NeuroQuantology | June 2022 | Volume 20 | Issue 6 | Page 4190-4194 | doi: 10.14704/nq.2022.20.6.

NQ22412
Md. Nazmul Hassan et al, A Detection and Segmentation of Medical Image Using Machine Learning Algorithms

A Detection and Segmentation of Medical Image

Using Machine Learning Algorithms
Md. Nazmul Hassan1*, Tabassum Progga2, Md. Ismail Jabiullah3

Abstract
Background and objective: Fatty liver disease (steatosis) is a common condition caused by an accumulation of fat in
the liver. A healthy liver has a small amount of fat. When fat accounts for 5% to 10% of the weight of our liver, it
becomes a problem. This disease is associated with a significant morbidity and fatality rate. Our proposed research is
to develop a machine learning model that can assist doctors in classifying high-risk patients, making an unique
diagnosis, and preventing and managing FLD. Methods: A total of 105 participants were recruited, with 55 of them
having their FLD tested. Classification models were used to discover the disease. Some of the models offered are linear
regression, logistic regression (LR), random forest (RF), K-Neighbors Classifier, K-Neighbors Regressor, and Decision
Tree Classifier. The confusion matrix was utilized to determine the relationship between the actual and predicted
values in the model. Results: The findings of this study suggest that machine learning algorithms are well- suited to
properly predicting FLD. The Logistic Regression model, according to our research, is the most accurate of the
regression models. Among the classification models, the Random-Forest-Classifier had a greater accuracy of 68
percent. Conclusion: Our technique could help identify FLD patients who could have a significant impact on treatment
patterns. The application of this model for early prediction could save money on treatment and reduce medical costs.
4190
Key Words: Fatty Liver Disease (FLD), Ultrasound images, Machine Learning.
DOI Number: 10.14704/nq.2022.20.6.NQ22412 NeuroQuantology 2022; 20(6):4190-4194

Introduction demanding the development of more efficient

Fatty liver disease is a common condition. It has a methods. As a result, accurate risk identification and
significant death and morbidity rate. FLD has been early detection of FLD could have major implications
established as an economic expense. In the long run, for diagnosis, prevention, and, eventually, proper
FLD produces non-cholestatic cirrhosis and treatment.
hepatocellular carcinoma [1]. Furthermore, the Machine learning (ML) is a different approach that
prevalence of FLD has risen in lockstep with might be able to solve these problems. Machine
diabetes, metabolic syndrome, and obesity [2]. learning (ML) is a discipline of computer science
Higher FLD prevalence has been associated to that uses computer algorithms to detect patterns in
higher economic costs. Biopsy has been used to large volumes of data and predict various outcomes
stratify persons with this disease in recent years, [4]. Machine learning approaches have developed as
and it has been recognized as the gold standard of a promising tool for prediction and decision-making
diagnosis. This surgery, however, is more painful in a variety of areas [5]. Due to the availability of
and costly. It has the potential to cause various clinical data, ML has also played an important role in
adverse effects or sampling problems when utilized. medical decision-making [6, 7]. Machine learning
Ultrasonography's accuracy in illness diagnosis is can enhance prediction and uncover latent features
highly dependent on the operator [3], that aren't visible but can be extracted from other
data.

Corresponding author: Md. Nazmul Hassan

Address: 1,2,3Department of Computer Science and Engineering, Daffodil International University, Daffodil Smart City, Ashulia,
Dhaka, Bangladesh
E-mail: [email protected], [email protected] ,[email protected]

eISSN 1303-5150 www.neuroquantology.com

NeuroQuantology | June 2022 | Volume 20 | Issue 6 | Page 4190-4194 | doi: 10.14704/nq.2022.20.6.NQ22412
Md. Nazmul Hassan et al, A Detection and Segmentation of Medical Image Using Machine Learning Algorithms

