamental role in
formulating research hypothesis, preparing research design, and collecting and analysing
the data.
It is suggested that novice researchers can use methodologies such as ANN, ANN-FL
which have wide acceptance and have obtained higher accuracy results or can choose
techniques such as AIS and PSO which have not been explored enough hitherto and can
give improved results either alone or when integrated with some other intelligent
techniques. For hepatitis diagnosis, ANN and ANN-CBR would be more suitable; for
liver cirrhosis diagnosis, ANN and DM would be more suitable; for liver cancer and fatty
liver diagnosis, ANN would be more suitable; for liver disorders data set, ANN-FL
would be more suitable; for hepatitis data set and hepatobiliary disorders data set, again
ANN would be more suitable. Though there are a number of other integration
methodologies such as AIS-GA, ANN-PSO, CBR-DM and CBR-GA which have not
been widely implemented until now. So, it is completely a researcher’s decision to decide
with which techniques to proceed based on his background knowledge and information
he had grabbed from this study. It is hoped and anticipated that the humble effort made in
this study will assist in the accomplishment of developing accurate and precise decision
making tools to diagnose liver disorders.
References
Ansari, S., Shafi, I., Ansari, A., Ahmad, J. and Shah, S.I. (2011) ‘Diagnosis of liver disease
induced by hepatitis virus using artificial neural networks’, IEEE 14th International
Multitopic Conference (INMIC), 22–24 December, Karachi, pp.8–12.
Arsene, C.T.C. and Lisboa, P.J. (2012) ‘Bayesian neural network applied in medical survival
analysis of primary biliary cirrhosis’, IEEE 14th International Conference on Computer
Modelling and Simulation, 28–30 March, Cambridge, pp.81–85.
Autio, L., Juhola, M. and Laurikkala, J. (2007) ‘On the neural network classification of medical
data and an endeavour to balance non-uniform data sets with artificial data extension’,
Computers in Biology and Medicine, Vol. 37, No. 3, pp.388–397.
Azaid, S.A., Fakhr, M.W. and Mohamed, A.F.A. (2006) ‘Automatic diagnosis of liver diseases
from ultrasound images’, IEEE International Conference on Computer Engineering and
Systems, 5–7 November, Cairo, pp.313–319.
Babu, G.S. and Suresh, S. (2013) ‘Meta-cognitive RBF network and its projection based learning
algorithm for classification problems’, Applied Soft Computing, Vol. 13, No. 1, pp.654–666.
Badawi, A.M., Derbala, A.S. and Youssef, A.B.M. (1999) ‘Fuzzy logic algorithm for quantitative
tissue characterization of diffuse liver diseases from ultrasound images’, International Journal
of Medical Informatics, Vol. 55, No. 2, pp.135–147.
Bologna, G. (2003) ‘A model for single and multiple knowledge based networks’, Artificial
Intelligence in Medicine, Vol. 28, No. 2, pp.141–163.
Bucak, I.O. and Baki, S. (2010) ‘Diagnosis of liver disease by using CMAC neural network
approach’, Expert Systems with Applications, Vol. 37, No. 9, pp.6157–6164.
Calisir, D. and Dogantekin, E. (2011) ‘A new intelligent hepatitis diagnosis system: PCA–
LSSVM’, Expert Systems with Applications, Vol. 38, No. 8, pp.10705–10708.
Celikyilmaz, A., Turksen, I.B., Aktas, R., Doganay, M.M. and Ceylan, N.B. (2009) ‘Increasing
accuracy of two-class pattern recognition with enhanced fuzzy functions’, Expert Systems with
Applications, Vol. 36, No. 2, pp.1337–1354.
Ceylan, R., Ceylan, M., Ozbay, Y. and Kara, S. (2011) ‘Fuzzy clustering complex-valued neural
network to diagnose cirrhosis disease’, Expert Systems with Applications, Vol. 38, No. 8,
pp.9744–9751.
