Available via license: CC BY-SA 4.0
Content may be subject to copyright.
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
161
INTERNET OF MEDICAL THINGS ENABLED CLOUD-BASED BREAST CANCER
IDENTIFICATION WITH MACHINE LEARNING
K. Parveen1, S. Y. Siddiqui2, M. Daud 3 and G. Abbas
1,3Department of Computer Science, University of Engineering and Technology (UET), Lahore
2Minhaj University Lahore, Pakistan
3Department of Animal Production, Riphah College of Veterinary Sciences, Lahore.
1Email of Corresponding author: kausarnawaz6@gmail.com
ABSTRACT: Breast cancer occurs when cells in the breast grow out of control. Breast cancer can
spread outside the breast through lymph vessels and blood vessels when it spreads to other parts of the
body, it is said to have metastasized. Most breast cancer cases are reported in women who are 50 years
and/or o40 years older. According to facts and figures shared by WHO (World Health Organization), it
impacts 2.1 million women every year and also causes the greatest number of cancer-related deaths
amongst women. Whilst breast cancer rates are higher among women in more developed regions, rates
are increasing in nearly every region globally. Different machine learning algorithms have been
applied to the dataset like Naïve Bayes (NB), J48 Decision tree, K-Nearest Neighbor (KNN) and ANN
(Gradient Descent) have been applied among them ANN (Gradient Descent) produces the optimal
results among these classification algorithms. The proposed Internet of Medical Things Enabled
Cloud-Based Breast Cancer Identification with Machine Learning system model with 98.07 %
accuracy has been achieved. For the proposed model 97.64 % sensitivity and 98.32 % specificity have
been recorded. From the results produced by the proposed expert system, it's satisfactory to utilize it
for breast cancer diagnosis. The Proposed system model will be helpful for the diagnosis of breast
cancer.
Keywords: Internet of medical things (IoMT), ANN (Gradient Descent), prediction model, Breast cancer.
(Received 02.03.2022 Accepted 21.05.2022)
INTRODUCTION
A significant number of (about 2.1 million)
women develop breast cancer every year, and more than
50% become dead due to frequent causes of cancer-
related deaths among women, worldwide. Every year
incidences of breast cancer are reported to increase in
developing countries where early detection is inadequate
but also in developed countries i.e. about 12% in the USA
(WHO, 2019). Early detection of this fatal problem is
important to prevent breast cancer incidences and
survival. In developing countries, the majority of women
are prone to this problem due to a lack of awareness and
resources in health systems for timely detection and
diagnosis (NBCF, 2019). Henceforth, for positive
outcomes connected to breast cancer diagnosis;
inexpensive, and available screening facilities should be
planned for identifying its early symptoms and signs
(Ahmad, A. 2019).
An innovation is witnessed in cancer screening
by Computer Aided Diagnosis (CAD) of specific
(biomarkers) cancer features. Patricio et al. (2018)
reported that these techniques vary from the conventional
test methods of mammography/clinical breast
investigations as these let more vigorous predictive
approaches and outcomes for treatment preferences. Such
strategies need the identification of the potential
biomarkers achieved from the routine blood investigation
of clinical samples to scrutinize their sensitivity,
accuracy, and specificity. Such information/prediction
model patterns are established to augment huge/complex
data sets using various statistical methodologies. Curently
these screening procedures are gaining more significance
in providing important contributions to fetching early
screening quality procedures (Patrício et al., 2018).
A number of studies have authenticated that the
use of machine-learning methods is considered to
increase the accuracy of foreseeing cancer incidences and
their prognosis. Furthermore, these techniques aid to
improve the basic knowledge of cancer
development/progression at the primary level. Machine
learning has been used in various cancer research for the
last 3 decades, however, its applications have become
more common in distinctive and predicting cancer-
causing factors, which is currently indispensable in
determining the probability of rising breast cancer in
women (Osareh et al., 2010). These causative agents
included molecular biomarkers, metabolic parameters,
cellular parameters, and anthropometric data e.g. body
mass index, age). Combining these factors with clinical
information (such as a patient’s general health), a good
set of screening parameters can be developed through a
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
162
machine learning. A range of causative agents has been
investigated for early detection, dependent on their
sensitivity and specificity with the increasing breast
cancer incidences (Brody et al., 2003).
