Intelligent Model to Predict Early Liver Disease

using Machine Learning Technique

Taher M. Ghazal
Network and Communication Aziz Ur Rehman Muhammad Saleem
Technology Lab, Center for Cyber
2022 International Conference on Business Analytics for Technology and Security (ICBATS) | 978-1-6654-0920-9/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICBATS54253.2022.9758929

School of Computer Science, School of Computer Science,

Security, Faculty of Information NCBA&E NCBA&E.
Science and Technology UKM, Lahore, Pakistan Lahore, Pakistan
Malaysia [email protected] [email protected]
[email protected]

Munir Ahmad
Faisal Mehmood,
School of Computer Science, Shabir Ahmad Department of IT Convergence
NCBA&E Department of Computer Engineering, Engineering, Gachon University
Lahore, Pakistan Gachon University Korea
[email protected] Korea [email protected]
[email protected]

Abstract—Liver Disease (LD) is the main cause of death learning has had a important influence on biomedical
worldwide, affecting a large number of people. A variety of research [7-9]. Improved detection and prediction of disease
factors affect the liver, resulting in this disease. The diagnosis of in the biomedical field can be achieved through the use of
this condition is both expensive and time-consuming. Machine machine learning techniques that are both objective and
Learning offers a lot of potential in terms of automated disease accurate [10].
diagnosis. As a result, the purpose of this research is to assess
the efficacy of various Machine Learning (ML) algorithms to Problems in medicine can be solved quickly and accurately
lower the high cost of liver disease diagnosis through prediction. using machine learning techniques. Getting better at
With the current rise in numerous liver disorders, it's more predicting results and lowering medical diagnostic costs are
important than ever to detect liver disease early on. This the primary goals of this research. Because of this, we used a
research proposed intelligent model to predict liver disease variety of classification techniques to determine whether or
using machine learning technique. This proposed model is more not a patient had liver disease. Various machine learning
effective and comprehensive in terms of performance of 0.884 techniques, such as KNN, SVM, etc. have been used, and the
accuracy, and 0.116 miss-rate. concert of these methods has been evaluated on numerous
metrics.
Keywords—CLD, ML, GBD, DL, FNS
Deep Learning (DL) is most enhanced model of ML. Current
I. INTRODUCTION years have seen the rise of DL in image recognition, speech
When it comes to digesting food, the liver is the utmost recognition, also natural language processing compared to
important organ in the human body. It is the combination of traditional machine learning [11].
viruses and alcohol use that damages the liver and puts a Convolutional Neural Networks (CNNs) are a new class of
person in a life-threatening situation. Liver diseases include DL algorithms that may be applied to medical image analysis
cirrhosis and other forms of hepatitis, as well as tumors and because they improve the enactment of artificial intelligence
cancers. Cirrhosis, a disease of the liver, is the leading cause and ML applications. There are multiple layers in CNNs, and
of death in this group [1]. Because of this, liver disease is a each layer is connected to the other by a minimal amount of
major public health issue around the world. Around 2 million processing. ML algorithms that mimic the arrangement of an
people die each year from liver disease around the world [2]. animal's graphic cortex are regarded as extremely powerful
Global Burden of Disease (GBD) scheme, issued in BMC and useful. There are many advantages to the CNN technique,
Medicine, estimates that one in four deaths from cirrhosis and including the fact that it practices raw figures as input rather
one million deaths from liver cancer occur every year in 2010 than requiring a manual withdrawal step like traditional ML
[3]. Every year, approximately 400,000 patients in China lose methods, and it is more accurate than traditional classification
their lives to liver diseases, which affect approximately 300 methods [12-13].
million people [4]. By Marcellin and others, liver disease is
one of the world's most under-recognized public health In only a rare clinical CLD studies to date, deep learning
issues, requiring immediate action and large-scale screenings techniques have been employed. Transfer learning also Fully
[5]. Connected Network (FCNet) approaches were applied to US
liver images by Meng et al. in this study [14].
Chronic HCV impurity is typically a slow, enlightened
disease with insufficient or no indications for numerous years II. LITERATURE REVIEW
following infection. Approximately patients are infected for
Several studies have been done in the past few years on liver
a long time and do not suffer any important liver injury, while
disease diagnosis. According to Rajesh and colleagues, an
others rapidly improvement to liver cirrhosis also may mature optimal hierarchical feature fusion using PeSOA algorithm is
hepatocellular carcinoma [6].
used to categorize liver cancer concerns using a Probabilistic
Predicting liver disease can be difficult. For the purposes of Neural Network (PNN) [15].
early detection and diagnosis of liver disease, machine