Intelligent techniques and applications in liver disorders 67
Chang, P.C., Fan, C.Y. and Dzan, W.Y. (2010) ‘A CBR-based fuzzy decision tree approach for
database classification’, Expert Systems with Applications, Vol. 37, No. 1, pp.214–225.
Chang, P.C., Lin, J.J. and Liu, C.H. (2012) ‘An attribute weight assignment and particle swarm
optimization algorithm for medical database classifications’, Computer Methods and
Programs in Biomedicine, Vol. 107, No. 3, pp.382–392.
Chowdhury, N.A., Khatun, M. and Hashem, M.M.A. (2007) ‘On integrating fuzzy knowledge
using a novel evolutionary algorithm’, IEEE 10th International Conference on Computer and
Information technology (ICCIT), 27–29 December, Dhaka, pp.1–6.
Chuang, C.L. (2011) ‘Case-based reasoning support for liver disease diagnosis’, Artificial
Intelligence in Medicine, Vol. 53, No. 1, pp.15–23.
Comak, E., Polat, K., Gunes, S. and Arslan, A. (2007) ‘A new medical decision making system:
least square support vector machine (LSSVM) with fuzzy weighting pre-processing’, Expert
Systems with Applications, Vol. 32, No. 2, pp.409–414.
Dehuri, S. and Cho, S.B. (2010) ‘Evolutionarily optimized features in functional link neural
network for classification’, Expert Systems with Applications, Vol. 37, No. 6, pp.4379–4391.
Dogantekin, E., Dogantekin, A. and Avci, D. (2009) ‘Automatic hepatitis diagnosis system based
on linear discriminant analysis and adaptive network based on fuzzy inference system’, Expert
Systems with Applications, Vol. 36, No. 8, pp.11282–11286.
Dong, Y., Xia, Z. and Xia, Z. (2008) ‘A two-level approach to choose the cost parameter in support
vector machines’, Expert Systems with Applications, Vol. 34, No. 2, pp.1366–1370.
Eastwood, M. and Gabrys, B. (2012) ‘Generalised bottom-up pruning: a model level combination
of decision trees’, Expert Systems with Applications, Vol. 39, No. 10, pp.9150–9158.
Elizondo, D.A., Birkenhead, R., Gamez, M., Garcia, N. and Alfaro, E. (2012) ‘Linear separability
and classification complexity’, Expert Systems with Applications, Vol. 39, No. 9, pp.7796–
7807.
Falco, I.D. (2013) ‘Differential evolution for automatic rule extraction from medical databases’,
Applied Soft Computing, Vol. 13, No. 2, pp.1265–1283.
Fan, C.Y., Chang, P.C., Lin, J.J. and Hsieh, J.C. (2011) ‘A hybrid model combining case-based
reasoning and fuzzy decision tree for medical data classification’, Applied Soft Computing,
Vol. 11, No. 1, pp.632–644.
Floares, A.G. (2009) ‘Intelligent clinical decision supports for interferon treatment in chronic
hepatitis C and B based on i-BiopsyTM’, IEEE Proceedings of International Joint Conference
on Neural Networks (IJCNN), 14–19 June, Atlanta, Georgia, USA, pp.855–860.
Gadaras, I. and Mikhailov, L. (2009) ‘An interpretable fuzzy rule-based classification methodology
for medical diagnosis’, Artificial Intelligence in Medicine, Vol. 47, No. 1, pp.25–41.
Gorunescu, F., Belciug, S., Gorunescu, M. and Badea, R. (2012) ‘Intelligent decision-making for
liver fibrosis stadialization based on tandem feature selection and evolutionary-driven neural
network’, Expert Systems with Applications, Vol. 39, No. 17, pp.12824–12832.
Hamamoto, I., Okada, S., Hashimoto, T., Wakabayashi, H., Maeba, T. and Maeta, H. (1995)
‘Prediction of the early prognosis of the hepatectomized patient with hepatocellular carcinoma
with a neural network’, Computers in Biology and Medicine, Vol. 25, No. 1, pp.49–59.