Through conventional screening-methods i.e.
mammography, these parameters can be possibly be
identified in routine blood samples. This research study is
therefore aimed to combine the predictive and
probabilistic approach of machine learning technique
with potential screening candidates or biomarkers of
breast cancer to gain statistically organized data about
their specificity and sensitivity in cancer screening. In
addition, using machine learning we could develop a
comparative study to analyze the most favorable and least
favorable screening candidates for achieving refined and
more sophisticated outcomes. These outcomes would
provide promising alternative solutions and therefore
have great applications in early breast cancer detection
and prevention(Birdwel et al., 2001).
Breast malignant growth has gotten one of the
most widely recognized sicknesses among ladies that
prompt passing. Breast malignancy can be analyzed by
ordering tumors. There are two distinct sorts of tumors,
example, dangerous and kind tumors. Doctors need a
dependable determination strategy to recognize these
tumors. In any case, for the most part, it is hard to
separate tumors even by specialists. Subsequently, the
computerization of the demonstrative framework is
required for diagnosing tumors. Numerous specialists
have endeavored to apply AI calculations for identifying
the survivability of malignant growths in individuals and
it is additionally been demonstrated by the analysts that
these calculations work better in distinguishing disease
analysis. This paper sums up the use of AI calculations in
recognizing diseases in humans. In this review area, 2
gives the data of the neural system, and its learning rules.
Area 3 determines about writing survey dependent on
Artificial Neural Network (ANN). Section 4 indicates
other related chips away at breast cancer utilizing neural
systems. Area 5 infers with other AI calculations and
their sorts, with related work on those calculations
(Praseetha, et al., 2019; Khan, et al., 2020; Siddiqui, et
al., 2021).
LITERATURE REVIEW
The contribution of the Internet of Medical
Things (IoMT) in the healthcare domain is to enhance
medical and healthcare services for individual health and
healthcare departments. Their study reviewed the
researchers who successfully applied IoMT-based
techniques specifically in different medical fields to
provide accurate, productive, and reliable service in a
digitized format. They proposed that IoMT is a
challenging technique with huge potential that would
dynamically revolutionize our healthcare structure (Joyia
et al., 2017).
The researchers demonstrated the efficacy of
IoMT in monitoring big data in healthcare departments.
In their research, they critically examined the
productivity, time management, and all other features of
IoMT enabled devices and software. They concluded that
IoMT enabled wearables or mobile apps are user-friendly
and support fitness, disease symptom tracking, health
education, and can raise preventive strategies. They
believed it IoMT is a novel personalized preventive
technology that can store, analyze and predict large data
files in medical departments by interlinking and reducing
inefficiencies (Ross, et al., 2016).
In artificial intelligence, machine learning is the
branch used for high precision in medical applications. In
this study, the researchers surveyed different types of
machine learning techniques to highlight the success of
its predictive properties in cancer prediction and
progression by comparing machine learning with other
detection and predictive methods, this research survey
found out that machine learning techniques substantially
improve result accuracy by almost 15 to 20 percent in
cancer studies such as breast cancer. Moreover, these
techniques are gaining marked importance in
understanding how cancer develops, its root causes, and
the development of personalized medicines against its
incidence and recurrence. Due to its widespread
applications, the researchers highly recommend machine
learning methods for studying cancer-related difficulties
(Cruz, et al., 2006).
The importance of computer-aided detection
(CAD) in screening for breast cancer is to improve the
early-stage diagnostic rate. After undergoing an extensive
review study they revealed that CAD improves detection
rates, breast tumor imaging, and time-consuming double
reading observations. However, regarding these benefits,
the reviewers declare that the applications of CAD
methods are still limited in clinical settings due to
inadequate perceptions about their effectiveness in cancer
detection. In the future, it is hoped that by implementing
CAD as a diagnostic strategy incidence of breast cancer
can be progressively (Ahmad, et al., 2013) .
