A REVIEW ON THE PREDICTION OF CONGENITAL HEART DISEASE USING DEEP LEARNING AND MACHINE LEARNING TECHNIQUES.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 511
A REVIEW ON THE PREDICTION OF CONGENITAL HEART DISEASE
USING DEEP LEARNING AND MACHINE LEARNING TECHNIQUES.
Nimi S Das1, Dr. Deepambika V. A.2
1 PG Student, Dept. of Electronics & Communication Engineering, LBSITW, Kerala, India
2 Assistant Professor, Dept. of Electronics & Communication Engineering, LBSITW, Kerala, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Congenital heart disease(CHD) is a leading cause
of newborn mortality and morbidity around the world. Early
detection and management can dramatically minimize the
risk of negative result. Chest X-Ray(CXR) is a useful
examination for medical practitioners to diagnose CHD. The
CXR is a simple, rapid and inexpensive examination that
provides useful diagnostic information clearly displays heart
shapes and sizes with low doses of radiation. The system is
built on deep learning and machine learning techniques and
proposes an efficient and accurate assistance system for
medical practitioners to diagnose CHD. As a result, the deep
learning assisted Convolutional Neural Network(CNN) has
been devised and applied for decision support systems that
assist doctors in diagnosing CHD successfully. Another aspect
of the problem that has been studied in this study is the
prediction of CHD using machine learning techniques. As a
result of the established prediction models and deep learning
categorization, very precise and reliable CHD diagnosis may
be made, reducing the frequency of misdiagnosis that might
cause patients to panic.
Key Words: Convolutional Neural Network,ChestX-Ray,
Cardiomegaly
1.INTRODUCTION
A congenital heart defect (CHD), often referred to as a
congenital heart anomaly or congenital heart disease, is a
birth disorder in the structure of the heart or major arteries.
The signs and symptoms vary depending on the type of
problem. Symptoms might range from non-existent to
potentially fatal. Rapid breathing, bluishskin(cyanosis),low
weight gain, and tiredness are all possible signs. Certain
illnesses during pregnancy, such as rubella, use of certain
medications or drugs, such as alcohol or cigarettes, tight
parental relationships, and poornutritional statusorobesity
in the mother are all risk factors. A risk factor is having a
parent with a congenital cardiac defect. Adolescentsaged 13
to 17 years old experienced the greatest increase in
prevalence, followed by adults aged 18 to 40 years old.
CHD is divided into two categories. They are acyanotic and
cyanotic, respectively. Congenital heart disease can be
caused by a variety of conditions. There are two types of
pulmonary vascularity: high (pulmonary plethora) and
diminished (pulmonary vascularity). The aetiologies of
cyanotic congenital heartdiseasecanbeseparatedintothose
with enhanced pulmonary vascularity(pulmonaryplethora)
and those with normal vascularity.
A chest x-ray (CXR) can detect a ventricular septal defect or
cardiomegaly. CXR is easy to use, takes less time, is
inexpensive, and emits minimal quantities of radiation. The
information gathered from diseased children's chest x-rays
can be utilized to predict CHD and treat it as soon as
possible. Changes in the heart can be caused by a variety of
heart diseases. Changes in the structure of the heart can be
caused by a variety of disorders.
Fig -1: Manually made masks for localizing cardiomegaly
Deep learning has been widely employed in the prediction
and analysis of congenital cardiac disorders, withnoticeable
improvements. A methodforautomaticimageinterpretation
is deep learning, a branch of machine learning. Deep
learning–based analysis haslatelybeenappliedinnumerous
medical settings using imaging modalities, such asdiagnosis
of heart problems, as deep learning became a rapidly
evolving paradigm for computer vision. Previous research
has shown that a deep learning based approach can be used
to objectively recognise diseases or discoveries in a variety
of imaging modalities, with one of thestudiesdemonstrating
that deep learning based analysis has the potential to
outperform clinicians. Given thecapabilitiesofdeeplearning
as demonstrated in earlier studies, it was expectedthatdeep
learning-based analysis might quantitatively predict CHD
from CXR in patients with congenital heart disease.
Using machine learning techniques for this prediction and
handling of data can become very efficient for medical
people. Diabetes, smoking, or excessive drinking, high
cholesterol, high blood pressure, or obesity are all factors
that can increase the risk of heart disease. Working in the