Several machine learning (ML) techniques are original US images could be implemented to get
currently being used in disease prediction with more promising results. Furthermore, other
significantly higher accuracy than traditional classification methods like neural network and
methods, including logistic regression (LR), random support vector machine classifiers should be
forest (RF), artificial neural networks (ANNs), K- implemented to enhance the efficiency of the
nearest neighbors (KNNs), support vector machines, process. R. Ribeiro and J. Sanches published an
extreme gradient boosting (XGBoost), and linear automatic classification system for the diagnosis of
discriminant analysis (LDA) [8]. liver steatosis (FLD) from ultrasound pictures in
The diagnosis of fatty liver ultrasonic imaging varies June 2009 [11]. The features were chosen so that
due to the use of different ultrasound equipment, clinicians could diagnose the condition simply by
inadequate picture quality, and physical differences looking at the ultrasound images. The outcome of
among patients. On the other hand, a prediction their planned work is quite promising in terms of
model based on readily available clinical sensitivity and specificity. This project has several
characteristics would aid doctors in properly limits as well. More patient data is needed to
identifying and making actionable preventive, early construct a more accurate and robust model. To
diagnosis, and focused intervention decisions. The generate more promising findings, feature
use of models based on electronic medical data to extraction methods on the original US photos could
predict FLD has not been researched thoroughly on be used. Other classification approaches, such as
a large scale. That is why our goal is to use modern neural networks and support vector machine
machine learning techniques, notably in the classifiers, should also be used to improve the
classification method, to develop a predictive model process' efficiency. Based on the features of liver
for fatty liver disease. ultrasonography imaging and clinical fatty liver
diagnostic criteria A SVM classification model is
Background Study built by Li, G., Luo, Y., Deng, W., Xu, X., Liu, A., and
Ma, H., Xu, C. F., Shen, Z., Yu, C. H., and Li, Y. Song, E., (2008) [12]. They were able to diagnose
M. (2018) [9] conducted a cross sectional fatty and normal livers with 97.1 percent and 84 4191
investigation utilizing machine learning approaches percent accuracy, respectively. Although this model
to predict Non Alcoholic Fatty Liver Disease has decent accuracy, more work is needed to add
(NAFLD). They employed ultrasonographic methods more enhancing features. Future study could focus
to diagnose NAFLD, which has certain drawbacks on how to get consistent diagnosis and quantitative
but is still the most widely used method with good analysis results from ultrasound images acquired on
accuracy. Their findings offer valuable insight into various B-scan equipment. Pei, X., Deng, Q., Liu, Z.,
the use of innovative AI approaches to predict Yan, X., and Sun, W., 2021 [13] found that when the
NAFLD. Their model has some flaws as well, such as variables are few, ML models, particularly XGBoost,
the lack of patient liver tissue biopsy data. Patients' have superior accuracy for FLD prediction. This
liver tissue biopsy data could be added in future method may be more beneficial in real-world clinical
studies to construct a more accurate predictive ML practice. In order to predict FLD in a variety of
model. Andrade, A., Silva, J. S., Santos, J., and Belo- datasets, this model will need to be tested in the
Soares, P. (2012) [10] evaluated the effectiveness of future. The authors of [14] wanted to create a
three classifier algorithms for hepatic steatosis machine learning model that could help doctors
diagnosis (Fatty Liver). This study provides a semi- classify high-risk patients and make a novel
automated categorization technique for evaluating diagnosis, as well as prevent and manage FLD.
steatotic liver tissues based on B-scan ultrasound Chieh-Chen Wu, Wen-Chun Yen, Wen-Ding Hsu, Md.
data. Several features were extracted using Artificial Mohaimenul Islam, Phung Anh(Alex) Nguyen,
Neural Networks (ANN), Support Vector Machines Tahmina Nasrin Poly, Yao-Chin Wang, Hsuan-Chia
(SVM), and k-Nearest Neighbors (k-NN) and then Yang Their approach has the ability to detect FLD
used in three different classifiers (kNN). The early, which would improve therapy precision and
classifiers were trained using the 10-cross effectiveness. They only obtained data from a single
validation approach. A stepwise regression-based medical facility. External validation and multicenter
feature selection method was also applied, which datasets, on the other hand, may improve
enhanced prediction accuracy. To build a more performance and dependability. They were unable
accurate and robust model more patients data is to distinguish between patients with fatty and non-
required. Feature extraction methods upon the fatty liver disease due to a lack of data. They simply