Delicate Computing strategies assume a
significant job in choice in applications with loose and
dubious information. The utilization of delicate figuring
disciplines is quickly developing for the determination
and anticipation in clinical applications. Among the
different delicate processing methods, the machine
learning framework exploits the fuzzy set hypothesis to
give register questionable words. In a fluffy master
framework, information is spoken to as a lot of
unequivocal etymological standards. Analysis of bosom
malignancy experiences vulnerability and imprecision
related to loose information measures and inadequacy of
information on specialists. Notwithstanding, there are a
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
163
few innovation situated examinations revealed for bosom
disease determination, and hardly any investigations have
accounted for the bosom malignant growth visualization.
Soft Computing approach systems fuzzy system
13 (Hussain et al., 2019), 14 (Fatima, et al. 2019; Atta et
al. 2019) neural network (Khan et al. 2019), and swarm
intelligence (Khan et al., 2015), and evolutionary
computing (Khan, et al., 2015) like genetic algorithm
(Ali et al., 2016), (Umair, et al., 2015; Umair et al., 2013;
Kashif et al., 2018; Alqudah et al., 2019) are the strong
candidate in the field of smart heath and smart cities.
Proposed Model: The suggested model consists of 2
major phases i.e. Training phase and the Relation phase.
The Training phase is further divided into three major
layers; (1) Data Acquirement layer (2) Pre-processing
layer (3) Application layer. In the Data Acquisition layer,
we use multiple sensors which are IoMT enabled with the
parameters of our research study. They are connected
with the IoMT-enabled devices to store the data in a
database that is called raw data. It's a wireless linked data
in which the pre-processing is applied using
Normalization, Moving Average, and Mean (See figure
1).
After the pre-processing, the application layer is
activated. This application layer is further divided into
two other layers; the Prediction layer and the
Performance evaluation layer. In the Prediction layer, we
use Machine learning which is going to predict our model
or train our model. After training, we evaluate the
performance of our trained model in the performance
evaluation layer for RMSE, Accuracy, and Regression.
Here we have to check if the learning criterion is met or
not. If the criterion does not meet then the prediction
layer is retained again and again until our learning criteria
are met. Once the learning criterion is accomplished the
trained model is stored on a cloud. The data then enters
the Validation phase that can be conducted anywhere or
in any place such as a hospital. The validation is on a
client-server interconnected with IoMT devices. It
imports the trained model from the cloud and gives input
for prediction, the input is then further checked for
accuracy of prediction. If the diagnosis is positive then it
is recommended to a doctor and if it is negative then it is
discarded and the system moves back to the next record.
Figure 1 Theoretical Framework of the study
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
164
Mathematical Model: The ANN architecture with the
Back-Propagation algorithm incorporates the input layer,
hidden layer, and output layer are being used for
convergence and bit per data (BPD) rate. The steps
incorporate initialization of weight, Back Propagation of
error and updating of weight and bias, and Feedforward.
The input is fetched using the equation below
Whereas the equation below represents the output
activation function
Here denotes the desired output. Whereas, as a
calculated output. However, the change in weight for the
output layer is calculated using the formula below
The equation below denotes the chain rule.
After substituting the values in the equation above, the
value of weight changed is obtained using the equation
below.
Where,
Where,
The output and the hidden layer and updation of the
weight and bias between layers is visualized through the
equation below
Dataset Evaluation: The Breast Cancer Dataset (BCD)
utilized by us has been given to the University of
California, Irvine (UCI). 11 traits exist there, what's
more, the first is the ID that we will evacuate (it's
anything but an element we need to take care of in our
grouping). The nine measures are as talked about before
in bosom malignant growth order segment, they are
intended to decide whether lump acts amiable otherwise
defame, the preceding element comprises of parallel
worth (2 aimed at the amiable lump in addition to 4
aimed at defame lump).
