Hindawi

Wireless Communications and Mobile Computing

Volume 2022, Article ID 9060340, 13 pages
https://doi.org/10.1155/2022/9060340

Research Article
An Accurate Heart Disease Prognosis Using Machine Intelligence
and IoMT

Jamshid Pirgazi ,1 Ali Ghanbari Sorkhi ,1 and Majid Iranpour Mobarkeh 2

1
Department of Computer Engineering, University of Science and Technology of Mazandaran, Behshahr, Iran
2
Department of Computer Engineering and IT, Payam Noor University, Tehran, Iran

Correspondence should be addressed to Jamshid Pirgazi; [email protected]

Received 25 February 2022; Revised 10 May 2022; Accepted 18 June 2022; Published 1 July 2022

Academic Editor: Mohammad R Khosravi

Copyright © 2022 Jamshid Pirgazi et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

In recent years, Internet of Medical Things (IoMT) and machine learning (ML) have played a major role in the healthcare industry
and prediction of in time diagnosis of diseases. Heart disease has long been considered one of the most common and lethal causes
of death. Accordingly, in this paper, a multiple-step method using IoMT and ML has been proposed for diagnosis of heart disease
based on image and numerical resources. In the first step, transfer learning based on convolutional neural network (CNN) is used
for feature extraction. In the second step, three methods of distributed stochastic neighbor embedding (t-SNE), F-score, and
correlation-based feature selection (CFS) are utilized to select the best features. In the end, a combination of outputs of three
classifiers including Gaussian Bayes (GB), support vector machine (SVM), and random forest (RF) according to the majority
voting is employed for diagnosis of the conditions of heart disease patients. The results were evaluated on the two UCI
datasets. The results indicate the improvement of performance compared to other methods.

1. Introduction As previously mentioned, IoT and IoMT play a great

part in the healthcare in prediction of time and chronic ill-
According to the World Health Organization (WHO) statis- ness diagnosis. The volume of information required by the
tics, cardiovascular disease is one of the leading causes of death healthcare, security factors, power of processing, and accu-
worldwide, accounting for 17.9 million deaths each year [1]. racy of information is very important in terms of diagnostic
The main causes of heart disease are various unhealthy activ- prediction for many illnesses. To tackle these challenges, AI
ities such as high cholesterol, obesity, an increase in triglycer- algorithms in previous researches are used to increase the
ide levels, and high blood pressure, among others. Sleep precision of patients’ data [5].
problems, irregular heartbeat, swollen legs, and, in some cases, IoMT refers to disease diagnosis without human inter-
weight gain of 1 to 2 kilograms per day all increase the risk of vention through the development of intelligent sensors,
heart disease [2, 3]. All these symptoms are common within smart devices, and advanced lightweight communication
various diseases leading to death in the near future; therefore, protocols. IoMT-based healthcare, swallowable sensor track-
the correct diagnosis is difficult. ing, mobile health, smart hospitals, and improved treatment
Smart healthcare presents healthcare platforms which of chronic diseases have been shown in [6].
make use of tools such as IoT, wearable appliances, and IoMT is a new network-based technique for connecting
wireless Internet connection for signing in health evidences medical devices and their applications to healthcare infor-
and resource connection, organizations, and individuals. mation technology systems. In [7], in addition to providing
IoT, artificial intelligence (AI), big data, cloud networks, treatment to orthopedic patients, the IoMT approach exam-
5G, and advanced biotechnology are some of the smart ines the possibilities of facing with COVID-19 pandemic.
healthcare networks used in disease screening and diagnosis In the recent years, ML is widely utilized in healthcare
and medical research [4]. industry to analyze big data for initial prediction of diseases
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 Wireless Communications and Mobile Computing