Hashem, A.M., Rasmy, M.E.M., Wahba, K.M. and Shaker, O.G. (2010) ‘Prediction of the degree
of liver fibrosis using different pattern recognition techniques’, IEEE 5th Cairo International
Biomedical Engineering Conference (CIBEC), 16–18 December, Cairo, pp.210–214.
Hashem, A.M., Rasmy, M.E.M., Wahba, K.M. and Shaker, O.G. (2012) ‘Single stage and
multistage classification models for the prediction of liver fibrosis degree in patients’,
Computer Methods and Programs in Biomedicine, Vol. 105, No. 3, pp.194–209.
Hayashi, Y. and Setiono, R. (2002) ‘Combining neural network predictions for medical diagnosis’,
Computers in Biology and Medicine, Vol. 32, No. 4, pp.237–246.
Hayashi, Y., Setiono, R. and Yoshida, K. (2000) ‘A comparison between two neural network rule
extraction techniques for the diagnosis of hepatobiliary disorders’, Artificial Intelligence in
Medicine, Vol. 20, No. 3, pp.205–216.
68 A. Singh and B. Pandey
Icer, S., Kara, S. and Guven, A. (2006) ‘Comparison of multilayer perceptron training algorithms
for portal venous Doppler signals in the cirrhosis disease’, Expert Systems with Applications,
Vol. 31, No. 2, pp.406–413.
Jen, C.H., Wang, C.C., Jiang, B.C., Chu, Y.H. and Chen, M.S. (2012) ‘Application of classification
techniques on development an early-warning system for chronic illnesses’, Expert Systems
with Applications, Vol. 39, No. 10, pp.8852–8858.
Jeon, J.H., Choi, J.Y., Lee, S. and Ro, Y.M. (2013) ‘Multiple ROI selection based focal liver lesion
classification in ultrasound images’, Expert Systems with Applications, Vol. 40, No. 2,
pp.450–457.
Kahramanli, H. and Allahverdi, N. (2009) ‘Extracting rules for classification problems: AIS based
approach’, Expert Systems with Applications, Vol. 36, No. 7, pp.10494–10502.
Kohara, S., Tateyama, T., Foruzan, A.H., Furukawa, A., Kanasaki, S., Wakamiya, M. and Chen,
Y.W. (2010) ‘Application of statistical shape model to diagnosis of liver disease’, IEEE 2nd
International Conference on Software Engineering and Data Mining (SEDM), 23–25 June,
Chengdu, China, pp.680–683.
Kulluk, S., Ozbakır, L. and Baykasoglu, A. (2013) ‘Fuzzy DIFACONN-miner: a novel approach
for fuzzy rule extraction from neural networks’, Expert Systems with Applications, Vol. 40,
No. 3, pp.938–946.
Lee, C.C., Chung, P.C. and Chen, Y.J. (2005) ‘Classification of liver diseases from CT images
using BP-CMAC neural network’, Proceeding of the 9th IEEE International Workshop on
Cellular Neural Networks and their Applications (CNNA), 28–30 May, pp.118–121.
Leung, K.S., Lee, K.H., Wang, J.F., Ng, E.Y.T., Chan, H.L.Y., Tsui, S.K.W., Mok, T.S.K., Tse,
P.C.H. and Sung, J.J.Y. (2011) ‘Data mining on DNA sequences of hepatitis B virus’,
IEEE/ACM Transactions on Computational Biology and Bioinformatics, March/April, Vol. 8,
No. 2, pp.428–440.
Li, B.N., Chui, C.K., Chang, S. and Ong, S.H. (2012) ‘A new unified level set method for semi-
automatic liver tumor segmentation on contrast-enhanced CT images’, Expert Systems with
Applications, Vol. 39, No. 10, pp.9661–9668.
