International Journal of Computer Applications (0975 – 8887)

Volume 184 – No.6, April 2022

Machine Learning Algorithm Early Detection of Liver

Cancer: A Review

Simran Jain Madan Lal Saini, PhD

Department of Computer Science Department of Computer Science
Poornima University, Jaipur, Rajasthan Poornima University, Jaipur, Rajasthan

ABSTRACT Keywords
One of the most highly prevalent cancers today is liver Machine Learning, Liver Cancer, Feature Selection,
cancer.The segmentation of a liver tumour is a critical step Accuracy
in making an early detection and recommending a
treatment. It has always been tedious to segment data by 1. INTRODUCTION
hand, so cancer detection techniques now use a variety of The liver is the largest in the body, located above and to
machine learning algorithms, such as decision trees, the left of the stomach and directly beneath the lung. It
Support Vector machine, artificial neural networks, performs many vital functions such as the production of
random forests, Logistic Regressions and genetic bile, metabolism, and absorption of bilirubin helps in
algorithms. These algorithms are all used in the cancer blood clotting, metabolism of fats and carbohydrates,
detection process. The purpose of this review article is to stores vitamins and minerals, produces albumin. The liver
conduct a comprehensive and comparative analysis of is a very complex organ, so that it can experience multiple
machine learning algorithms for diagnosing and predicting problems. So, the consequences of the diseased liver can
liver cancer in the medical field, which have already been be perilous and it can lead to a variety of illnesses,
used to predict liver disease by a number of authors, and to including alcoholic liver disease, fascioliasis, cirrhosis,
highlight the most frequently used features, classifiers, hepatitis, fatty liver disease, and liver cancer.[1] The below
techniques, fundamental ideas, and accuracy. figure shows structure of liver in fig. 1.

Fig 1:Structure of liver

The rate of incidence and death due to cancer of the liver non-alcoholic fatty liver disease, aflatoxins exposure, and
has been increasing steadily. According to cancer statistics, persistent alcoholism are all risk factors for liver cancer.
there have been 18.1 million additional cancers diagnosed As a result, early cancer detection and therapy are critical
and 9.6 million cancer deaths in 2018. Asia accounts for challenges.[3]Blood tests, liver biopsy, CT scans, MRI,
over half among all cancer cases as well as more than half ultrasounds can be done to detect liver disease or cancer.
of cancer deaths worldwide, according to these figures. By After inspecting all of these, doctors or practitioners decide
2020, liver cancer will rank sixth among all cancers whether is always a difficult task, as it requires
diagnosed globally and would be the third largest cause of experienced physicians and takes ample amount of time.
death.[2]Most of the time, liver cancer does not show any Machine learning plays a vital function in disease
symptoms till the liver damage is extensive. But if any diagnosis and treatments. It can help to extract valuable
symptoms occur, they may include loss of weight and information from medical datasets and build a model
appetite, pain in upper abdominal or abdominal swelling, toidentify the patients. For instance, various research has
yellowing of the skin(jaundice), nausea, vomiting, etc. been conducted to determine the prevalence of liver
Liver cancer happens due to mutations in DNA, which problems in patients using machine learning and data
results in cell growth in an Uncontrolled manner HBV and mining techniques. [4][5]
HCV infection, cirrhosis, inherited liver illnesses, diabetes,