leading to the improvement of the quality of healthcare [8, optimal DL for the lung cancer classification, brain imaging,
9]. ML can be used to solve complex health issues and give and Alzheimer’s disease is introduced. The researches show
accurate results. Healthcare industry is one of the largest that medical image classification is based on optimal feature
industries in which ML has shown to be functional. Creating selection using the DL by combining preprocessing, feature
accurate and multidimensional datasets are very important selection, and classification. The primary goal of model
and play a critical role in the functionality of ML algorithms. extraction is to select an effective feature for medical image
IoMT enables medical facilities and healthcare products to classification. The opposition-based crow search (OCS)
share real-time data to create a great volume of data for approach is recommended to enhance the efficiency of the
ML [10]. DL classifier. In addition, multitextured, gray-level features
Lately, large amount of research data and patients’ cases are chosen for analysis. Finally, it is claimed that the optimal
have become accessible. There are many open sources for features made better the result of classification.
gaining access to patients’ records, and research can be done This study presents a method based on data collected by
to be able to use computer technologies for patient identifi- IoT. In this regard, a general method is presented for numer-
cation and accurate disease diagnosis in order to prevent ical and image data. At first, the proposed method examines
the lethality of these illnesses. Today, ML and AI are well the type of data resource. If input data were from image
recognized to play major roles in healthcare industry, and resources, in the first step, features are extracted from this
various models of ML and deep learning (DL) can be type of resource using transfer learning. CNN-based deep
employed to classify and diagnose diseases or to predict network is used for this purpose. Fully connected layer has
results. Complete analysis of genome data can easily be done been utilized for feature extraction, whereas if the input data
using different models of ML [11–13]. were from numerical sources, the first step is ignored. The
Several studies have utilized different models of ML for proposed method’s next steps include feature selection and
classification and diagnosis of heart diseases. CART auto- classification phases, which are independent of the input
matic classifier based on classification and regression of con- resource. In the feature selection step, three methods of dis-
gestive heart failure [14], using deep neural network for best tributed stochastic neighbor embedding (t-SNE), F-score,
feature selection and ECG performance improvement [15], and correlation-based feature selection (CFS) have been
proposing a clinical decision support system for diagnosis used. An individual classifier has been trained for each
of heart failures and its prevention during initial stages of method of feature selection. In this paper, three classifiers
the disease [16], and also rule-based natural language pro- of SVM, GB, and RF have been employed. In the end, voting
cessing (NLP) [17] are among these researches. is used for final label selection. The results demonstrate that
In today’s digital age, healthcare generates a large the proposed method performs well.
amount of patient data. For physicians, manual control of The rest of this paper is organized as follows. Section 2
these data is difficult, whereas IoT can manage the produced discusses previous research in this area. Section 3 examines
data very efficiently. IoT records large amounts of data and the proposed method and its details. Section 4 compares
is capable of diagnosing diseases using machine algorithms the performance of the proposed method to some of the suc-
with the purpose of applying different methods of ML on cessful models in this field, and Section 5 concludes the
the produced data. A ML approach is proposed for initial paper.
heart disease prediction in relation to IoT [10].
Cardiac image processing approaches which are
obtained from DL manage and supervise large medical data 2. Literature
gathered by the IoT. Deep IoMT is a common DL and IoT
platform that is in charge of extracting precise cardiac image With the recent advances in medical data processing and
data of usual instruments and devices. Energy depletion, machine learning, many researchers have been consistently
finite battery life, and high PLR (packet loss ratio) are critical active in this field. One of the most challenging medical data
issues that must be addressed in universal medical care. is data related to heart diseases which have drawn many
Wearable devices must be stable (i.e., have a longer battery researchers’ attention. In [20, 21], multiple machine learning
life), energy efficient, and valid in order to improve an methods were examined for the prediction of heart diseases
affordable and inclusive healthcare environment. In this in which recursive neural network (RNN) and decision tree
regard, a new efficient approach based on the consciously (DT) were reported to have gained the best results.
enhanced efficient-aware approach (EEA) of self-adaptive In [22], deep neural network (DNN) with the name of
power control to decrease energy utilization while increasing Heart Evaluation for Algorithmic Risk-reduction and Opti-
validity and battery life is proposed in [18]. For remote car- mization five (HEARO-5) was proposed. This method which
diac imaging of elderly patients, a new common DL-IoMT is consisted of regularization has shown positive results on
framework (DL-based layered architecture for IoMT) has UCI dataset. In [23], for classifying imbalanced clinical data,
also been proposed. a neural network with a convolution layer was used. This
Medical image classification is critical in the prediction study takes advantage of a two-step approach feature weight
and early detection of critical illnesses. Medical imaging is based on least absolute shrinkage and selection operator
the most essential record of patient’s health which helps to (LASSO) and then identification of critical features based
control and cure illnesses, which is one of the important on majority voting for achieving more accuracy in classified
applications of IoMT. In [19], an improved classification of imbalanced data.
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 3