Li, D.C. and Liu, C.W. (2010) ‘A class possibility based kernel to increase classification accuracy
for small data sets using support vector machines’, Expert Systems with Applications, Vol. 37,
No. 4, pp.3104–3110.
Li, D.C., Liu, C.W. and Hu, S.C. (2010) ‘A learning method for the class imbalance problem with
medical data sets’, Computers in Biology and Medicine, Vol. 40, No. 5, pp.509–518.
Li, D.C., Liu, C.W. and Hu, S.C. (2011) ‘A fuzzy-based data transformation for feature extraction
to increase classification performance with small medical data sets’, Artificial Intelligence in
Medicine, Vol. 52, No. 1, pp.45–52.
Lin, R. H. (2009) ‘An intelligent model for liver disease diagnosis’, Artificial Intelligence in
Medicine, Vol. 47, No. 1, pp.53–62.
Lin, R.H. and Chuang, C.L. (2010) ‘A hybrid diagnosis model for determining the types of the
liver disease’, Computers in Biology and Medicine, Vol. 40, No. 7, pp.665–670.
Luo, Y., Li, J.Q., Zheng, D.W., Tan, Z.P., Zhou, H., Deng, Q.P., Liu, Y.T., Ou, A. and Yin, J.
(2011) ‘Application of data mining technology in excavating prevention and treatment
experience of infectious diseases from famous herbalist doctors’, IEEE International
Conference on Bioinformatics and Biomedicine (Workshops (BIBMW), 12–15 November,
Atlanta, GA, pp.784–790.
Luukka, P. (2009) ‘Classification based on fuzzy robust PCA algorithms and similarity classifier’,
Expert Systems with Applications, Vol. 36, No. 4, pp.7463–7468.
Luukka, P. (2011) ‘Fuzzy beans in classification’, Expert Systems with Applications, Vol. 38, No.
5, pp.4798–4801.
Luukka, P. and Leppalampi, T. (2006) ‘Similarity classifier with generalized mean applied to
medical data’, Computers in Biology and Medicine, Vol. 36, No. 9, pp.1026–1040.
Intelligent techniques and applications in liver disorders 69
Mala, K. and Sadasivam, V. (2005) ‘Automatic segmentation and classification of diffused liver
diseases using wavelet based texture analysis and neural network’, Annual IEEE INDICON
Conference, 11–13 December, Chennai, India, pp.216–219.
Mezyk, E. and Unold, O. (2011) ‘Mining fuzzy rules using an artificial immune system with fuzzy
partition learning’, Applied Soft Computing, Vol. 11, No. 2, pp.1965–1974.
Ming, L.K., Kiong, L.C. and Soong, L.W. (2011) ‘Autonomous and deterministic supervised fuzzy
clustering with data imputation capabilities’, Applied Soft Computing, Vol. 11, No. 1,
pp.1117–1125.
Neshat, M., Yaghobi, M., Naghibi, M.B. and Esmaelzadeh, A. (2008) ‘Fuzzy expert system design
for diagnosis of liver disorders’, IEEE International Symposium on Knowledge Acquisition
and Modeling, 21–22 December, Wuhan, pp.252–256.
Neshat, M. and Zadeh, A.E. (2010) ‘Hopfield neural network and fuzzy Hopfield neural network
for diagnosis of liver disorders’, 5th IEEE International Conference on Intelligent Systems
(IS), 7–9 July, London, pp.162–167.
Obot, O.U. and Udoh, S.S. (2011) ‘A framework for fuzzy diagnosis of hepatitis’, IEEE World
Congress on Information and Communication Technologies (WICT), 11–14 December,
Mumbai, pp.439–443.
Obot, O.U. and Uzoka, F.M.E. (2009) ‘A framework for application of neuro-case-rule base
hybridization in medical diagnosis’, Applied Soft Computing, Vol. 9, No. 1, pp.245–253.