heart, for example, becomes harder orrequiresextra effortif
a person has high blood pressure. If we try to control the
above-mentioned causes, we may be able to lower our risks
of developing a cardiac disease.Machinelearningalgorithms
have proven to be particularly effective tools for CHD study
prediction. Traditional statistical modelsfarebetterinterms
of prediction. Machinelearningtechniquesprovidereal-time
insights, and when combined with the explosion of
computing power, theyareassistinghealthcarepractitioners
in diagnosing patients more rapidly and precisely. Minimize
medical and diagnostic errors, predict bad reactions, and
reduce healthcare costs for both providers and patients by
developing innovative newpharmaceuticalsandtreatments.
2. LITERATURE REVIEW
B. Mihai-Sorin et al. [1] The utilisation of two different types
of network architectures, namely LeNet and Network in
Network(NiN), is discussed in this study. Multipledatabases
will be used to test the networks in this article. One of them
comprises photographs representing burn wounds from
paediatric cases, another contains a large number of art
images and additional facial databases were used. Simple
structures like LeNet and NiN have proventobetrustworthy
for low-complexity classification, buttheyareinsufficient for
more challenging jobs.
Tataru Christine et al. [2] Using digital image processing
techniques and an expertradiologist,deeplearning wasused
to the unique CXR dataset to construct a simple
preprocessing pipeline. As a result, built a pipeline that can
use CXR photos to apply three neural network architectures
that have proven successful in classification tasks:
GoogLeNet, InceptionNet, and ResNet. In final model's
accuracy are presented. There are several approaches to
improve the above model. By, further preprocessing,suchas
removing lungs from pictures or clipping edges of CXR
images to highlight lung regions. To direct toward clinical
use, weighting examples so that sensitivity rather than
specificity is highlighted. The level of uncertainty in a
physician's ground truth diagnosis is taken into account.
incorporation of a segmentation component so that the
network can learn small, particular properties instead of
only macro features.
Krizhevsky A et al. [3] The 1.2 millionhigh-resolutionphotos
in the ImageNet LSVRC-2010 contest were used to train a
massive, deep convolutional neural network to categorise
them into 1000 separate classes. It did not employ any
unsupervised pre-training in our tests to keepthingssimple,
even though it expect to help, especially if it get enough
computational capacity to greatly increase the size of the
network without a matching increase in the amount of
labelled data. The results have improved as we have grown
our network and trained it for longer periods of time, but
it still have a long way to go before it can match the infero-
temporal pathway of the human visual system. Eventually
want to employ very big and deep convolutional nets on
video sequences, where the temporal structure provides
very useful information that ismissingorlessvisibleinstatic
images.
Rubin Jonathan et al. [4] The results of training and testing a
series of deep convolutional neural networks onthisdataset
to recognise numerous common thoracic disorders are
presented in this research. On such a massive collection of
chest x-ray images, which is over four times the size of the
largest previously available chest x-ray corpus, CNNs are
trained for this task (ChestX-Ray14). The CNN training
approach could be improved by using techniques known to
increase image-based classification performance, such as
data augmentation and pixel normalisation and when
making a categorization judgement, Here only consider
radiograph pixel information, To make an accurate final
assessment for numerous illnesses meticulous study of a
patient's history and current clinical record is essential.
Abdelilah Bouslama et al. [5] In this paper, it provides a full
process for detecting Cardiomegaly disease from X-Ray
pictures automatically. The procedure is broken down into
four steps. Here, Cardiomegaly was detected with an
accuracy of between 93 and 94 percent.
S. Tuli et al. [6] HealthFog, an unique framework for
integrating ensemble deep learning with Edge computing
devices, was proposed and deployed for a real-world
application of autonomous Heart Disease analysis.FogBus,a
fog-enabled cloud framework, is used to deploy and test the
proposed model's performance in terms of energy
consumption, network bandwidth, latency, jitter, accuracy,
and execution time. By introducing a new Fog-based Smart
Healthcare SystemforAutomaticDiagnosisofHeartDiseases
using deep learning and IoT dubbed HealthFog, It is just
focused on the healthcare aspects for heart patients. Prior
studies on such Heart Patient analyses did not use deep
learning and hence had a low prediction accuracy,rendering
them useless in practise. Using unique communication and
model distribution strategies like ensembling, this study
allows large deep learning networks to be incorporated in
Edge computing paradigms, allowing for excellent accuracy
with very low latencies.
R. Poplin et al. [7] Here Deep-learning models were
constructed using retinal fundus pictures from 48,101
patients in the UK Biobank. The UK Biobank clinical
validation dataset had a mean age of 56.98.2 years,whilethe
EyePACS-2K clinical validation dataset had a mean age of
54.910.9 years. The findings show that using deep learning
to retinal fundus images alone may predict various
cardiovascular risk variables, such as age, gender, andblood
pressure. This is corroborated by our preliminary MACE
prediction results, which are comparable to the composite
SCORE risk calculator in terms of accuracy.
K. M. Z. Hasan et al. [8] For the identificationofheartdisease,
a novel classifier SDA -Sparse Discriminant Analysis