42
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.6, April 2022

Fig. 2: Progression of liver damage

Fig. 3:Steps of Machine Learning Algorithm

2. MACHINE LEARNING value to be used as a classifier to differentiate between two

classes. For instance, let us say the expectation value of a
ALGORITHMS data instance is more than 0.25, so if it comes below that, it
Machine learning employs algorithms in their most basic will be under class A, otherwise class B.
form, which study incoming data, make reasonable
forecasts, and improve their operations. These algorithms 2.2. Decision tree
generate more precise estimates as they are fed with It is among the oldest and best-known ML algorithms.
additional data. We can classify Machine Learning Contains decision logic for classifying tree-structured data
Algorithms into three part that is supervised, semi- items such as tests and results. A node in a DT tree often
supervised and unsupervised.The base algorithm is trained has multiple levels, with the root node acting as the
in monitored machine learning techniques through a primary node. Each internal node represents an input
named training data set. Then, the unlabelled test data set variable or property test (at least one child).[6]
is placed in the trained algorithm and categorized into like Continues the test and branching procedure until the leaf
categories. Classification and regression problems can both criteria is met. Finally, the leaf or terminal node represents
be solved using supervised learning methods. In the first the selection's outcome. As a result, they are easy to
type of problem variable or the output can be categorized understand and learn and are used in various medical
into group or classes, that is in classification, the output is diagnostic procedures. Furthermore,when traversing the
always discrete in nature. For example, 'zero' or 'one' or tree for classification, all test results for each node provide
'unhealthy' and 'healthy.' In regression issues, the sufficient knowledge to make expert decisions about the
associated output variable is an exact value, such as an class in the sample.[7]
individual's risk of developing liver cancer. In the figure,
the basic steps of Machine Learning have described.The
generally used machine learning classification techniques 2.3. Support Vector Màchine
for disease forecast are briefly described in the subsections SVM is a machine learning supervised learning technique
below. that uses a mathematical approach to handle classification
and regression issues. By using hyperplanes, SVM divides
2.1. Logistic regression data points plotted in an n-dimensional space into
categories. To classify data points, the margin between the
It is a supervised technique for classification that is
hyperplanes must be increased. There are a variety of
securely verified and it is an expansion of traditional
kernels for separating linear and nonlinear data points in a
regression.But it can only model binary variables that
multidimensional space.
show the occurrence or non-occurrence of particular
category events. Hence, we need to assign a threshold

43
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.6, April 2022

2.4. Random Forest conjunction with an optimization technique such as

We can say that it is a collection of decision tree which gradient descent, the MLP network employs
uses bagging method. By applying an average model backpropagation (BP).This method of supervised learning
technique, bagging is utilized to increase the model's is commonly recognised as the most effective way to train
stability and accuracy. Each tree is made up of many neural networks. The BP technique identifies errors by
random vectors that can be used to vote for the best class estimating the differences among expected and calculated
to predict. The addition of randomization to the model values. After that, the error is transmitted back through the
stops it from overfitting, resulting in a more accurate network, revising the weights and getting the best
classification analysis conclusion.[8] outcomes. The technique is monitored until the difference
reaches a pre-set threshold. The neural network procedure
has gained significant the most consideration as an
2.5. K Nèarest Neighbours efficient and accurate approach that produces results
The K-Nearest Neighbors strategy is a distribution-free considered superior to those obtained using conventional
approach for pattern recognition classification and techniques due to its autonomy from whatever prior
regression. In both cases, the input is the k nearest practise knowledge of the nature of interactions among inputs and
instances in the feature space.When it is used for output.[13]
categorization, the analysis performed on the data
determines class membership. A piece of property is
classified by a majority vote of its immediate 2.9. Gradient Boosting
neighbours.[9] The object is assigned to the most often The learning mechanisms of the Gradient Boosting
occurring class among its k nearest neighbours, where k is Machines (GBM) adapt the new models in a sequence to
a positive number that is typically small. If k equals 1, the allow a more accurate estimation of the response variables.
item is assigned to the class of its single nearest neighbour. The main idea of this algorithm is to create new base
If the K-NN is employed as a regression model, the learners that are maximally correlated with the negative
outputs of the KNN are the object's attribute values. This is gradient of the set loss function. The loss functions used
the average of the k closest neighbours. can be any; However, to give you an idea, if the error
function is the traditional squared error loss, the learning
approach leads to a gradual adjustment of the error. In
2.6. Naive Bayes general, the researcher has complete control over the loss
It is a set of simple classifiers based upon implementing function, with a variety of previously derived loss
the Bayes hypothesis to features that must have crucial functions and the ability to develop activity-specific losses.
(naive) independence. One of the most fundamental Thanks to their great versatility, GBMs can be adapted to
Bayesian network models, it has been the subject of any data-driven activity. On the other hand, it leaves a lot
extensive research since the 1960s. In the early 1960s, it of room for the model designer, so choosing the best loss
was originally introduced to the text retrieval community function is a matter of trial and error.[14]
(but not under that name). It can also be utilised in
automatic medical diagnosis. Naive Bayes classifiers 3. LITERATURE REVIEW
demand a linear association between the frequency of
parameters and the number of factors in a training Rahman et al. [15] researched to determine the usefulness
challenge. [10] of numerous machine learning techniques have been
successful in lowering the primary prediction costs
associated with the identification of chronic liver
2.7. C4.5 illnesses.Six distinct algorithms were employed in this
C4.5 is based on the entropy assessment of the Information study: decision tree, support vector, bayesian networks,
Gain Ratio (IG). The IG report is used to select a test logistic regression, and random forest. The accuracy,
attribute from each node in the tree. The test task is precision, recall, f-1 score, and specificity of numerous
characterized by the highest percentage of information categorization algorithms were evaluated. The LR shows
gain ratio for the current node.[11] highest accuracy of 75%, then RF 74%, DT 69%, SVM
64%, KNN 62%, and least accuracy was shown by NB
2.8. MLP 53%., The research revealed that the LR had the best
McCulloch and Pitts established the Artificial Neural accuracy.
Network (ANN) in 1943. The ANN is a complex computer
model biologically based nervous systems. It is composed Rabbi et al. [4] For classifying the Indian Liver Patient
of numerous basic components called neurons that are Dataset, we used four distinct types of MLA: Logistic
organised into input, hidden, and output layers. Each Regression, Decision Tree, Random Forest, and Extra
neuron in the network has its own set of input values, a Trees (ILPD). Pearson Correlation Coefficient based
barrier, and an activation function. Changing consequences feature selection (PCC-FS) is applied to eliminate
connect neurons in different layers according to a learning irrelevant features from the dataset. Along with it to
process, resulting in a particular desired output from a increase the predictive performance AdaBoost one of the
specific input. Neurons are frequently combined in the boosting algorithms was used. The analysis is evaluated in
most straightforward artificial neural network to allow data respect of accuracy, ROC, F-1 score, precision, &
to flow from input neurons to hidden neurons and work. A recall.After comparing experimental results, we have
multilayer perceptron (MLP) is indeed a form of a found that boosting on ET provides the highest accuracy of
convolutional neural network composed of many layers of 92.19%.
neurons that are entirely coupled.[12] MLP networks can
approximate and satisfy universal approximation for any Cao et al. [16] proposed an early test method using the
form of constrained piece - wise linear combination with clinical laboratory data set. GBDT i.e., Gradient boosting
sufficient training and hidden layer numbers. When used in trees, was used to select essential features, followed by