Ozsen, S. and Gunes, S. (2009) ‘Attribute weighting via genetic algorithms for attribute weighted
artificial immune system (AWAIS) and its application to heart disease and liver disorders
problems’, Expert Systems with Applications, Vol. 36, No. 1, pp.386–392.
Ozyilmaz, L. and Yildirim, T. (2003) ‘Artificial neural networks for diagnosis of hepatitis disease’,
Proceedings of the IEEE International Joint Conference on Neural Networks, 20–24 July,
Vol. 1, pp.586–589.
Parka, Y.J., Chuna, S.H. and Kim, B.C. (2011) ‘Cost-sensitive case-based reasoning using a
genetic algorithm: application to medical diagnosis’, Artificial Intelligence in Medicine, Vol.
51, No. 2, pp.133–145.
Perez, M.A., Marquez, C.Y., Nieto, O.C. and Cruz, A.J.A. (2012) ‘An associative memory
approach to medical decision support systems’, Computer Methods and Programs in
Biomedicine, Vol. 106, No. 3, pp.287–307.
Polat, K. and Gunes, S. (2007a) ‘A hybrid approach to medical decision support systems:
combining feature selection, fuzzy weighted pre-processing and AIRS’, Computer Methods
and Programs in Biomedicine, Vol. 88, No. 2, pp.164–174.
Polat, K. and Gunes, S. (2007b) ‘Medical decision support system based on artificial immune
recognition immune system (AIRS), fuzzy weighted pre-processing and feature selection’,
Expert Systems with Applications, Vol. 33, No. 2, pp.484–490.
Polat, K., Sahan, S., Kodaz, H. and Gunes, S. (2007) ‘Breast cancer and liver disorders
classification using artificial immune recognition system (AIRS) with performance evaluation
by fuzzy resource allocation mechanism’, Expert Systems with Applications, Vol. 32, No. 1,
pp.172–183.
Qasem, S.N. and Shamsuddin, S.M. (2011) ‘Radial basis function network based on time variant
multi-objective particle swarm optimization for medical diseases diagnosis’, Applied Soft
Computing, Vol. 11, No. 1, pp.1427–1438.
Ramirez, M.C., Martinez, C.H., Fernandez, J.C., Briceno, J. and Mata, M. (2012) ‘Multi-objective
evolutionary algorithm for donor–recipient decision system in liver transplants’, European
Journal of Operational Research, Vol. 222, No. 2, pp.317–327.
Revesz, P. and Triplet, T. (2010) ‘Classification integration and reclassification using constraint
databases’, Artificial Intelligence in Medicine, Vol. 49, No. 2, pp.79–91.
Revett, K., Gorunescu, F., Gorunescu, M. and Ene, M. (2006) ‘Mining a primary biliary cirrhosis
dataset using rough sets and a probabilistic neural network’, 3rd International IEEE
Conference on Intelligent Systems, September, London, pp.284–289.
70 A. Singh and B. Pandey
Rouhani, M. and Haghighi, M.M. (2009) ‘The diagnosis of hepatitis diseases by support vector
machines and artificial neural networks’, International Association of Computer Science and
Information Technology – Spring IEEE Conference, 17–20 April, Singapore, pp.456–458.
Sarkar, B.K., Sana, S.S. and Chaudhuri, K. (2012) ‘A genetic algorithm-based rule extraction
system’, Applied Soft Computing, Vol. 12, No. 1, pp.238–254.
Sartakhti, J.S., Zangooei, M.H. and Mozafari, K. (2012) ‘Hepatitis disease diagnosis using a novel
hybrid method based on support vector machine and simulated annealing (SVM-SA)’,
Computer Methods and Programs in Biomedicine, Vol. 108, No. 2, pp.570–579.
Stoean, C., Stoean, R., Lupsor, M., Stefanescu, H. and Badea, R. (2011a) ‘Feature selection for a
cooperative coevolutionary classifier in liver fibrosis diagnosis’, Computers in Biology and
Medicine, Vol. 41, No. 4, pp.238–246.