In [24], to increase the performance of the classifier, Transfer learning is an issue of great significance which
feature selection approaches based on fast correlation- focuses on knowledge retention of problem-solving and its
based feature selection (FCBF) were used to choose effi- usage to solve a different but related problem. Since datasets
cient features. In this method, classification is done using are not sufficiently available, CNN network is not initially
K-nearest neighbor (KNN), SVM, Naive Bayes (NB), RF, trained; thus, pretrained network weights aid to solve more
and multilayer perceptron (MLP) optimized using particle issues concerning feature extraction or configuration. Very
swarm optimization (PSO) with ant colony optimization deep networks are costly to be trained. More complex
(ACO) [25]. NB, SVM, and RF methods were employed models require more time for training using hundreds of
for extraction and classification of the most relevant fea- systems with expensive CPUs.
tures in [26, 27]. Transfer learning maps a model that has already been
A k-means method with particle swamp was proposed in trained in specific areas to a new model in new domains;
[28] for detecting hazard factors in coronary heart disease thus, the time required for training by using this method is
treatment (CAD). The extracted data are classified using reduced [32]. Furthermore, in complex models, transfer
MLP, multinomial logistic regression (MLR), and algorithms learning decrease the need for a large number of training
of phase rule, as well as C4.5. It was claimed that the results samples. Because the number of images available in the field
demonstrated the appropriate accuracy of the proposed of heart disease is limited, this method is used to compute
method on the datasets presented by medical college in the initial weights from the well-known ImageNet dataset.
India. In [29], heart disease prediction has been done using The ResNet, AlexNet, VGG-16, and VGG-19 architectures
methods of data mining, ML, and DL, and neural network trained on ImageNet are evaluated based on a set of valida-
method was claimed to be more functional than other tions. VGG-16 architecture has shown the best performance
methods. In [30], genetic algorithms and neural networks due to experimental results. As shown in Figure 2, this paper
were employed for diagnosis of heart disease. uses CNN-based transfer learning to extract features.
3.2. Feature Selection. As it is shown in Figure 1, in this sec-
3. Proposed Method tion, the feature vector extracted from the previous step is
The general procedure of the proposed method is shown in used as the input for feature selection. In this step, three
Figure 1. As it can be seen, this method is made up of three methods of feature selection including t-SNE, F-score, and
major steps. In the first step, two different approaches with CFS are used which are further elaborated in the following.
respect to the input resource are used. If data are numerical, 3.3. Correlation-Based Feature Selection (CFS). As a filter
only feature vector gets used for the next step; however, if method, CFS classifies and evaluates feature subsets based
data are image, the feature vector must be extracted. For on subsets that are highly correlated with the class but unre-
the purpose of extracting features from images, transfer lated to one another [33]. Irrelevant features should be
learning based on CNN has been used. In this stage, fully ignored if they have a low correlation with the class. Aside
connected layer is utilized after convolution layers for fea- from that, the duplicated features can be identified because
ture extraction. The second step of the proposed method is they are closely related to the remaining ones. The feature
made up of feature selection. This step is independent of can be accepted if it predicts the label that no other features
the input resource. Three methods of t-SNE, F-score, and predict. The evaluation function of CFS’ feature subset is as
CFS have been put to use for feature selection. In the third follows:
step of the proposed method, for each feature vector of the
previous step, three different classifiers of SVM, GB, and
RF are used. In the end, majority voting has been used for kr cf
M s = qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð1Þ
selection of the favorable output. Labels of the three classi- k + k ð k − 1 Þr f f
fiers used in the last step are the input of the current step.
Eventually, the final input label is selected. In the following,
different sections of the proposed method will be described. In this equation, M s shows the heuristic “merit” of a fea-
ture subset S including k features, and also, ƒ!
r cf and r f f rep-
3.1. Feature Extraction Based on Image Resource. The extrac-
resent the mean feature-class correlation (f ∈ S) and the
tion of features is a critical issue in classification [31]. As
average feature intercorrelation, respectively. The calculation
illustrated in Figure 1, one of the main steps of the proposed
from this equation has the usage to predict not only the fea-
method is feature extraction. In the step of feature extrac-
ture subsets but also the redundant ones [34].
tion, if the resource is image, it must turn into a feature vec-
tor. Methods based on DL are among the most successful 3.4. F-Score. F-score by evaluating the difference between
methods for feature extraction; however, unfortunately, the two real numbers sets presents a simple feature selection fil-
numbers of images related to heart diseases are very low; ter method [35] which for feature i is calculated as follows:
therefore, in this step, transfer learning has been utilized
for feature extraction (Figure 2). A pretrained CNN network À k Á2
is used in this step as well. This network is merely used for ∑mk=1 xi − xi
F − scoreðiÞ = À k Á nk  k 2 : ð2Þ
feature extraction that the output of fully connected layer ∑m 1/n − 1 ∑ x − 
x k
k=1 j=1 j,i i
is selected as the feature vector.
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 Wireless Communications and Mobile Computing

EEG sensor
ECG sensor

Mobile gateway IoT gateway

Intrabody
communication Numerical data
Input data
Type of
collected by resource
Coordinator
node
sensor

IMU sensor

Image
Transfer learning
with CNN

Features vector
Features vector with fully connected layer

Feature selection step

All Features

t-SNE F-score CFS

Classification step

SVM RF GB
Positive class

Margin

Negative class
L1
L2 L

Voting

Normal Abnormal

Figure 1: An overview of the proposed method.

 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 5

Model

Labels
Imagenet

Flatten Fully Somax

connected
Feature learning Classfication
Input
Transfer learning
knowledge

Pre-trained Features
CNN

Figure 2: Using transfer learning to extract features form image resources.

Table 1: Hyperparameters of the basic classifiers. This process is undertaken through three main steps. They
are as follows:
Methods Parameters Amounts
C_SVM 1 (1) In the first step, the interpoint similarity in high-
Kernel_SVM Radial basis function (RBF) dimensional space is measured. To better under-
SVM Degree_SVM 3 stand this, suppose a set of scattered data points in
a two-dimensional space. For each data point of xi ,
Gamma_SVM Scale
the Gaussian distribution is spread around that point
Coef0_SVM 0
by the user. Then, the density of all xi points will be
Priors_GB None computed based on that Gaussian distribution.
GB
Var-smoothing_GB 1e-08 Then, renormalization is applied to all data points.
This will result in a set of Pij probabilities for all data
Min_samples_split_RF 2
RF points. These probabilities are proportional to their
Min_samples_leaf_RF 1
similarities. This actually means that if x1 and x2
data points possess a similar value under the Gauss-
ian circle, their proportions and similarities will be
equal consequently; hence, the local similarities will
In the above equation, m refers to the number of classes, hold true in the structure of high-dimensional space
nk shows the samples number of class k, xi presents the
mean of feature i among data, also xki demonstrates the (2) The second step is quite similar to the first; but con-
versely, Student’s T-distribution with one level of
mean of feature i in class k, and xkj,i shows the amount of fea-
freedom is used instead of Gaussian distribution
ture i in the sample j of the class k. If F-score related to a fea- which is also known as the Cauchy distribution. This
ture is high, it shows that the respected feature includes will result in a second set of Qij probabilities in a
proper information which belongs to classification.
low-dimensional space
(3) The last step is associated with the reflection of high-
3.5. Distributed Stochastic Neighbor Embedding (t-SNE). dimensional space probabilities Pij through low-
This method is an unsupervised nonlinear method which
dimensional space probabilities Qij in the best possi-
is used for discovery and reduction of data dimensions. In
other words, it will provide the user with an understanding ble manner. The basic requirement here is the simi-
of the manner of data organization in a high-dimensional larity of the two mappings. The difference between
space. This method has been introduced in 2008 by Laurens two-dimensional space probability distributions is
van der Maatens and Geoffery Hinton [36]. The main differ- computed through the Kullback-Leibler (KL) diver-
ence between this method and principal component analysis gence criteria. This study does not elaborate upon
(PCA) is that PCA is a method of reducing the linear dimen- KL. The only point to be considered is that it is an
sions which attempts to maximize the variance and preserve asymmetrical approach in which the effective com-
the large distance between the pares, while t-SNE preserves parison of Pij and Qij values does not suffice. Eventu-
PCA in preserving the small distance between pares by using ally, the optimal value of the KL cost function is
local similarities. t-SNE algorithm computes a similarity found using gradient descent
measure between the pare of samples in large-dimensional
data and low-dimensional space. Then, it attempts to opti- 3.6. Classification. An ensemble classifier is used on the
mize these two similarity measures using a cost function. reduced feature vector. In these types of classifications,
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 Wireless Communications and Mobile Computing