The dataset utilized to conduct this research is
the updated version of the Wisconsin Diagnostic Breast
Cancer Data Set (WDBC) downloaded from Kaggle. It
predicts that either it is the benign or malignant state of
breast cancer. The features of the datasets were
determined through an analysis of an image of an FNA
i.e. fine needle aspirate of a breast mass. The analysis
performed incorporates the following traits. However, the
real-valued features from – numbered below, describe the
characteristics of the cell nuclei.
Table1 Dataset Attributes.
Sr.No.
Name of Attributes
Symbol
Data type
1
Id
ID
Integer
2
diagnosis (M=1, B=0)
diagnosis
Integer
3
radius_mean
Radius
Numeric
4
texture_mean
Texture
Numeric
5
compactness_mean
compactness
Numeric
6
area_mean
Area
Numeric
7
smoothness_mean
smoothness
Numeric
8
symmetry_mean
symmetry
Numeric
9
concavity_mean
concavity
Numeric
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
165
10
concave points_mean
concave points
Numeric
11
perimeter_mean
perimeter
Numeric
12
fractal_dimention_mean
fractal_dimention
Numeric
The data set values are numeric so to apply
different classification techniques it is first converted to a
stringto a wordvector then numeric tonominal. Data
Values is converted from numeric to nominal to
implement other techniques which are shown in figure 2
that are classified into healthy and Diseased form.
Figure 2 Healthy VS Diseased
RESULTS AND DISCUSSIONS
A confusion matrix is a technique for
summarizing the performance of a classification
algorithm. Calculating a confusion matrix can give you a
better idea of what your classification model is getting
right and what types of errors it is making.
Table 3ANN (Gradient Descent) Confusion Matrix.
a – Malignant
b – Benign
a – Malignant
207 (TP)
5 (FN)
b – Benign
6 (FP)
351 (TN)
The ANN Gradient Descent is employed to
classify that either the patient had a benign or malignant
tumor. Performance of ANN (Gradient Descent) can be
visualized by Table 4 and Table 5.
Figure 3 shows the performance analyses of all
four machine learning algorithms i.e (Naïve Bayes, J48
Decision tree, KNN, and ANN (Gradient Decent).
It is shown from the experiment performed that Naïve
Bayes accuracy is 78.38% that is less than the ANN that
has an accuracy of 98.07% and also less than the KNN
having an accuracy of 85.59%. It is concluded that
decision trees are used both for numeric and categorical
data which are shown in figure 4. It is shown that the
least significant attribute is class tests in decision trees
that have an accuracy of 81.72% while predicting breast
cancer causes so it can be neglected.
Healthy Diseased
Dataset 357 212
357
212
0
50
100
150
200
250
300
350
400
Healthy vs Diseased
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
166
Table 4:Performance results of ANN (Gradient Decent).
Algorithm
Sensitivity/Recall
%
Specificity
%
Accuracy
%
ANN (Gradient Decent)
97.64
98.32
98.07
Table 5:Performance Analysis of Machine Learning Algorithms.
Algorithm
Sensitivity/Recall
Specificity
Accuracy
Naïve Bayes Classifier
69.81
83.47
78.38
J48 Decision Tree
74.53
85.99
81.72
KNN
82.50
87.26
85.59
ANN Gradient Descent
97.64
98.32
98.07
Figure 3 Performance Analyses of Machine Learning Algorithms
Figure 4 Performance Analysis Results.
Naïve Bayes J48 Decision
tree KNN
ANN
Gradient
Decent
%78.38 81.72 85.59 98.07
0
10
20
30
40
50
60
70
80
90
100
Performance Analysis of Machine Learning Algorithms
Naïve Bayes
Classifier
J48 Decision
Tree KNN ANN Gradient
Decent
Sensitivity 69.81 74.53 82.5 97.64
Specificity 83.47 85.99 87.26 98.32
Accuracy 78.38 81.72 85.59 98.07
0
10
20
30
40
50
60
70
80
90
100
Performance Analysis Results
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
167
Table 6:Comparative Analysis of Results with Different Classifiers
Classifier
Accuracy
Supervised Fuzzy Clustering (Abonyi & Szeifert, 2003).