Stoean, R., Stoean, C., Lupsor, M., Stefanescu, H. and Badea, R. (2011b) ‘Evolutionary-driven
support vector machines for determining the degree of liver fibrosis in chronic hepatitis C’,
Artificial Intelligence in Medicine, Vol. 51, No. 1, pp.53–65.
Su, C.T., Chen, L.S. and Yih, Y. (2006) ‘Knowledge acquisition through information granulation
for imbalanced data’, Expert Systems with Applications, Vol. 31, No. 3, pp.531–541.
Su, C.T. and Yang, C.H. (2008) ‘Feature selection for the SVM: an application to hypertension
diagnosis’, Expert Systems with Applications, Vol. 34, No. 1, pp.754–763.
Sun, Y., Lu, J. and Yahagi, T. (2005) ‘Ultrasonographic classification of cirrhosis based on
pyramid neural network’, IEEE Canadian Conference on Electrical and Computer
Engineering, 1–4 May, Saskatoon, Sask, pp.1678–1681.
Tan, K.C., Yu, Q., Heng, C.M. and Lee, T.H. (2003) ‘Evolutionary computing for knowledge
discovery in medical diagnosis’, Artificial Intelligence in Medicine, Vol. 27, No. 2, pp.129–
154.
Torun, Y. and Tohumoglu, G. (2011) ‘Designing simulated annealing and subtractive clustering
based fuzzy classifier’, Applied Soft Computing, Vol. 11, No. 2, pp.2193–2201.
Uttreshwar, G.S. and Ghatol, A.A. (2009) ‘Hepatitis B diagnosis using logical inference and
generalized regression neural networks’, IEEE International Advance Computing Conference
(IACC), Patiala, India, 6–7 March, pp.1587–1595.
Wang, C.H., Hong, T.P. and Tseng, S.S. (1998) ‘Integrating fuzzy knowledge by genetic
algorithms’, IEEE Transactions on Evolutionary Computation, November, Vol. 2, No. 4,
pp.138–149.
Wu, C.C., Lee, W.L., Chen, Y.C., Lai, C.H. and Hsieh, K.S. (2012) ‘Ultrasonic liver tissue
characterization by feature fusion’, Expert Systems with Applications, Vol. 39, No. 10,
pp.9389–9397.
Yan, W., Lizhuang, M., Xiaowei, L. and Ping, L. (2008) ‘Correlation between child-pugh degree
and the four examinations of traditional Chinese medicine (TCM) with liver cirrhosis’, IEEE
International Conference on BioMedical Engineering and Informatics, 27–30 May, Sanya,
pp.858–862.
Zhang, Y. and Rockett, P.I. (2011) ‘A generic optimising feature extraction method using
multiobjective genetic programming’, Applied Soft Computing, Vol. 11, No. 1, pp.1087–1097.
  • ... Liver infection is a general term that covers all the potential issues that bring about the liver to neglect to perform its assigned capacities. This study mainly discusses about five types of liver diseases such as alcoholic liver damage (ALD), liver cirrhosis (LC), primary hepatoma (PM), cholelithiasis (C) [3] and HCC [4]. ...
    Chapter
    Full-text available
    Liver diseases have produced a big data such as metabolomics analyses, electronic health records, and report including patient medical information, and disorders. However, these data must be analyzed and integrated if they are to produce models about physiological mechanisms of pathogenesis. We use machine learning based on classifier for big datasets in the fields of liver to Predict and therapeutic discovery. A dataset was developed with twenty three attributes that include the records of 7000 patients in which 5295 patients were male and rests were female. Support Vector Machine (SVM), Boosted C5.0, and Naive Bayes (NB), data mining techniques are used with the proposed model for the prediction of liver diseases. The performance of these classifier techniques are evaluated with accuracy, sensitivity, specificity.