Table 2: Description of Cleveland dataset [39].

No. Name of attribute Description

1 Age Age in years
2 Sex Male is equal 1 and female is equal 0
3 CP Type of chest pain
4 Trestbps A criterion which shows resting blood pressure
5 Chol A criterion which shows serum cholesterol
6 FBS A variable which is boolean, when fasting blood sugar > 120 mg/dl is true otherwise it is false
7 Restecg A criterion which shows resting electrocardiographic results
8 Thalach A criterion which shows maximum heart rate
9 Exang A binary variable which shows exercise-induced angina
10 Oldpeak A criterion which shows ST depression
11 Slope A criterion which shows the peak exercise ST segment
12 CA A criterion which shows major vessel number
13 Thal A criterion which shows heart rate
14 NUM A criterion which shows heart disease status

assumption in Bayesian classifier is statistical independence

between features and in most cases maximizes the perfor-
mance of the acquisition. In this classifier, model parameters
are estimated with a small set of training data. Random for-
est is a simple machine learning technique that usually pro-
duces outstanding results even when its hyperparameters are
not adjusted. This technique is one of the most extensively
used machine learning algorithms for both regression and
classification because of its simplicity and usability [37, 38].
This method works based on building a large number of
decision trees. In the proposed method, the classifications
are combined by voting according to label repetitions. The
main reason for choosing three different classifiers, SVM,
BG, and RF, as the basic classifier which is the main compo-
nent in constructing ensemble classifiers is “diversity.” All of
these classifiers are trained differently leading to the increase
of the level of classification diversity and ensemble
generalization.

Figure 3: Sample image from dataset.

4. Experimental Results
This section summarizes the results of experiments con-
combination of a number of basic classifiers creates an accu- ducted to evaluate the suggested method’s performance. It
rate and robust classification. One of the most common should be noted that all the presented methods and analysis
ways to combine classifiers is majority voting. As shown in of their results are done on same datasets and similar hard-
Figure 1, since the diversity of the consisting classifiers gives ware. All the implementation is done on a computer with
rise to the power of an ensemble classifier, the SVM, BG, and Core (TM) i7 M620 CPU, 4GB memory card, and T4
RF are suggested as basic classifiers. Therefore, it is expected graphic card with Python as programming language as well
that the sample data to be covered in the maximum range as Keras framework. It also should be mentioned that
and the generalizability of the classification to be increased. Scikit-learn-0.22.0 toolbox has been used for classification
It is better not to use the classification with the similar and all the parameters in this toolbox also have been utilized
results in group classification. In order to reduce the classifi- by default. For instance, SVC employs the “one vs. one”
cation error, it is important to choose the appropriate classi- approach for ensemble classification. Table 1 shows the
fier and combination strategy. main classifier parameters.
Support vectors in the SVM model are the most impor-
tant component of the model, which is obtained through 4.1. Database. The Cleveland dataset from UCI is used to
convex optimization. In this model, the classification margin evaluate the proposed method. This dataset is available at
creates the maximum distance within classes. The main http://archive.ics.uci.edu/ml/datasets.php. Cleveland dataset
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 7

Table 3: Descriptions of echocardiogram dataset [4].

Name of attribute Description

Survival This variable indicates the number of months the patient survives
Still-alive A variable which is binary, still-alive is shown by 1 and dead by 0
Age at heart attack Age of heart attack occurrence (in years)
Pericardial effusion A variable which is binary. Fluid around the heart is shown by 1 and no fluid by 0
Fractional shortening A criterion which measures contractility around the heart
Epss Another criterion which measures contractility (E-point septal separation)
Lvdd A criterion which measures the size of the heart (left ventricular end-diastolic dimension)
Wall motion score A criterion which measures the movement of the left ventricle segments
Wall motion index This criterion depends on number of segments seen that can be used instead of the wall motion score
Mult An ignorable var which is derivative
Name Patient’s name
Group Meaningless
Alive at 1 A variable which is boolean, patient was dead after one year is shown by 0 and patient was alive at one year by 1

(RE), precision (PR), and F1 are used to show the perfor-

Shuffle split
mance of the proposed method [40].
0 Generally, accuracy (ACC) refers to a model’s ability to
CV iteration