eISSN 1303-5150 www.neuroquantology.com

NeuroQuantology | June 2022 | Volume 20 | Issue 6 | Page 4190-4194 | doi: 10.14704/nq.2022.20.6.NQ22412
Md. Nazmul Hassan et al, A Detection and Segmentation of Medical Image Using Machine Learning Algorithms

utilized a classification strategy, however a deep Model Building

learning approach may have been used to improve Predictive classifier algorithms correctly identified
prediction. FLD patients. Classification techniques such as
random forest (RF), decision tree classifier (DTC),
Methodology linear regression, and logistic regression (LR) were
Population Under Study used to develop prediction models. Because of the
We gathered data from the Family Care Foundation following feature, we choose these four models.
Located in Sabamoon Tower, 152, 2, Green Rd, Leo Breiman and Adele Culter invented the random
Dhaka-1205 as part of a liver protection initiative. forest (RF) ensemble classification technique in
All patients who had undergone initial fatty liver 1999 [15]. A significant number of decision trees
treatment were included in the study in December make up this system. The tree is built separately
2021. Patients were excluded if they were using the general technique of bootstrap
a) under 30 years old, b) had a history of mental aggregation (i.e. bagging) and is a sample of the
illness, or c) had a history of substance abuse. d) training set picked at random. The final conclusion
incomplete examination process e) an ultrasound is determined by a simple majority vote of all trees.
test revealed a suspected case of fatty liver. RF has proven to be a highly accurate approach in a
multitude of applications, including medical
diagnostics.
Machine Learning Computer simulations of biological
The main purpose of this study was to employ neural networks are known as artificial neural
classification machine learning models to select networks (ANN). It's a powerful nonlinear model
prognostic indicators for fatty liver disease that's been demonstrated to provide accurate
prediction. During this phase, we divided our predictions in several CDS [16]. This model is made
machine learning technique into four parts: 1. Data up of artificial brain units called "perceptrons" [17].
preprocessing: Examples of data preprocessing Dendrites carry the signal into the neuron in ANN,
include data cleansing, missing data resolution, data just as they do in a conventional neural cell. Signal 4192
transformation, and data imbalance reduction. 2. transmission is replicated from an input layer to a
Variables selection: a technique for selecting the sequence of hidden layers, and ultimately to an
most relevant collection of variables for use in model output layer. Each layer, on the other hand, has a
formulation (help to reduce overfitting, improve huge number of perceptrons, and the perceptrons
accuracy, and reduce the training time). 3. Model between layers are linked by different weights that
construction: select an appropriate classification for can be adjusted during the training process. During
improved prediction. this process, it learns from a large number of
samples in the training dataset until each input
Data Preprocessing matches the corrected input. Logistics regression
(LR) is a discrete choice model that belongs to the
This study comprised a total of 120 patients, with 62 multivariate analysis group. It is the most widely
of them being diagnosed with fatty liver disease. used approach of empirical analysis in sociology,
Because data preparation is a key stage in machine biostatistics, clinical medicine, quantitative
learning, we eliminated any variables with more psychology, econometrics, and marketing, and it is
than 50% missing values. regularly used to compare with machine learning
We used the information gain ranking technique to research [18]. It provides a lot of benefits, including
determine the weight of each variable in this a lot of power and precision.
approach. It aided in determining the efficacy of
factors included in the training dataset. Only
variables with a score greater than 0 were included Model Assesment
in the final model. The confusion matrix was used to figure out how the
actual and predicted values in the model related.
[19] Table 1 shows the structure of confusion matrix
of the research.