  • Article
    The logical thinking of medical practitioner involves a lot of subjective decision making and its complexity makes traditional quantitative approaches of analysis inappropriate. The computer based diagnostic tools and knowledge base certainly helps for early diagnosis of diseases. The intelligent decision making systems can appropriately handle both the uncertainty and imprecision. This paper discusses about the application potential of artificial intelligence in medical diagnosis. The fuzzy expert system has been presented specific to liver disease diagnosis.
  • Conference Paper
    A benchmark medical study is realized for a Primary Biliary Cirrhosis (PBC) dataset by using two different versions of a Bayesian Neural Network (BNN) entitled Partial Logistic Artificial Neural Network for Competing Risks with Automatic Relevance Determination (PLANN-CR-ARD). The two BNN versions are based on two different compensation mechanisms which are designed to preserve the numerical stability of the PLANN-CR-ARD model and to calculate the marginalized network results. The predictions of the PLANN-CR-ARD models are comparable to the non-parametric estimates obtained through the survival analysis of the PBC dataset. The input variables from the PBC dataset which can have a strong influence on the outcome of the disease are determined. The PLANN-CR-ARD models can be used to investigate the non-linear inter-dependencies between the predicted outputs and the input data which consist of the characteristics of the PBC patients.
  • Conference Paper
    Full-text available
    In computational anatomy, statistical shape model is used for quantitative evaluation of the variations of an organ shape. This paper is focused on construction of Statistical Shape Model of the liver and its application to computer assisted diagnosis. We prove the potential application of statistical shape models in classification of normal and cirrhosis livers. First, statistical shape model of liver is constructed. Then the coefficients of the model are used to recognize whether liver is normal or abnormal.
  • Conference Paper
    In this paper, we use support vector machine (SVM) and artificial neural networks to diagnosis hepatitis diseases. Furthermore, we use those networks to identify the type and the phase of disease. Considering the most important hepatitis cases leads us to six classes: hepatitis B (two phases), hepatitis C (two phases), non-viral hepatitis and no-hepatitis. For this purpose, we design various networks including RBF, GRNN, PNN, LVQ and SVM. The performance of each of them has studied and the best method is selected for each of classification tasks. The overall accuracy of diagnosis system is near 97%.
  • Conference Paper
    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.
  • Conference Paper
    Full-text available
    Liver biopsy is considered as mandatory for the management of patients infected with the hepatitis C virus (HCV), particularly for staging of fibrosis degree. However, due to its invasive nature and limitations of sampling error, the tendency is to substitute the liver biopsy with non-invasive method. The objective of this study is to combine the serum biomarkers and histopathological findings to develop a classification model that can predict the hepatic fibrosis stage. The best developed classification model was able to predict the different fibrosis grades with accuracy of 93.7%. This accuracy represents a substantial improvement over previous works and would pave the way to utilize classification models as a clinically non-invasive and reliable method to assess the degree of liver fibrosis.
  • Article
    In this paper, a new approach based on Differential Evolution (DE) for the automatic classification of items in medical databases is proposed. Based on it, a tool called DEREx is presented, which automatically extracts explicit knowledge from the database under the form of IF-THEN rules containing AND-connected clauses on the database variables. Each DE individual codes for a set of rules. For each class more than one rule can be contained in the individual, and these rules can be seen as logically connected in OR. Furthermore, all the classifying rules for all the classes are found all at once in one step. DEREx is thought as a useful support to decision making whenever explanations on why an item is assigned to a given class should be provided, as it is the case for diagnosis in the medical domain. The major contribution of this paper is that DEREx is the first classification tool in literature that is based on DE and automatically extracts sets of IF-THEN rules without the intervention of any other mechanism. In fact, all other classification tools based on DE existing in literature either simply find centroids for the classes rather than extracting rules, or are hybrid systems in which DE simply optimizes some parameters whereas the classification capabilities are provided by other mechanisms. For the experiments eight databases from the medical domain have been considered. First, among ten classical DE variants, the most effective of them in terms of highest classification accuracy in a ten-fold cross-validation has been found. Secondly, the tool has been compared over the same eight databases against a set of fifteen classifiers widely used in literature. The results have proven the effectiveness of the proposed approach, since DEREx turns out to be the best performing tool in terms of highest classification accuracy. Also statistical analysis has confirmed that DEREx is the best classifier. When compared to the other rule-based classification tools here used, DEREx needs the lowest average number of rules to face a problem, and the average number of clauses per rule is not very high. In conclusion, the tool here presented is preferable to the other classifiers because it shows good classification accuracy, automatically extracts knowledge, and provides users with it under an easily comprehensible form.