1 accurately predict the output label. Equation (3) depicts the

accuracy criterion. It also should be mentioned that variance
Class
and mean in 10 numbers of repetitions are considered to cal-
Group culate accuracy for 10-fold cross validation. This criterion
0 20 40 60 80 100
examines the training level and functionality of the model,
although it has no further information regarding the model
Sample index
accurate functionality.
Testing set
TP + TN
Training set Accuracy = : ð3Þ
total examples
Figure 4: Stages of sample selection in testing and training sets.
In equation (4), precision criterion is shown that is
owns 76 attributes and 303 samples. Nonetheless, only 14 appropriate for amounts with high false positive.
attributes of Cleveland dataset were put to use for training
TP
and testing. These features are further elaborated in PR = : ð4Þ
Table 2. These types of data have been used as numerical TP + FP
resources in the present paper.
In equation (5), recall (sensitivity) criterion is shown that
In the following, echocardiogram images have been
is appropriate for amounts with high false negative.
employed as image resources. Figure 3 shows some examples
of these images. The suitable attributes are described in
TP
Table 3. UCI database was used for echocardiography image RE = : ð5Þ
retrieval using 66 normal images from 30 participants and TP + FN
66 abnormal images from 30 subjects [4]. When the vari-
In equation (6), specificity criterion is shown.
ables of “survival” and “still-alive” are combined together,
it shows whether the patient has stayed alive at least one year TN
after the heart attack or not. Spe = : ð6Þ
In the experiments performed to evaluate the proposed TP + FN
method, 10-fold cross-validation was used. The steps for
F1 criterion is shown in equation (7). This criterion also
building a training and test set are described in Figure 4.
contains accuracy and recall (sensitivity) criteria. F1
Accordingly, in each repetition, 10% of the data were used
approaches 0 and 1, respectively, in its worst and best cases.
as a test set and the rest as a training set. In addition, 10%
of the training image sets have been used to create the vali- 2 ∗ RE ∗ PR
dation set. F1 = : ð7Þ
PR + RE

4.2. Evaluation Criteria. Several quantitative criteria includ- In the aforementioned equations, TP presents the num-
ing specificity (Spe), accuracy (ACC), recall (sensitivity) ber of images which is correctly allocated to C i class by
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 Wireless Communications and Mobile Computing

Heart disease data

60 200 70
80
175
60

Number of patients

Number of patients
50
Number of patients

Number of patients
150
60 50
40 125
40
30 100 40
30
75
20
50 20
20
10 10
25
0 0 0 0
40 60 0.00 0.25 0.50 0.75 1.00 0 1 2 3 4 5 100 125 150 175 200
Age (years) Sex (0 = female, 1 = male) Type of chest pain [cp] Resting blood pressure [restbp]

80
100 250
140 70
Number of patients

Number of patients

80 200 120 60

Number of patients

Number of patients
100 50
60 150
80 40
40 100 60 30
40 20
20 50
20 10
0 0 0 0
200 300 400 500 0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 2.0 100 150 200
Serum cholesterol [Chol] Fasting blood bugar [s] Resting electrocardiography [restecg] Maximum heart rate achieved [thalach]

200 140 175

140
175 120 150
120
Number of patients

Number of patients

Number of patients

Number of patients
150 100 125
100
125
80 80 100
100
60 60 75
75
40 40 50
50
25 20 20 25
0 0 0 0
0.00 0.25 0.50 0.75 1.00 0 2 4 6 0.0 0.5 1.0 1.5 2.0 0 1 2 3
Exercise induced angina [exang] Exercise induced ST depression [oldpeak] Slope of peak exercise ST segment [slope] Major vessels colored by fluoroscopy [ca]

160 160
140 140
Number of patients

Number of patients

120 120
100 100
80 80
60 60
40 40
20 20
0 0
0 1 2 3 0 2 4
al Heart disease [num]

Figure 5: Histogram of the number of patients based on different attributes of Cleveland dataset.

classifier and FN presents the number of images from class patients per attribute. As it is evident, the amount of most
C i which are wrongly allocated to other classes using classi- attributes is imbalanced among patients.
fier. FP presents the number of images belonging to class C i Figure 6 shows the frequency of attributes according to
which are allocated to other classes. TN criterion is the num- the individuals’ condition (healthy or sick). With respect to
ber of images which do not belong to class C i nor allocated the aforementioned figure, it is certain that amounts of some
to this class using classifier. of the attributes have more significant relationships with the
condition of samples and show more separability toward
4.3. Results. In this section, we investigate the proposed individuals’ conditions. This relationship and separability,
method’s performance on two datasets with varying input however, is less noticeable in some of the attributes.
resources. In the first dataset, data are numerical and The system’s performance can be influenced by choosing
extracted from Cleveland dataset. As it was previously men- the right features. Three feature selection approaches are
tioned, these types of data directly go into the step of feature employed in this case: t-SNE, F-score, and CFS.
selection as inputs. In this section, in order to show the influ- As stated in the proposed method, for the three classi-
ence of each attribute, the attributes of this dataset are exam- fiers SVM, RF, and GB, the extracted features based on t-
ined. Figure 5 illustrates the histogram of the number of SNE, F-score, and CFS methods have been used, respectively.
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 9

Heart disease data

0.07
8 2.5
0.06 0.025
7
2.0
0.05 6 0.020
0.04 5 1.5
0.015
4
0.03
1.0
3 0.010
0.02
2
0.5 0.005
0.01 1
0.00 0 0.0 0.000
20 40 60 80 0.00 0.25 0.50 0.75 1.00 0 1 2 3 100 150 200
Age (years) Sex (0 = female, 1 = male) Type of chest pain [CP] Resting blood pressure [restbp]