978-1-6654-0920-9/22/$31.00 ©2022 IEEE

Authorized licensed use limited to: Fatima Jinnah University. Downloaded on June 16,2023 at 07:46:33 UTC from IEEE Xplore. Restrictions apply.
Models for the finding of liver illnesses are built using a Traditional serum-based biomarker indices like APRI,
Fuzzy Neural System (FNS). In order to distinguish between Fibrosis-4 (FIB-4), AST to ALT ratio, and NAFLD fibrosis
normal and disordered cells, they applied a regularized set of score were used in medical care to find the momentous
data to the FNS classifier [16]. fibrosis as well as cirrhosis in patients previously analyzed by
various etiologies of chronic liver disease, with diffident
Biswas et al. planned a DL framework for perceiving Fatty
accuracy and a high percentage of indeterminate scores.
Liver Disease, which they tested on a large dataset. They
These tests have not been found to be suitable for routine use
compared their system to traditional ML systems such as in cirrhosis screening because, while their specificity is
Support Vector Machine(SVM) and Evolutionary Learning
adequate, their sensitivity is inadequate. Based on blood tests,
Machine(ELM) [17]. Venkata et al. compared the
ML algorithms have been used to sense fibrosis or cirrhosis
performance of some classification algorithms on dissimilar secondary to various liver disease etiologies [21-27].
factors that cause liver disease, including SGOT, SGPT, and
ALP, in their research [18]. Kaur et al. used six ML To efficiently classify patients with liver disease, machine
classification algorithms, ranked and averaged the results, learning algorithms have been used for pattern recognition.
and discovered that Bagging and J48 classifiers work greatest Vanderbeck et al used 47 unique liver biopsy images to
on their dataset [19]. manual comments through two pathologists to create a
classification algorithm in an effort to automate disease
Multiple liver diseases have been identified, predicted, and
staging. The algorithm was able to recognize key liver biopsy
assessed using machine learning techniques. Studying the
sorts (macro steatosis, bile ducts, portal veins, also sinusoids)
still images acquired from Shear Wave Elastography (SWE) using a color analysis protocol, with a recall rate of >82
and classifying patients via the inclusion or exclusion of
percent [28].
essential liver disease, researchers used a support vector
machine classification algorithm. Using an inverse mapping III. METHODOLOGY
sequence that correlates quantitatively strong color data with
gained stiffness standards, this algorithm was clever to There may be more cures or a longer life expectancy if
precisely distinguish healthy topics from those by chronic disease is found earlier. This prospect has led to public health
liver disease at 87.3%, by a sensitivity of 93.5% and a services that suggest screening populations for precise live
specificity of 81.2%, when associated to standard gold liver diseases. In this research work an intelligent model is being
biopsy [20]. proposed to predict liver disease more efficiently by using
machine learning technique.
Fig. 1. Proposed model for Liver Disease Prediction

Figure 1 shows the that the proposed model in training phase parameters to collect the data from liver patient and store it
has data collection, data preprocessing layer and training into a database. These stored data may involve missing data
model. The data collection layer contains various input or noisy data as the communication source is Internet of thing

Authorized licensed use limited to: Fatima Jinnah University. Downloaded on June 16,2023 at 07:46:33 UTC from IEEE Xplore. Restrictions apply.
(IoT). The following layer is the data preprocessing layer that
is very significant layer in which activity handle the missing (8)
values through moving average as well as normalization to
eliminate the noisy data. Later this procedure output of the
preprocessing is referred to the training model, in which where,
Artificial Neural Network (ANN) algorithm of ML is used to
identify the performance that if learning criteria meet is
fulfilling means yes, then go to data cloud computing Use the chain rule to update the weights between the input
database. If learning criteria is not fulfilling means no, then and hidden layers.
the training model will be retrained and so on.
Using ANN Proposed liver disease detection system total,
three layers were used: input, hidden, and output. The back
propagation algorithm, which comprised initialization of
weight, feed forward, back propagation of error, and update
of weight and bias, has made significant progress. The
activation function of each neuron in the hidden layer is f(x)
=Sigmoid (x). The proposed liver disease detection system's
sigmoid function is written as
In above equation, represents the constant,
(1)