  • Article
    Ultrasound imaging is one of the most widely used imaging modality for the purpose of visualizing the human soft tissues. Especially, liver imaging application is of great importance in the areas of diagnostic ultrasound. In ultrasound liver image, the classification of lesions depends heavily on the characteristics of the lesions including internal echo, morphology, edge, echogenicity, and posterior echo enhancement. These characteristics are differently observed according to ROI selection methods that may indeed significantly impact the classification performances. Currently developed ROI selection methods have limitation for guaranteeing robust classification performance for focal liver lesions, mainly due to the inherent difficulties that represent all ultrasonic appearances of characteristics of lesion. In order to obtain better and more stable classification performances, we propose a new and novel approach, so-called multiple-ROI based focal liver lesion classification. The proposed approach properly combines the advantages of existing ROI selection methods to represent well various ultrasonic appearances of liver lesions including internal echo, morphology, edge, echogenicity, and posterior echo enhancement. To verify the effectiveness of the proposed ROI selection approach, extensive and comparative experiments have been performed using a total of 150 ultrasound images. Each ultrasound image contains one corresponding focal liver lesion so that a total of 150 focal liver lesions is used, comprising of 50 cysts, 50 hemangiomas, and 50 malignancies. Experimental results show that the proposed multiple-ROI-based approach can achieve the enhanced and stable classification performance regardless of features being used. In addition, our proposed method outperforms other existing classification methods designed for focal liver lesion classification. Especially, the proposed approach attains classification accuracy of up to 80% over well-known challenging task of classifying the hemangiomas and malignancies.
  • Article
    Artificial neural networks (ANNs) are mathematical models inspired from the biological nervous system. They have the ability of predicting, learning from experiences and generalizing from previous examples. An important drawback of ANNs is their very limited explanation capability, mainly due to the fact that knowledge embedded within ANNs is distributed over the activations and the connection weights. Therefore, one of the main challenges in the recent decades is to extract classification rules from ANNs. This paper presents a novel approach to extract fuzzy classification rules (FCR) from ANNs because of the fact that fuzzy rules are more interpretable and cope better with pervasive uncertainty and vagueness with respect to crisp rules. A soft computing based algorithm is developed to generate fuzzy rules based on a data mining tool (DIFACONN-miner), which was recently developed by the authors. Fuzzy DIFACONN-miner algorithm can extract fuzzy classification rules from datasets containing both categorical and continuous attributes. Experimental research on the benchmark datasets and comparisons with other fuzzy rule based classification (FRBC) algorithms has shown that the proposed algorithm yields high classification accuracies and comprehensible rule sets.
  • Article
    A generalisation of bottom-up pruning is proposed as a model level combination method for a decision tree ensemble. Bottom up pruning on a single tree involves choosing between a subtree rooted at a node, and a leaf, dependant on a pruning criterion. A natural extension to an ensemble of trees is to allow subtrees from other ensemble trees to be grafted onto a node in addition to the operations of pruning to a leaf and leaving the existing subtree intact. Suitable pruning criteria are proposed and tested for this multi-tree pruning context. Gains in both performance and in particular compactness over individually pruned trees are observed in tests performed on a number of datasets from the UCI database. The method is further illustrated on a churn prediction problem in the telecommunications domain.