4.0
0.008 8 3.5
0.007 0.020
3.0
0.006 6
2.5 0.015
0.005
2.0
0.004 4
1.5 0.010
0.003
0.002 2 1.0
0.005
0.001 0.5
0.000 0 0.0 0.000
200 300 400 500 0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 2.0 100 150 200
Serum cholesterol [Chol] Fasting blood sugar [s] Rest ECG alach

1.4 2.00
8 3.0
1.2 1.75
2.5 1.50
6 1.0
2.0 1.25
0.8
1.00
4 1.5
0.6
0.75
0.4 1.0
2 0.50
0.2 0.5 0.25
0 0.0 0.0 0.00
0.00 0.25 0.50 0.75 1.00 0 2 4 6 0.0 0.5 1.0 1.5 2.0 0 1 2 3 4
Exang Oldpeak Slope Ca

2.5

2.0

1.5

1.0

0.5

0.0
0 1 2 3
al

Disease
No disease

Figure 6: Frequency of different attributes in two conditions of the healthy and the sick in Cleveland dataset.

Each classifier’s features are chosen using a validation set. obtained in both types of input sources (image or numerical),
Table 4 displays the outcomes of each approach in the valida- the t-SNE feature has the best performance in the SVM classi-
tion set. It should be noted that the mean accuracy for 10 iter- fication, the F-score feature in the RF classification, and the
ations is reported in this table. According to the results CFS feature in the GB classification, respectively.
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 Wireless Communications and Mobile Computing

Table 4: Results of different classifiers based on different feature selection methods in validation set.

Accuracy
Type of data Method
t-SNE F-score CFS
SVM 90.12 (±0.032) 88.34 (±0.570) 81.22 (±0.0078)
Numerical resources RF 85.12 (±0.0322) 86.43 (±0.120) 83.11 (±0.056)
GB 90.21 (±0.0167) 78.45 (±0.077) 88.25 (±0.110)
92.60(±0.570) 93.12 (±0.061) 95.65 (±0.018) SVM
Image resources 94.16(±0.420) 96.32 (±0.045) 89.32 (±0.130) RF
95.78(±0.220) 86.25 (±0.190) 90.74 (±0.470) GB

Table 5: Results of the proposed method in comparison with other methods based on numerical resources in Cleveland dataset.

Method Accuracy Precision Recall Specificity F-score

Logistic regression [8] 83.3 — 86.3 82.3 —
K-neighbors [8] 84.8 — 85.0 77.7 —
SVM [8] 83.2 — 78.2 78.7 —
Random forest [8] 80.3 — 78.2 78.7 —
Decision tree [8] 82.3 — 78.5 78.9 —
DL [8] 94.2 — 82.3 83.1 —
K-nearest neighbor [5] 75.73 — — — —
Decision trees [5] 72.45 — — — —
Random forest [5] 75.73 — — — —
Multilayer perceptron [5] 67.54 — — — —
Naïve Bayes [5] 76.26 — — — —
Linear support vector machine [5] 77.73 — — — —
Faster R-CNN with SE-ResNeXt-101 [4] 98.00 96.16 98.47 96.02 97.58
Proposed method 98.7 96.61 99.18 96.65 98.48

1
0.9
0.8
0.7
0.6
Accuracy

0.5
0.4
0.3
0.2
0.1
0
1 11 21 31 41
Iteration
VGG-16
VGG-19
AlexNet
ResNet

Figure 7: The initial results based on different architectures in image classification.

 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 11

Table 6: Results of the proposed method in comparison with other methods based on image resources.

Method Accuracy Precision Recall Specificity F-score

VGG-19 [4] 95.23 93.96 94.80 93.19 95.58
ResNeXt-101 [4] 96.15 94.00 95.42 92.98 95.99
Inception-ResNet-v2 [4] 96.48 94.07 96.14 94.11 96.04
SE-ResNet-101 [4] 97.94 95.18 97.31 95.03 98.25
Faster R-CNN with SE-ResNeXt-101 [4] 99.15 98.06 98.95 96.32 99.02
Proposed method 99.84 98.64 99.61 97.19 99.12

Table 7: Results of the proposed method with the different voting method.

Type of data Accuracy Method

97.32 Proposed method-weighted majority voting
Numerical resources
98.7 Proposed method-majority voting
98.00 Proposed method-weighted majority voting
Image resources
99.84 Proposed method-majority voting