(2)

The Input and output layer is taken

(3)

The following equation represent back propagation error

where, represents the output that is
desired and output that is estimated activation function for the After simplification the equation above can be
output layer is shown below.
(9)
(4)
where,

(5)

The layer is written as in equation 6 rate of change in weight

for the output.
(10)

Above equation is used to modernize the weights between the

hidden also output layers. The weights among the hidden and
input layers are updated using the equation below.
(6)

(11)
Equation 6 can be written as follows after applying the Chain
Rule approach
In validation phase the stored learned data is imported from
cloud to predict the liver disease. It will be checked if disease
(7) is found or not. If yes message will be displayed that disease
is found in case of No procedure will be discarded.
Equation 8 can be generated by replacing the values in
equation 7 for the value of weight changed. IV. SIMULATION RESULTS
In this research an intelligent disease prediction model
entangled with machine learning approaches is proposed in

Authorized licensed use limited to: Fatima Jinnah University. Downloaded on June 16,2023 at 07:46:33 UTC from IEEE Xplore. Restrictions apply.
which Artificial Neural Network (ANN) algorithm is being
used to predict the liver disease in intelligent way. The dataset (15)
is collected from UCI Machine Learning data repository,
which is used to predict real time patient data either disease
is found or not.
(16)
TABLE 1: TRAINING OF THE PROPOSED MODEL THROUGHOUT THE
PREDICTION OF LIVER DISEASE (ANN)
Proposed Model Training (17)
Total samples
Result
(45872)

Expected Predicted Predicted

(18)
output Positive Negative

True Positive False Positive

(TP) (FP)
(19)
Input
8960 Positive 7031 1929

False Negative True Negative (20)

(FN) (TN)
It is shown in table 1 that the proposed system prediction of
36912
Negative
1886 35026 liver disease throughout the training period. During training,
a total of 45872 samples are used, which are separated into
8960,36912 positive and negative samples, respectively.
TABLE 2: VALIDATION OF THE PROPOSED MODEL THROUGHOUT THE There are 7031 true positives that are successfully predicted
PREDICTION OF LIVER DISEASE (ANN) and no liver disease is identified, but 1929 records are
mistakenly predicted as negatives, indicating liver disease.
Proposed Model Validation
Similarly, 36912 samples are obtained, with negative
showing liver disease and positive indicating no liver disease,
Total samples
(19660)
Result with 35026 samples properly identified as negative showing
intrusion and 1886 samples inaccurately predicted as positive
Expected Predicted Predicted indicating no liver disease despite the existence of liver
output Positive Negative disease.
True Positive False Positive The proposed model predicts intrusion during the validation
(TP) (FP) phase, as shown in table 2. Throughout validation, a total of
Input 19660 samples are used, which are separated into 3891,15769
3891 Positive 2879 1012 positive and negative samples, respectively. It is determined
that 1012 samples have true positives that are successfully
False Negative True Negative predicted and no liver disease is found, however 2879 records
(FN) (TN) are mistakenly predicted as negatives, showing liver disease.
Likewise, a total of 15769 samples are gathered, with
15769
1251 14518 negative specifying liver disease and positive indicating no
Negative
liver disease, with 14518 samples correctly predicted as
ANN approach is used on the dataset of 65532sets of records; negative indicating liver disease and 1251 samples
moreover, the dataset is separated into training organizes of imperfectly identified as positive representing no liver
70% (45872 samples) and 30% (19650 samples) for the stated disease found the existence of liver disease.
resolves training, and validation. Different measures used for
TABLE 1: PERFORMANCE OF PROPOSED LIVER DISEASE DETECTION SYSTEM
performance calculation with different metrics are resultant IN TRAINING AND VALIDATION (ANN)
through the formulas:
Mis
s- Fall- PPV
Sensi Speci
Accura Rate out LR (Pre NP
(12) ANN
cy
tivity ficity
(%) FPR +
LR-
cisio V
TPR TNR
FN n)
R
0.08 0.05 15. 0.08 0.78 0.94
(13) Training 0.916 0.784 0.948
4 1 37 8 8 7
Validatio 0.11 0.07 9.3 0.12 0.69 0.93
0.884 0.739 0.920
n 6 9 5 6 7 4