44
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.6, April 2022

training and testing. Two type of Classification algorithm marker-controlled floodplain fragmentation and a Gaussian
was used one is Support Vector Machine and other GBDT. mixture model to differentiate cancer-affected regions.
The Kappa index approaches near perfection, and the Finally, 99.38 percent accuracy was reached for
accuracy exceeds 90%, according to the findings. classification using DNN.
Das et al.[17]. MLP and decision tree classifiers were used
to analyse CT images of hepatocellular and metastatic Kalsoom et al. [24]proposed an combination of
carcinoma. Fourier analysis was employed to extract unsupervised machine learning technique and supervised
important features from the LBP histogram. Finally, the mechanism for properly segmenting tumours in liver. The
decision tree classifier achieved 95.02 percent accuracy features like LBP and HOG have been extracted, and
out of the two machine learning classifiers. classification is done using KNN. The overall accuracy of
97% is achieved. As compared to SVM and Ensemble,
Meng et al. [18] examined MRI images of liver cancer. At which have shown the accuracy of 85% and 49%,
first, Histogram based feature has been used for the respectively, KNN has performed better.
extraction of feature, then for the prediction SVM was
used. This technique predicted primitive stage liver cancer Jacob et al. [25] compared four algorithms SVM, Logistic
with an accuracy of 86.67 percent and benign type tumours Regression, KNN, and ANN. Performance was calculated
with an accuracy of 80 percent. using different evaluation metrics. The ANN fared the
best, with a precision of 98 percent.
Naeem et al. [19] The researchers analysed a fused two-
dimensional dataset of computed tomography and Saritha et al. [26]proposed a new algorithm called
magnetic resonance images in benign hepatocellular separation of points by plane, it uses a technique of maths
adenoma and hemangioma cysts, as well as malignant to separate N number of data points by a certain number of
hepatocellular carcinoma and hepatoblastoma liver hyperplanes. The data set used was the Real Data Set of
cancers. The received dataset was pre-processed, and LFT, taken by the “MEDCIS PathLabs India Private Ltd.”
Gabor filters were employed to minimise noise. Hyderabad. The proposed method can predict the disease
Ten significant hybrid features were identified using the with 85.1% accuracy with training and testing time of 1
likelihood of error plus average correlation feature sec, respectively.
selection technique. Then tenfold cross-validation
procedure was used procedure to deploy this improved Roslina et al. [27]The author used the carrier vector
hybrid feature dataset to ML classification algorithm that machine (SVM) and the wrapper method to predict the
is Support Vector Machine, J48, multilayer perceptron prognosis of hepatitis disease. Prior to the classification
(MLP), and random forest (RF). MLP had 95.78 percent process, they used packaging methods to remove the noise
MRI accuracy and a 97.44 percent CT accuracy overall. characteristics. First, SVM selects features to achieve
Out of all four, MLP has achieved 99 percent accuracy. better accuracy. Feature selection is implemented to reduce
noise or irrelevant data. Through the experimental results,
Krishna et al. [20] CT images were preferred over the they saw an increased accuracy rate of the cost of clinical
more often utilised Positron - Emission and Magnetic laboratory tests with minimal run time. He achieved the
Resonance because they produce higher-quality images goal by combining wrapper method and SVM techniques
and are less expensive. For the feature extraction SFTA, Initially the accuracy was 72.73%, while after feature
that is, Segmentation based Fractal Texture Analysis was selection method the increase accuracy was 74.55%
used. Then for the classification, Naive Bayes and SVM
were used. Out of the two classifiers, SVM performed 4. RESULT AND DISCUSSIȮN
better with an accuracy of 92.5% Numerous methods and algorithms of diagnosing liver
cancer are evaluated and compared to a variety of observed
Li et al. [21] proposed an automated process using CNN to data. The TABLE 1 shows how different studies utilised
section lesions from CT images. Then it was compared different feature extraction approaches and classifiers.
with AdaBoost, Random Forest, and support vector From the analysis we can observed that Image Dataset
machine. Handcrafted characteristics incorporating mean, shows better accuracy as compared to the Numerical
variance, and contextual variables were used to train these Algorithms based on CT scan pictures are by far the most
classifiers with the thirty-portal phase enhanced CT images efficient and provide up to 99 percent accuracy.
were examined experimentally using leave just one testing
data. The average Dice Similarity Coefficient (DSC),
precision, and recall achieved 80.06% ± 1.63%, 82.67% ±
1.43%, and 84.34% ± 1.61%, respectively. By analysing
the output, it is observed that CNN has performed better.