The proposed method’s results are shown in Table 5. As used for receiving input data based on numerical and image
is obvious, the proposed method outperformed all of the resources. In this paper, to diagnose the condition of heart
other methods. disease patients, a hybrid method based on feature extrac-
In the following, the performance of the proposed tion from images using transfer learning, feature selection
method based on the image resource is examined. It was using t-SNE, F-score, and CFS, and classification using the
noted in the proposed method section that the choice of combined output of three classifiers including GB, SVM,
convolutional network design affects the method’s perfor- and RF using majority voting is used. It was indicated that
mance; hence, four different architectures were investigated: feature selection or a subset of suitable features is a funda-
AlexNet, ResNet, VGG-16, and VGG-19. Training occurs mental part of these types of systems and highly influences
solely in the fully connected layers, which is identical to an the accuracy of their performance.
MLP network used for classification, and the convolutional
layers needed to extract the feature are not learned due to Data Availability
the usage of transfer learning. The output layer has the same
number of layers as the number of classes and is made up of The datasets used and/or analyzed during the current study
two layers. The accuracy performance of each type of archi- are available from the corresponding author on reasonable
tecture with 50 repetitions to train fully connected layers is request.
shown in Figure 7. This comparison shows that the VGG-
16 architecture performs better, and as a result, this architec- Conflicts of Interest
ture has been used to extract features. The results show that
a fully connected neural network (e.g., MLP) reports accu- The authors declare that they have no conflicts of interest.
racy of 96.4% for image classification, and this approach
can improve performance. References
Table 6 shows the results of the proposed method, and as
[1] World Health Organization, Cardiovascular Diseases, WHO,
it can be seen, the proposed method has proved to have a
Eneva, Switzerland, 2020.
suitable performance on these types of data.
[2] American Heart Association, Classes of Heart Failure, Ameri-
In this section, the voting method is evaluated with two
can Heart Association, Chicago, IL, USA, 2020, https://www
different perspectives. In the proposed method, the same .heart.org/en/health-topics/heart-failure/what-is-heartfailure/
weight for each classifier is considered. Table 7 shows the classes-of-heart-failure.
results obtained from the proposed method based on [3] American Heart Association, Heart Failure, American Heart-
weighted majority voting with different weights for each Association, Chicago, IL, USA, 2020, https://www.heart.org/
classifier. As can be seen in the below table, the proposed en/health-topics/heart-failure.
method has performed better. [4] S. Manimurugan, S. Almutairi, M. M. Aborokbah et al., “Two-
stage classification model for the prediction of heart disease
5. Conclusion using IoMT and artificial intelligence,” Sensors, vol. 22, no. 2,
p. 476, 2022.
Many researchers have been interested in using ML to diag- [5] A. Kishor and W. Jeberson, “Diagnosis of heart disease using
nose heart diseases in recent years. In this paper, IoMT is internet of things and machine learning algorithms,” in
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 Wireless Communications and Mobile Computing