(14) It is shown in tables 3 (ANN) that proposed system learning

technique performance in term of accuracy, TPR, TNR, FNR
and PPV throughout the training and validation phase. It
clearly displays that the proposed system throughout training
gives 0.916, 0.784, 0.948, 0.084, 0.051, 15.37, 0.088, 0.788

Authorized licensed use limited to: Fatima Jinnah University. Downloaded on June 16,2023 at 07:46:33 UTC from IEEE Xplore. Restrictions apply.
and 0.947 in term of accuracy TPR, TNR, FNR and PPV [12] Brattain LJ, Telfer BA, Dhyani M, Grajo JR, Samir AE. Machine
respectively. And throughout validation the purposed model learning for medical ultrasound: status, methods, and future opportunities
[published online ahead of print 2018/03/02]. Abdom Radiol (NY).
gives 0.884, 0.793, 0.920, 0.116, 0.079, 09.35, 0.126, 0.679 2018;43(4):786-799.
and 0.934 in term of accuracy TPR, TNR, FNR and PPV
[13] Litjens G, Kooi T, Bejnordi BE, et al. A survey on deep learning
respectively. In addition, some more statistical measure of in medical image analysis [published online ahead of print 2017/08/05]. Med
purposed system is added to identify the values such as fall Image Anal. 2017; 42:60-88.
out likelihood positive ratio, likelihood negative ratio and [14] Meng D, Zhang L, Cao G, Cao W, Zhang G, Hu B. Liver fibrosis
negative predict value. classification based on transfer learning and FCNet for ultrasound images.
IEEE Access. 2017; 5:5804-5810.
It is clearly shown the performance of the proposed model
[15] G, R. and Priyadharson A, S. (2018). Liver cancer detection and
throughout prediction of early liver disease in terms of better classification based on optimum hierarchical feature fusion with PeSOA and
accuracy, and miss-rate is 88.4, and 0.116 respectively. PNN classifier. Biomedical Research, 29(1).
[16] Spector, A. Z. 1989. Achieving application requirements. In
V. CONCLUSION Distributed Systems, S. Mullender, Ed. ACM Press Frontier Series. ACM,
Preventing complications and rapid worsening can be New York, NY, 19-33. DOI= http://doi.acm.org/10.1145/90417.90738.
prevented with early detection of the liver disease and a [17] Biswas M., Kuppili V., Edla D.R., Suri H.S., Saba L., Marinhoe
precise diagnosis. With an accurate diagnosis, it's easier to R.T., Sanches J.M., Suri J.S. (2018) Computer Methods and Programs in
Biomedicine, 155 , pp. 165-177.
decide on the best course of treatment for the patient.
Therefore, in this research work an intelligent model is [18] VenkataRamana, B., Babu, M. and Venkateswarlu, N. (2011). A
Critical Study of Selected Classification Algorithms for Liver Disease
introduced to predict the early liver disease by using machine Diagnosis. International Journal of Database Management Systems, 3(2),
learning (ANN) approach. The proposed model shows better pp.101-114.
performance in terms of 88.4 accuracy, and 0.116 miss-rate. [19] Kaur, A. and Kaur, I. (2014). Empirical Evaluation of Machine
Learning Algorithms for Fault Prediction. Lecture Notes on Software
Funding: This research is supported by the National
Engineering, pp.176-180.
Research Foundation of Korea (NRF) Grant funded by the
[20] Gatos I, Tsantis S, Spiliopoulos S, et al. A Machine-Learning
Korean government under Grant NRF- Algorithm Toward Color Analysis for Chronic Liver Disease Classification,
2021R1I1A1A01045177. Employing Ultrasound Shear Wave Elastography. Ultrasound Med Biol.
2017;43(9):1797-1810.
REFERENCES
[21] Wei R, Wang J, Wang X, et al. Clinical prediction of HBV and
[1] K. Sumeet, J.J. Larson, B. Yawn, T.M. Therneau, W.R. Kim, HCV related hepatic fibrosis using machine learning. EBioMedicine.