Rajesh et al. [22] examined HCC liver cancer using KNN,

Naïve Bayes, DT, RF, and SVM. They proposed a model
to predict the HCC. 80.64% accuracy was achieved by RF
without any cross-validation accuracy whereas all other
classifiers were employed using 10-fold cross validation.
Das et al. [23] The Pictures of a liver were analysed
primarily for the purpose of differentiating cancer lesions
and predicting the various types of liver disease, which
include hemangioma, hepatocellular carcinoma, and
metastatic carcinoma. To segment the liver, we used part

45
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.6, April 2022

Table.1: Comparison and summary report

Sr.No. Paper Dataset Feature Extraction Classifier Accuracy

1 [15] UCI Machine Learning Repository - LR 75%
2 [4] Indian Liver Patient Dataset (ILPD). Pearson Correlation Coefficient ET 92.19%
Gradient boosting decision
3 [16] clinical laboratory data set SVM 90%
trees
CT images of hepatocellular and
4 [17] LBP histogram Fourier DT 95.02%
metastatic cancer
5 [18] MRI Images Histogram SVM 86.67%
error probability plus average
6 [19] Fused 2D dataset of CT and MRI MLP 99%
correlation
Segmentation based Fractal Naïve
7 [20] CT Images 92.5%
Texture Analysis Bayes
UCI machine learning repository
8 [22] - RF 80.64%
named as HCC S
Gray level co-occurrence matrix
9 [23] CT images DNN 99.38%
(GLCM)
10 [24] 3DIRCADb Dataset. LBP and HOG feature KNN 97%
11 [25] Indian Liver Patient Dataset (ILPD) - ANN 92.8%
12 [27] UCI Machine Learning Repository Wrapper Method libSVM 74.55%
pp. 394–424, 2018, doi: 10.3322/caac.21492.