eISSN 1303-5150 www.neuroquantology.com

NeuroQuantology | June 2022 | Volume 20 | Issue 6 | Page 4190-4194 | doi: 10.14704/nq.2022.20.6.NQ22412
Md. Nazmul Hassan et al, A Detection and Segmentation of Medical Image Using Machine Learning Algorithms

Table 1: Confusion Matrix Representation

Positive Negative
Predicted true (+) TP TN
Predicted false (-) FP FN

Accuracy show that machine learning methods are well suited

To measure model accuracy, divide learning models to accurately forecast FLD. The Logistic Regression
in medical practice to provide tailored patient care. is the most accurate of the regression models. Fatty
Using correct machine learning prediction models, Liver Disease is a frequent complication of critical
physicians may be able to extract the minimum data illness that is associated with a higher risk of
necessary to make a therapeutic the total number of mortality and morbidity. In recent years, traditional
positive instances by the total number of cases. The diagnostic and treatment methods have contributed
accuracy parameter provides the percentage of to a better understanding of FLD. It can, however,
correctly detected cases. The model's accuracy is have negative repercussions and waste resources.
defined as Machine learning methods always provide greater
insight when compared to traditional statistical
models. We developed and evaluated a classification
Sensitivity: Sensitivity is defined as the ability to machine learning model to predict FLD. The
determine the degree of a trait in order to Random-Forest- Classifier, among the classification
appropriately classify a person with disorders. The models, has a higher accuracy of 68 percent. This
equation is as follows, study has a number of issues that must be
addressed. To begin with, we simply obtained data
Specificity: Specificity is defined as the degree to from one medical center. External validation and
which an attribute is used to appropriately classify a multicenter datasets, on the other hand, may
person who is free of ailments. increase performance and dependability. Second, 4193
there is a dearth of information. In our study, the
patients' BMI was not mentioned.
To characterize the quality of the predicted class,
quality measures such as sensitivity and specificity
Conclusion
are utilized. Three metrics are used to evaluate the
The findings show that a machine learning
sensitivity, specificity, and accuracy of a medical
classification model, especially the random forest
diagnostic model. Precision, sensitivity, and
model, may consistently identify fatty liver disease
specificity are the three criteria. The process
patients using only a few clinical variables. This
diagram of our proposed system is depicted in Fig. 1.
method may be more beneficial in real-world clinical
practice. Machine learning algorithms might be used
to better identify fatty liver for both prevention and
therapy. Future research is needed to demonstrate
that our model can predict FLD in a range of
datasets.

References
Lazo, M., & Clark, J. M. (2008, November). The epidemiology of
nonalcoholic fatty liver disease: a global perspective. In
Seminars in liver disease (Vol. 28, No. 04, pp. 339-350). ©
Thieme Medical Publishers.
Fig. 1: Process diagram Le, M. H., Devaki, P., Ha, N. B., Jun, D. W., Te, H. S., Cheung, R. C.,
& Nguyen, M. H. (2017). Prevalence of non-alcoholic fatty
Result Discussion liver disease and risk factors for advanced fibrosis and
The amount of data in healthcare is always mortality in the United States. PloS one, 12(3), e0173499.
Anstee, Q. M., Targher, G., & Day, C. P. (2013). Progression of
increasing, and machine learning enables massive
NAFLD to diabetes mellitus, cardiovascular disease or
volumes of data to be analyzed fast [20]. As a cirrhosis. Nature reviews Gastroenterology & hepatology,
consequence, there's an opportunity to use machine 10(6), 330-344.
choice [21]. The outcomes of this study Motwani, M., Dey, D., Berman, D. S., Germano, G., Achenbach, S.,
Al-Mallah, M. H., ... & Slomka, P. J. (2017). Machine learning