95.5%
CBRGenetics (Darzi, AsgharLiaei & Hosseini, 2011).
97.3%
Fuzzy Rule Classification (Gadaras & Mikhailov, 2009).
96%
RBF-SVM (Hu, Liu & Yu, 2008).
76%
The proposed model (ANN Gradient Decent)
98.07%
Figure 5 Comparative Analyses of Results with Different Classifiers
Secondly, ANN Gradient Decent shows a high
accuracy level of 98.07% with combining three attributes.
The results gained i.e. accuracy rate after the
experimentation were compared with the research
performed by variant researchers can be visualized in
Table 6 and Figure 5.
Conclusion: This research is conducted to diagnose
breast cancer in women. Certain features are captured to
classify a woman in the "benign" or a "malignant"
class.Dataset is collect from the UCI repository which
contains 569 medical instances. Different data mining
techniques and machine learning techniques have been
implemented on the dataset for breast cancer predictions.
Results produced from these machine learning techniques
show that classification is one of the most capable ways
to predict the presence of disease. Naïve Bayes, Decision
Tree, K-NN, and ANN Gradient Decent produced
accuracies 78%, 60%, 85 %, and 98.07 % respectively.
Comparative analysis of the proposed model is
carried out from where it's found that the proposed model
produced better results than that of the literature models,
which made this research worthful. As patients of breast
cancer are increasing day by day. The development of the
proposed system can help medical practitioners to
diagnose the patient who is suffering from such a deadly
disease, with easier procedures and better reliability. The
sensitivity and specificity produced by this system are
more than that of 98% which makes this system useable
in real-time.
REFERENCES
Ahmad, A. (2019). Breast cancer statistics: recent trends.
Breast Cancer Metastasis and Drug Resistance,
1-7
Ahmad, L. G., Eshlaghy, A. T., Poorebrahimi, A.,
Ebrahimi, M., & Razavi, A. R. (2013). Using
three machine learning techniques for predicting
breast cancer recurrence. J Health Med Inform,
4(124), 3.
Ali, M. N., Khan, M. A., Adeel, M., & Amir, M. (2016).
Genetic Algorithm based adaptive Receiver for
MC-CDMA system with variation in Mutation
Operator. International Journal of Computer
Science and Information Security (IJCSIS),
14(9).
Alqudah, A. M., Algharib, H. M., Algharib, A. M., &
Algharib, H. M. (2019). Computer aided
diagnosis system for automatic two stages
classification of breast mass in digital
mammogram images. Biomedical Engineering:
Supervis
ed Fuzzy
Clusterin
g
CBR
Genetics
Fuzzy
Rule
Classifica
tion
RBF-SVM Propose
d Model
Accuracy (%) 95.5 97.3 96 76 98.07
0
20
40
60
80
100
Comparative Analysis with different classifiers
Pakistan Journal of Science (Vol. 74 No. 3 September, 2022)
168
Applications, Basis and Communications,
31(01), 195000
Atta, A., Abbas, S., Khan, M. A., Ahmed, G., & Farooq,
U. (2020). An adaptive approach: Smart traffic
congestion control system. Journal of King Saud
University-Computer and Information Sciences,
32(9), 1012-1019.
Birdwell, R. L., Ikeda, D. M., O’Shaughnessy, K. F., &
Sickles, E. A. (2001). Mammographic
characteristics of 115 missed cancers later
detected with screening mammography and the
potential utility of computer-aided detection.
Radiology, 219(1), 192-20
Brody, J. G., & Rudel, R. A. (2003). Environmental
pollutants and breast cancer. Environmental
health perspectives, 111(8), 1007-1019
Cruz, J. A., & Wishart, D. S. (2006). Applications of
machine learning in cancer prediction and
prognosis. Cancer informatics, 2,
117693510600200030
Fatima, A., Adnan Khan, M., Abbas, S., Waqas, M.,
Anum, L., & Asif, M. (2019). Evaluation of
planet factors of smart city through multi-layer
fuzzy logic (MFL). The ISC International
Journal of Information Security, 11(3), 51-5.