The grey gradient co-occurrence matrix (GLCM) is an [3] “Liver cancer - Symptoms and causes - Mayo Clinic.”
approach for extracting features. has shown more accuracy https://www.mayoclinic.org/diseases-conditions/liver-
as compared to Segmentation based Fractal Texture cancer/symptoms-causes/syc-20353659 (accessed Jul.
Analysis on CT scan images, first has shown accuracy of 06, 2021).
99.38% former has shown 92.5% accuracy. Fused image [4] M. F. Rabbi, S. M. Mahedy Hasan, A. I. Champa, M.
of CT scan and MRI with error probability plus average Asifzaman, and M. K. Hasan, “Prediction of liver
correlation feature extraction technique has shown 99% disorders using machine learning algorithms: A
accuracy result. comparative study,” in 2020 2nd International
Conference on Advanced Information and
5. CONCLUSION Communication Technology, ICAICT 2020, Nov.
The failure to detect cancer at an advanced stage is the 2020, pp. 111–116. doi:
cause of death from liver cancer. As a result, early 10.1109/ICAICT51780.2020.9333528.
prediction is essential, that can be accomplished with the
[5] V. Sapra and M. Lal Saini, “Deep Learning Model for
use of computerization in cancer diagnosis. Early
Detection of Breast Cancer,” SSRN Electronic
identification of cancer is achievable with different
Journal, Mar. 2019, doi: 10.2139/SSRN.3383336.
machine learning techniques, which could save the lives of
many patients. This study summarises previously [6] V. Sapra, M. L. Saini, and L. Verma, “Identification
published works that used a different machine learning of Coronary Artery Disease using Artificial Neural
approach to detect and diagnose liver disease.Different Network and Case-Based Reasoning,” Recent
algorithms perform differently in different situations, but Advances in Computer Science and Communications,
dataset and feature selection are also essential in vol. 14, no. 8, pp. 2651–2661, Jun. 2020, doi:
improving prediction outcomes. Additionally, the paper 10.2174/2666255813999200613225404.
contains an assessment of a range of machine learning
strategies in use by multiple writers, with each [7] S. Uddin, A. Khan, M. E. Hossain, and M. A. Moni,
methodology providing both positive and negative results “Comparing different supervised machine learning
depending upon that datasets and features used, among algorithms for disease prediction,” BMC Medical
other aspects. With this investigation, we discovered that Informatics and Decision Making, vol. 19, no. 1, pp.
adopting a new combination or hybrid machine learning 1–16, 2019, doi: 10.1186/s12911-019-1004-8.
method may improve accuracy and performance, and that [8] P. S. Kohli and S. Arora, “Application of machine
in the future, we can investigate more factors to improve learning in disease prediction,” 2018 4th International
performance over the current technique. Conference on Computing Communication and
Automation, ICCCA 2018, pp. 9–12, 2018, doi:
6. REFERENCES 10.1109/CCAA.2018.8777449.
[1] Tim Newman, “The liver: Structure, function, and
disease,” Medicalnewstoday.Com, 2018. [9] Varun Sapra, Madan Lal Saini, “Deep learning
https://www.medicalnewstoday.com/articles/305075 network for identification of Ischemia using clinical
(accessed Jul. 04, 2021). data”, International Journal of Engineering and
Advanced Technology, ISSN: 2249-8958, Volume-8
[2] F. Bray, J. Ferlay, I. Soerjomataram, R. L. Siegel, L. Issue-5, June 2019.
A. Torre, and A. Jemal, “Global cancer statistics
2018:GLOBOCAN estimates of incidence and [10] Reshma S, “Chronic Kidney Disease Prediction using
mortality worldwide for 36 cancers in 185 countries,” Machine Learning,” International Journal of
CA: ACancer Journal for Clinicians, vol. 68, no. 6, Engineering Research and, vol. V9, no. 07, 2020, doi:

46
 International Journal of Computer Applications (0975 – 8887)
Volume 184 – No.6, April 2022

10.17577/ijertv9is070092. 2017 International Conference on Intelligent

Computing, Instrumentation and Control
[11] S. Sharma, J. Agrawal, and S. Sharma, “Classification Technologies, ICICICT 2017, vol. 2018-Janua, pp.
Through Machine Learning Technique: C4. 5 1066–1070, 2018, doi:
Algorithm based on Various Entropies,” International 10.1109/ICICICT1.2017.8342716.
Journal of Computer Applications, vol. 82, no. 16, pp.
28–32, 2013, doi: 10.5120/14249-2444. [21] W. Li, F. Jia, and Q. Hu, “Automatic Segmentation of
Liver Tumor in CT Images with Deep Convolutional
[12] “Computational Intelligence for Detection of Neural Networks,” Journal of Computer and
Coronary Artery Disease with Optimized Features”. Communications, vol. 03, no. 11, pp. 146–151, 2015,
[13] G. Singh and M. Sachan, “Multi-layer perceptron doi: 10.4236/jcc.2015.311023.
(MLP) neural network technique for offline [22] S. Rajesh, N. A. Choudhury, and S. Moulik,
handwritten Gurmukhi character recognition,” 2014 “Hepatocellular Carcinoma (HCC) Liver Cancer
IEEE International Conference on Computational prediction using Machine Learning Algorithms,”
Intelligence and Computing Research, IEEE ICCIC 2020 IEEE 17th India Council International
2014, pp. 1–5, 2015, doi: Conference, INDICON 2020, no. C, 2020, doi:
10.1109/ICCIC.2014.7238334. 10.1109/INDICON49873.2020.9342443.
[14] A. Mariot, S. Sgoifo, and M. Sauli, “I gozzi [23] A. Das, U. R. Acharya, S. S. Panda, and S. Sabut,
endotoracici: contributo casistico-clinico (20 casi),” “Deep learningbased liver cancer detection using
Friuli Med, vol. 19, no. 6, 1964. watershed transform and Gaussian mixture model
[15] A. K. M. S. Rahman, F. M. Javed Mehedi Shamrat, Z. techniques,” Cognitive Systems Research, vol. 54, pp.
Tasnim, J. Roy, and S. A. Hossain, “A comparative 165–175, 2019, doi: 10.1016/j.cogsys.2018.12.009.
study on liver disease prediction using supervised [24] A. Kalsoom, A. Moin, M. Maqsood, I. Mehmood, and
machine learning algorithms,” International Journal S. Rho, “An Efficient Liver Tumor Detection using
of Scientific and Technology Research, vol. 8, no. 11, Machine Learning,” pp. 706–711, 2021, doi:
pp. 419–422, 2019. 10.1109/csci51800.2020.00130.
[16] G. Cao, M. Li, C. Cao, Z. Wang, M. Fang, and C. [25] J. Jacob, J. Chakkalakal Mathew, J. Mathew, and E.
Gao, “Primary Liver Cancer Early Screening Based Issac, “Diagnosis of Liver Disease Using Machine
on Gradient Boosting Decision Tree and Support Learning Techniques,” International Research
Vector Machine,” ICIIBMS 2019 - 4th International Journal of Engineering and Technology, vol. 5, no. 4,
Conference on Intelligent Informatics and Biomedical pp. 4011–4014, 2018, [Online]. Available:
Sciences, pp. 287–290, 2019, doi: www.irjet.net
10.1109/ICIIBMS46890.2019.8991441.
[26] B. Saritha, N. Manaswini, D. Hiranmayi, S. V.
[17] A. Das, P. Das, S. S. Panda, and S. Sabut, “Detection Ramana, R. Priyanka, and K. Eswaran,
of Liver Cancer Using Modified Fuzzy Clustering and “Classification of Liver Data using a New
Decision Tree Classifier in CT Images 1,” vol. 29, no. Algorithm,” International Journal of Engineering
2, pp. 201–211, 2019, doi: Technology Science and Research, vol. 4, no. 9, pp.
10.1134/S1054661819020056. 330–334, 2017.
[18] L. Meng, C. Wen, and G. Li, “Support vector [27] A. H. Roslina and A. Noraziah, “Prediction of
machine based liver cancer early detection using hepatitis prognosis using support vector machines and
magnetic resonance images,” 2014 13th International wrapper method,” Proceedings - 2010 7th
Conference on Control Automation Robotics and International Conference on Fuzzy Systems and
Vision, ICARCV 2014, vol. 2014, no. December, pp. Knowledge Discovery, FSKD 2010, vol. 5, no. Fskd,
861–864, 2014, doi: pp. 2209–2211, 2010, doi:
10.1109/ICARCV.2014.7064417. 10.1109/FSKD.2010.5569542.
[19] S. Naeem et al., “Machine-learning based hybrid- [28] Varun Sapra, Madan Lal Saini, “Computational
feature analysis for liver cancer classification using Intelligence for Detection of Coronary Artery Disease
fused (MR and CT) images,” Applied Sciences with Optimized Features”, International Journal of
(Switzerland), vol. 10, no. 9, 2020, doi: Innovative Technology and Exploring Engineering,
10.3390/app10093134. ISSN: 2278-3075 Volume 8, Issue-6C, Pages 144-
[20] A. Krishna, D. Edwin, and S. Hariharan, 148, April 2019.
“Classification of liver tumor using SFTA based
Naïve Bayes classifier and support vector machine,”

IJCATM : www.ijcaonline.org 47