approach was presented. If the limits between classes are
nonlinear or if subgroups are available within each class,the
time complexity of this technique will be decreased by
optimal scoring analysis of LDA, and it will be
comprehensive to execute sparse discrimination by the
combination of Gaussians. The proposed strategy increased
prediction accuracy by 96%. The results significantly
outperform those reported in previous research in the
literature.
S. M. Awan et al. [9] This paper describe the various
techniques used in Artificial Neural Networks (ANN). The
accuracy is calculated and visualised, for example, the ANN
accuracy rate is 94.7 percent, but the accuracy rate with
Principle Component Analysis (PCA) is 97.7 percent. The
best prediction rate attained in each of the
techniques/methodologies is summarised by investigating,
analysing, and conducting an ensemble base technique, and
the results/prediction rate for each algorithm are reported
in a tabular form as well as graph representations for clear
understanding. It can be raised even more by tweaking the
settings and making them more suitable for each method
and data type.
R. Jin et al. [10] In this research, With a training set of data, a
deep learning algorithm was usedtocreatea neural network
model to predict the risk of cardiovascular disease, which
was then validated with another setofdata.Whencompared
to genuine diagnostic data, this model had a confidencelevel
of 70%, indicating that big data and deep learning can be
used to create a potent prediction tool.
3. CONCLUSIONS
The correct diagnosis of congenital cardiac disease can save
lives, while erroneous diagnosis can be fatal. Different
machine learning algorithms and deep learning are used in
this research to compare and analyse the outcomes of the
CHD dataset. More techniquestocombineML andDLmodels
with specific multimedia for the benefit of patients and
clinicians could be discovered.
ACKNOWLEDGEMENT
We would like to thank the Director ofLBSITWandPrincipal
of the institution for providing the facilities and support for
our work
REFERENCES
[1] B.Mihai-Sorin,F.Iulian-Ionuț,F.Laura,ConstantinVertan,
“The use of deep learning in image segmentation,
classification and detection”, The image processing and
analysis lab (LAPI), Politehnica University of Bucharest,
Romania, from: https://arxiv.org/pdf/1605.09612.pdf.
[2] Tataru Christine, Yi Darvin, Shenoyas Archana, Ma
Anthony. Deep Learning for abnormality detection in Chest
X-Ray images. From: http://cs231n.stanford.
edu/reports/2017/pdfs/527.pdf; June, 13, 2017.
[3] Krizhevsky A, Sutskever I, Hinton E. ImageNet
classification with deep convolutional neural networks.
from: https://www.nvidia.cn/content/tesla/pdf/ machine-
learning/imagenet-classification-with-deep-convolutional-
nn.pdf.
[4] Rubin Jonathan, Sanghavi Deepan,ZhaoClaire,Lee Kathy,
Qadir Ashequl, XuWilso Minnan. Large scale Automated
reading of frontal and lateral chest X-rays using dual
convolutional neural networks. from:
https://arxiv.org/pdf/1804. 07839.pdf; 24 Apr 2018.
[5] Abdelilah Bouslama, Yassin Laaziz, Abdelhak Tali.
“Diagnosis and precise localization of cardiomegaly disease
using U-NET” from:
https://doi.org/10.1016/j.imu.2020.100306
[6] S. Tuli, N. Basumatary, S. S. Gill, M. Kahani, R. C. Arya, G. S.
Wander, and R. Buyya, ‘‘HealthFog: An ensemble deep
learning based smart healthcare system for automatic
diagnosis of heart diseases in integrated IoT and fog
computing environments,’’ Future Gener. Comput. Syst., vol.
104, pp. 187–200, Mar. 2020.
[7] R. Poplin, A. V. Varadarajan, K. Blumer, Y. Liu, M. V.
McConnell, G. S. Corrado, L. Peng, and D. R. Webster,
‘‘Prediction of cardiovascularrisk factorsfromretinal fundus
photographs via deep learning,’’ Nature Biomed. Eng., vol. 2,
no. 3, pp. 158–164, Mar. 2018.
[8] K. M. Z. Hasan, S. Datta, M. Z. Hasan, and N. Zahan,
‘‘Automated predictionofheartdiseasepatientsusing sparse
discriminant analysis,’’ in Proc. Int. Conf. Electr., Comput.
Commun. Eng. (ECCE), Feb. 2019, pp. 1–6.
[9] S. M. Awan, M. U. Riaz, and A. G. Khan, ‘‘Prediction of
heart disease using artificial neural networks,’’VFASTTrans.
Softw. Eng., vol. 13, no. 3, pp. 102–112, 2018.
[10] R. Jin, ‘‘Predict the risk of cardiovascular diseases in the
future using deep learning,’’ Ph.D. dissertation, Dept. Elect.
Comput. Eng., Univ. Texas San Antonio,SanAntonio,TX,USA,
2018.
BIOGRAPHIES
Nimi S Das, currently pursuing
M.Tech in Signal Processing at LBS
Institute of Technology for
Women, Poojappura. Affiliated to
the APJ Abdul KalamTechnological
University, Kerala

Dr. Deepambika V. A, Assistant
Professor at LBS Institute of
Technology for Women,
Poojappura. Affiliated to the APJ
Abdul Kalam Technological
University, Kerala

A REVIEW ON THE PREDICTION OF CONGENITAL HEART DISEASE USING DEEP LEARNING AND MACHINE LEARNING TECHNIQUES.

More Related Content

Similar to A REVIEW ON THE PREDICTION OF CONGENITAL HEART DISEASE USING DEEP LEARNING AND MACHINE LEARNING TECHNIQUES. (20)

More from IRJET Journal (20)

Recently uploaded (20)

A REVIEW ON THE PREDICTION OF CONGENITAL HEART DISEASE USING DEEP LEARNING AND MACHINE LEARNING TECHNIQUES.