Hussain, S., Abbas, S., Sohail, T., Adnan Khan, M., &
Athar, A. (2019). Estimating virtual trust of
cognitive agents using multi layered socio-fuzzy
inference system. Journal of Intelligent & Fuzzy
Systems, 37(2), 2769-2784.
Joyia, G. J., Liaqat, R. M., Farooq, A., & Rehman, S.
(2017). Internet of medical things (IoMT):
Applications, benefits and future challenges in
healthcare domain. J. Commun., 12(4), 240-247.
Khan, F., Khan, M. A., Abbas, S., Athar, A., Siddiqui, S.
Y., Khan, A. H., ... & Hussain, M. (2020).
Cloud-based breast cancer prediction
empowered with soft computing approaches.
Journal of Healthcare Engineering, 2020
Khan, M. A., Abbas, S., Hasan, Z., & Fatima, A. (2018).
Intelligent transportation system (ITS) for smart-
cities using Mamdani fuzzy inference system.
no. January
Khan, M. A., Umair, M., & Choudhry, M. A. S. (2015).
GA based adaptive receiver for MC-CDMA
system. Turkish Journal of Electrical
Engineering & Computer Sciences, 23
Khan, M. A., Umair, M., & Choudry, M. A. S. (2015,
December). Island differential evolution based
adaptive receiver for MC-CDMA system. In
2015 International Conference on Information
and Communication Technologies (ICICT) (pp.
1-6). IEEE.
Khan, M. A., Umair, M., Saleem, M. A., Ali, M. N., &
Abbas, S. (2019). CDE using improved opposite
based swarm optimization for MIMO systems.
Journal of Intelligent & Fuzzy Systems, 37(1),
687-692.
Osareh, A., & Shadgar, B. (2010, April). Machine
learning techniques to diagnose breast cancer. In
2010 5th international symposium on health
informatics and bioinformatics (pp. 114-120).
IEEE.
Patrício, M., Pereira, J., Crisóstomo, J., Matafome, P.,
Gomes, M., Seiça, R., & Caramelo, F. (2018).
Using Resistin, glucose, age and BMI to predict
the presence of breast cancer. BMC cancer,
18(1), 1-8
Praseetha, S., BT, M., & Anusuya, S. (2019). Storage and
Security Issues of Medical Images using Cloud
Platform C. Server meant for Security. Int. J.
Innov. Technol. Explor. Eng, 8(12), 977-980.
Ross, C. L., Teli, T., & Harrison, B. S. (2016).
Electromagnetic Field Devices and Their Effects
on Nociception and Peripheral Inflammatory
Pain Mechanisms. Alternative Therapies in
Health & Medicine, 22(3)
Siddiqui, S. Y., Hussnain, S. A., Siddiqui, A. H.,
Ghufran, R., Khan, M. S., Irshad, M. S., &
Khan, A. H. (2020). Diagnosis of arthritis using
adaptive hierarchical Mamdani fuzzy type-1
expert system. EAI Endorsed Transactions on
Scalable Information Systems, 7(26).
Siddiqui, S. Y., Naseer, I., Khan, M. A., Mushtaq, M. F.,
Naqvi, R. A., Hussain, D., & Haider, A. (2021).
Intelligent breast cancer prediction empowered
with fusion and deep learning.
Umair, M., Khan, M. A., & Choudry, M. A. S. (2013,
January). GA backing to STBC based MC-
CDMA systems. In 2013 4th International
Conference on Intelligent Systems, Modelling
and Simulation (pp. 503-506). IEEE.
Umair, M., Khan, M. A., & Choudry, M. A. S. (2015).
Island genetic algorithm based MUD for MC-
CDMA system. In 2015 International
Conference on Information and Communication
Technologies (ICICT) (pp. 1-6). IEEE..