Underestimation of liver-related mortality in the United States. 2018;35:124-132.
Gastroenterology; (2013) 145:375–382, e371–372. [22] Cao Y, He K, Cheng M, et al. Two classifiers based on serum
[2] A.A. Mokdad, A.D. Lopez, S. Shahraz, R. Lozano, A.H. Mokdad, peptide pattern for prediction of HBV induced liver cirrhosis using MALDI-
J. Stanaway, et al, Liver cirrhosis mortality in 187 countries between 1980 TOF MS. Biomed Res Int. 2013;2013:814876.
and 2010: a systematic analysis. BMC Med 2014; 12:145. [23] Batool, T., Abbas, S., Alhwaiti, Y., Saleem, M., Ahmad, M., Asif,
[3] Byass, Peter, The global burden of liver disease: a challenge for M. and Elmitwally, N.S., 2021. Intelligent Model Of Ecosystem For Smart
methods and for public health. BMC medicine 12.1 (2014); 159. Cities Using Artificial Neural Networks. INTELLIGENT AUTOMATION
[4] F.-S. Wang, J.-G. Fan, Z. Zhang, B. Gao, H.-Y. Wang, The global AND SOFT COMPUTING, 30(2), pp.513-525.
burden of liver disease: the major impact of china, Hepatology 60 (6) (2014) [24] Saleem, M., Khan, M.A., Abbas, S., Asif, M., Hassan, M. and
2099–2108. Malik, J.A., 2019, July. Intelligent FSO link for communication in natural
[5] P. Marcellin, B. K. Kutala, Liver diseases: A major, neglected disasters empowered with fuzzy inference system. In 2019 International
global public health problem requiring urgent actions and large-scale Conference on Electrical, Communication, and Computer Engineering
screening, Liver 340 International 38 (S1) (2018). (ICECCE) (pp. 1-6). IEEE.
[6] T.J. Liang, B .Rehermann, L.B. Seeff, J.H. Hoofnagle. [25] Yip TC, Ma AJ, Wong VW, et al. Laboratory parameter-based
“Pathogenesis, natural history, treatment, and prevention of hepatitis C.”. machine learning model for excluding non-alcoholic fatty liver disease
Ann Intern Med, pp132:296, vol.305, 2000. (NAFLD) in the general population. Aliment Pharmacol Ther.
2017;46(4):447-456.
[7] L. A. Auxilia, Accuracy Prediction Using Machine Learning
Techniques for Indian Patient Liver Disease. 2018 2nd International [26] Siddiqui, S.Y., Athar, A., Khan, M.A., Abbas, S., Saeed, Y.,
Conference on Trends in Electronics and Informatics (ICOEI). IEEE (2018). Khan, M.F. and Hussain, M., 2020. Modelling, simulation and optimization
of diagnosis cardiovascular disease using computational intelligence
[8] Hashem, M. Esraa, S. Mai, A study of support vector machine
approaches. Journal of Medical Imaging and Health Informatics, 10(5),
algorithm for liver disease diagnosis. American Journal of Intelligent
pp.1005-1022.
Systems 4.1 (2014); 9-14.
[27] Islam MM, Wu CC, Poly TN, Yang HC, Li YJ. Applications of
[9] P. Sajda, "Machine learning for detection and diagnosis of
Machine Learning in Fatty Live Disease Prediction. Stud Health Technol
disease." Annu. Rev. Biomed. Eng. 8 (2006); 537-565.
Inform. 2018;247:166-170.
[10] S. M. Mahmud, et al. "Machine Learning Based Unified
[28] Pournik O, Dorri S, Zabolinezhad H, Alavian SM, Eslami S. A
Framework for Diabetes Prediction." Proceedings of the 2018 International
diagnostic model for cirrhosis in patients with non-alcoholic fatty liver
Conference on Big Data Engineering and Technology. ACM (2018).
disease: an artificial neural network approach. Med J Islam Repub Iran.
[11] Ihnaini, B., Khan, M.A., Khan, T.A., Abbas, S., Daoud, M.S., 2014;28:116.
Ahmad, M. and Khan, M.A., 2021. A smart healthcare recommendation
system for multidisciplinary diabetes patients with data fusion based on deep
ensemble learning. Computational Intelligence and Neuroscience, 2021.

Authorized licensed use limited to: Fatima Jinnah University. Downloaded on June 16,2023 at 07:46:33 UTC from IEEE Xplore. Restrictions apply.