Proceedings of Second International Conference on Computing, [21] A. Gavhane, G. Kokkula, I. Pandya, and K. Devadkar, “Predic-
Communications, and Cyber-Security, pp. 691–702, Springer, tion of heart disease using machine learning,” in 2018 Second
Singapore, 2021. International Conference on Electronics, Communication and
[6] S. Vishnu, S. R. J. Ramson, and R. Jegan, “Internet of medical Aerospace Technology (ICECA), pp. 1275–1278, Coimbatore,
things (IoMT)-an overview,” in 2020 5th international confer- India, 2018.
ence on devices, circuits and systems (ICDCS), Coimbatore, [22] N.-S. Tomov and S. Tomov, “On deep neural networks for
India, 2020. detecting heart disease,” 2018, https://arxiv.org/abs/1808
[7] R. P. Singh, M. Javaid, A. Haleem, R. Vaishya, and S. Ali, .07168.
“Internet of medical things (IoMT) for orthopaedic in [23] A. Dutta, T. Batabyal, M. Basu, and S. T. Acton, “An efficient
COVID-19 pandemic: roles, challenges, and applications,” convolutional neural network for coronary heart disease pre-
Journal of Clinical Orthopaedics and Trauma, vol. 11, no. 4, diction,” Expert Systems with Applications, vol. 159, article
pp. 713–717, 2020. 113408, 2020.
[8] R. Bharti, A. Khamparia, M. Shabaz, G. Dhiman, S. Pande, and [24] Y. Khourdifi and M. Bahaj, “Heart disease prediction and clas-
P. Singh, “Prediction of heart disease using a combination of sification using machine learning algorithms optimized by
machine learning and deep learning,” Computational Intelligence particle swarm optimization and ant colony optimization,”
and Neuroscience, vol. 2021, Article ID 8387680, 11 pages, 2021. International Journal of Intelligent Engineering and Systems,
[9] A. G. Sorkhi, Z. Abbasi, M. I. Mobarakeh, and J. Pirgazi, vol. 12, no. 1, pp. 242–252, 2019.
“Drug–target interaction prediction using unifying of graph [25] J. Patel, A. A. Khaked, J. Patel, and J. Patel, “Heart disease
regularized nuclear norm with bilinear factorization,” BMC prediction using machine learning,” in Proceedings of Sec-
Bioinformatics, vol. 22, no. 1, pp. 1–23, 2021. ond International Conference on Computing, Communica-
[10] Z. Al-Makhadmeh and A. Tolba, “Utilizing IoT wearable med- tions, and Cyber-Security, pp. 653–665, Springer,
ical device for heart disease prediction using higher order Singapore, 2021.
Boltzmann model: a classification approach,” Measurement, [26] A. N. Repaka, S. D. Ravikanti, and R. G. Franklin, “Design and
vol. 147, p. 106815, 2019. implementing heart disease prediction using Naives Bayesian,”
[11] S. Shalev-Shwartz and S. Ben-David, “Understanding machine in 2019 3rd International Conference on Trends in Electronics
learning,” in From 4eory to Algorithms, Cambridge University and Informatics (ICOEI), pp. 292–297, Tirunelveli, India,
Press, Cambridge, UK, 2020. 2019.
[12] T. Hastie, R. Tibshirani, and J. Friedman, “The elements of sta- [27] S. Bashir, Z. S. Khan, F. H. Khan, A. Anjum, and K. Bashir,
tistical learning,” in Data Mining, Inference, and Prediction, “Improving heart disease prediction using feature selection
Springer, Cham, Switzerland, 2020. approaches,” in 2019 16th International Bhurban Conference
[13] S. Marsland, Machine Learning, An Algorithmic Perspective on Applied Sciences and Technology (IBCAST), pp. 619–623,
CRC Press, Boca Raton, FL, USA, 2020. Islamabad, Pakistan, 2019.
[14] P. Melillo, N. De Luca, M. Bracale, and L. Pecchia, “Classifica- [28] L. Verma, S. Srivastava, and P. C. Negi, “A hybrid data mining
tion tree for risk assessment in patients suffering from conges- model to predict coronary artery disease cases using non-
tive heart failure via long-term heart rate variability,” IEEE invasive clinical data,” Journal of Medical Systems, vol. 40,
Journal of Biomedical and Health Informatics, vol. 17, no. 3, no. 7, p. 178, 2016.
pp. 727–733, 2013. [29] H. Sharma and M. A. Rizvi, “Prediction of heart disease using
[15] M. M. A. Rahhal, Y. Bazi, H. Alhichri, N. Alajlan, F. Melgani, machine learning algorithms: a survey,” International Journal
and R. R. Yager, “Deep learning approach for active classifica- on Recent and Innovation Trends in Computing and Commu-
tion of electrocardiogram signals,” Information Sciences, nication, vol. 5, no. 8, pp. 99–104, 2017.
vol. 345, pp. 340–354, 2016. [30] K. Uyar and A. İlhan, “Diagnosis of heart disease using genetic
[16] G. Guidi, M. C. Pettenati, P. Melillo, and E. Iadanza, “A algorithm based trained recurrent fuzzy neural networks,”
machine learning system to improve heart failure patient assis- Procedia Computer Science, vol. 120, pp. 588–593, 2017.
tance,” IEEE Journal of Biomedical and Health Informatics, [31] A. G. Sorkhi, J. Pirgazi, and V. Ghasemi, “A hybrid feature
vol. 18, no. 6, pp. 1750–1756, 2014. extraction scheme for efficient malonylation site prediction,”
[17] R. Zhang, S. Ma, L. Shanahan, J. Munroe, S. Horn, and Scientific Reports, vol. 12, no. 1, pp. 1–16, 2022.
S. Speedie, “Automatic methods to extract New York heart [32] S. M. R. Hashemi, H. Hassanpour, E. Kozegar, and T. Tan,
association classification from clinical notes,” in Proceedings “Cystoscopic image classification by unsupervised feature
of the 2017 IEEE International Conference on Bioinformatics learning and fusion of classifiers,” IEEE Access, vol. 9,
and Biomedicine (BIBM), pp. 1296–1299, Kansas City, MO, pp. 126610–126622, 2021.
USA, November 2017. [33] M. A. Hall, Correlation-Based Feature Selection for Machine
[18] T. Zhang, A. H. Sodhro, Z. Luo et al., “A joint deep learning Learning, Citeseer, 1999.
and internet of medical things driven framework for elderly [34] N. Sánchez-Marono, A. Alonso-Betanzos, and M. Tombilla-
patients,” IEEE Access, vol. 8, pp. 75822–75832, 2020. Sanromán, “Filter methods for feature selection–a compara-
[19] R. J. S. Raj, S. J. Shobana, I. V. Pustokhina, D. A. Pustokhin, tive study,” in International Conference on Intelligent Data
D. Gupta, and K. Shankar, “Optimal feature selection-based med- Engineering and Automated Learning, pp. 178–187, Springer,
ical image classification using deep learning model in internet of Berlin, Heidelberg, 2007.
medical things,” IEEE Access, vol. 8, pp. 58006–58017, 2020. [35] S. Ding, “Feature selection based F-score and ACO algorithm
[20] K. Saxena and R. Sharma, “Efficient heart disease prediction in support vector machine,” in 2009 Second International Sym-
system,” Procedia Computer Science, vol. 85, pp. 962–969, posium on Knowledge Acquisition and Modeling, vol. 1, pp. 19–
2016. 23, Wuhan, China, 2009.
 6302, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/9060340 by Saveetha Engineering College, Wiley Online Library on [08/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Wireless Communications and Mobile Computing 13

[36] P. Jiang, W. Ning, Y. Shi et al., “FSL-Kla: a few-shot learning-

based multi-feature hybrid system for lactylation site predic-
tion,” Computational and Structural Biotechnology Journal,
vol. 19, 2021.
[37] L. Breiman, “Random forests,” Machine Learning, vol. 45,
no. 1, pp. 5–32, 2001.
[38] M. Rahimi and M. A. Riahi, “Reservoir facies classification
based on random forest and geostatistics methods in an off-
shore oilfield,” Journal of Applied Geophysics, vol. 201, article
104640, 2022.
[39] M. A. Khan and F. Algarni, “A healthcare monitoring system
for the diagnosis of heart disease in the IoMT cloud environ-
ment using MSSO-ANFIS,” IEEE Access, vol. 8, pp. 122259–
122269, 2020.
[40] J. Pirgazi, A. R. Khanteymoori, and M. Jalilkhani,
“TIGRNCRN: trustful inference of gene regulatory network
using clustering and refining the network,” Journal of Bioinfor-
matics and Computational Biology, vol. 17, no. 3, article
1950018, 2019.