eISSN 1303-5150 www.neuroquantology.com

NeuroQuantology | June 2022 | Volume 20 | Issue 6 | Page 4190-4194 | doi: 10.14704/nq.2022.20.6.NQ22412
Md. Nazmul Hassan et al, A Detection and Segmentation of Medical Image Using Machine Learning Algorithms
for prediction of all-cause mortality in patients with Pei, X., Deng, Q., Liu, Z., Yan, X. and Sun, W., 2021. Machine
suspected coronary artery disease: a 5-year multicentre learning algorithms for predicting fatty liver disease.
prospective registry analysis. European heart journal, Annals of Nutrition and Metabolism, 77(1), pp.38-45.
38(7), 500- 507. Chieh-Chen Wu , Wen-Chun Yen , Wen-Ding Hsu , Md.
R. Nicole, “Title of paper with only first word capitalized,” J. Mohaimenul Islam , Phung Anh(Alex) Nguyen , Tahmina
Name Stand. Abbrev., in press. Nasrin Poly , Yao-Chin Wang, Hsuan-Chia Yang , Yu-Chuan
Raghupathi, W., & Raghupathi, V. (2014). Big data analytics in (Jack) Li , Prediction of Fatty Liver Disease using Machine
healthcare: promise and potential. Health information Learning Algorithms, Computer Methods and Programs in
science and systems, 2(1), 1-10. Biomedicine (2018)
M. Young, The Technical Writer’s Handbook. Mill Valley, CA: Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-
University Science, 1989. 32.
Hsieh, C. H., Lu, R. H., Lee, N. H., Chiu, W. T., Hsu, M. H., & Li, Y. C. Papadopoulos, M. C., Abel, P. M., Agranoff, D., Stich, A., Tarelli, E.,
(2011). Novel solutions for an old disease: diagnosis of Bell, B. A., ... & Krishna, S. (2004). A novel and accurate
acute appendicitis with random forest, support vector diagnostic test for human African trypanosomiasis. The
machines, and artificial neural networks. Surgery, Lancet, 363(9418), 1358-1363.
149(1),87– 93. Rosenblatt, F. (1958). The perceptron: a probabilistic model for
Ma, H., Xu, C. F., Shen, Z., Yu, C. H., & Li, Y., M. (2018). Application information storage and organization in the brain.
of machine learning techniques for clinical predictive Psychological review, 65(6), 386.
modeling: a cross-sectional study on nonalcoholic fatty Dreiseitl, S., & Ohno-Machado, L. (2002). Logistic regression and
liver disease in China. BioMed research international, 2018. artificial neural network classification models: a
Andrade, A., Silva, J. S., Santos, J., & Belo- Soares, P. (2012). methodology review. Journal of biomedical informatics,
Classifier approaches for liver steatosis using ultrasound 35(5- 6), 352-359.
images. Procedia Technology, 5, 763-770. Provost, F., & Kohavi, R. (1998). Glossary of terms. Journal of
Ribeiro, R., & Sanches, J. (2009, June). Fatty liver Machine Learning, 30(2-3), 271-274.
characterization and classification by ultrasound. In Condie, T., Mineiro, P., Polyzotis, N., & Weimer, M. (2013, April).
Iberian conference on pattern recognition and image Machine learning on big data. In 2013 IEEE 29th
analysis (pp. 354-361). Springer, Berlin, Heidelberg. International Conference on Data Engineering (ICDE) (pp.
Li, G., Luo, Y., Deng, W., Xu, X., Liu, A. and Song, E., 2008, August. 1242-1244). IEEE.
Computer aided diagnosis of fatty liver ultrasonic images Murdoch, T. B., & Detsky, A. S. (2013). The inevitable application
based on support vector machine. In 2008 30th Annual of big data to health care. Jama, 309(13), 1351-1352.
International Conference of the IEEE Engineering in
Medicine and Biology Society (pp. 4768-4771). IEEE. 4194

eISSN 1303-5150 www.neuroquantology.com