IRJET- A System to Detect Heart Failure using Deep Learning Techniques

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 06 | June 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 384
A System to detect Heart Failure using Deep Learning Techniques
Shubhangi Khade1, Anagha Subhedar2, Kunal Choudhary3, Tushar Deshpande4, Unmesh Kulkarni5
1Professor, Dept. of Computer Engineering, Modern Education Society’s College of Engineering, SPPU, Pune,
Maharashtra, India
2,3,4,5Students, Dept. of Computer Engineering, Modern Education Society’s College of Engineering, SPPU, Pune,
Maharashtra, India
----------------------------------------------------------------------***-----------------------------------------------------------------------
Abstract - Cardiovascular diseases or Congestive Heart
Failure is one of the leading causes of deaths all over the
world. It accounts for almost a million people’s deaths on a
yearly basis. Also, 3-5% of hospital admissions are due to
heart failure occurrence. This is an alarming situation when
something needs to be done to impede the progression of the
disease hence boosting the quality of life. Although traditional
machine learning methods have been implemented previously,
this is a diligent effort in the direction of a prior revelation of
the disease which might help in reducing the number of
deaths. This paper proposes the use of algorithms like Boosted
Decision Tree (for detection), CNN (for subtype estimation),
and finally predicting possible unfortunate events. The
primary focus is on accuracy of detection of CHF, prevention
being the major concern.
Key Words: Congestive Heart Failure, Deep-
Reinforcement Learning, Convolution Neural Network,
Electrocardiogram, Boosted Decision Tree
1. INTRODUCTION
According to the recent studies, heart disease appeared to
be one of the leading causes of deaths all over the world.
Heart diseases or cardiovascular diseases refer to the
condition which directly or indirectly affect the functionality
of the heart. Conditions involving blocked blood vessels
which may lead to heart attack and other strokes also come
under heart diseases. This article focuses specifically on
heart failure, which is one of the forms of heart disease.
Heart failure in simple form can be explained as the inability
of the heart muscles to pump blood efficiently. This may lead
to an unhealthy heart and thus an unhealthy life. Detection
of heart failure in a patient is necessary because the
condition of the heart worsens day by day if not given the
right treatment. Once the heart failure is detected in a
patient, it is not possible to cure it. One can only increase the
life expectancy up to some (or many) years depending on
the treatment he/she receives and the stage or severity level
of heart failure. Hence, it is important to detect heart failure
in a patient at an early stage so as to get the right treatment
appropriately.
In the past few years, several traditional machine learning
algorithms were implemented to solve this problem of
detecting heart failure at an early stage. However, the main
drawback of these systems was that the models they built
were static in nature. And hence, it affects the accuracy of
the model if patients with new or outlying traits are given to
the system.
To tackle this problem, the use of CNN, along with the
traditional machine learning algorithms is demonstrated in
this paper. Making use of the deep learning algorithms
increases the accuracy of the system as well. The use of
machine learning algorithms like boosted decision tree is
used at the initial stage to find the probabilities of a patient
being prone to heart failure. The CNN layer comes into
picture to accurately detect the heart failure once the
probability obtained in the decision tree surpasses a
threshold. The next section will help you in better
understanding of the proposed system.
2. LITERATURE SURVEY
Some recent studies related to heart failure have been
mentioned below.
Firstly, Elfadil et al. [2] proposed a technique for detecting
heart failure in patients using spectral analysis and neural
networks. Their approach was to divide the power spectral
densities into six regions (From R1 to R6), and to use a
neural network with 6 input nodes and only a single output
node. The inputs to these 6 input nodes are the densities
which are divided into the six regions. The output node is
just a single node which classifies into two classes, normal
or CHF. The accuracy rate of the network obtained was
83.65%.
Yang et al. [3] proposed a scoring model for diagnosis of
heart failure at an early stage. The model was based on
Support vector machine, and the Bayesian principal
component analysis was used for assigning data at missing
values. The model classified the patients into three groups,
which are, healthy group, heart failure prone group and
heart failure group. The accuracy of the model was found to
be 74.4%.

The authors Shouman et al. [4] proposed a system for
detection of heart diseases using single and hybrid data
mining techniques. The proposed system determines gaps in
previous studies on heart disease diagnosis and cure. The
model aims to methodically fill those gaps to explore further
advancement in diagnosis using data mining techniques.
The authors Miao et al. [5] developed a system using deep
neural networks to enhance efficiency and reliability of
diagnosis of heart diseases. The model utilizes multiple layer
architecture of deep learning. The propose system has a
classification model which uses training data and a model
for prediction which makes prediction with help of a
dataset. The testing results of the system showed
truthfulness of 83.67%, sensitivity 93.51% and specificity of
72.86%.
Chang—Sik Son, et al. [6] worked towards early diagnosis of
Congestive Heart Failure in emergency rooms. They
designed a decision-making model which uses Rough Sets
(RS) and Decision Trees. Among the data, two subsets were
determined: RS-based and LR-based. 10-fold cross
validation method was conducted to compare the decision
making models. The generated model was found to
outperform the other models and was 97.5% accurate.
Heart Failure subtypes detection is of utmost importance
and Alonso-Betanzos, et al. [7] proposed a paradigm that
does so by using Ejection Fraction (EF). Based on the metric
Ejection Fraction, nearly half of the HF patients have
preserved ejection fraction and other half reduced ejection
fraction. These are the major two subtypes patients are
distinguished into.
Two basic categories of CHF are: Systolic CHF and diastolic
CHF. Yalcin Isler [8] makes use of Heart Rate Variability
(HRV) analysis to discriminate patients accordingly. Use of
Nearest Neighbor and Multi-Layer Perceptron (MLP) helps
to achieve an accuracy of 96.43%.
The authors Mahmud et al. [9] have provided a survey on
application of Deep Learning (DL), Reinforcement Learning
(RL) and their combination Deep-Reinforcement Learning
(deep RL) on biological data. Also, a comparison is carried
out on the basis of performance when DL techniques are
applied on different datasets.
The authors Bhurane et al. [10] proposed an automated
approach for the diagnosis of CHF using ECG signals. Short
ECG segments were made use of for the experiments. Using
frequency localized filter banks, five different features were
extracted. They have used Quadratic Support Vector
Machine (QSVM) for training and classification purpose.
Accuracy was found to be 99.66%.
U. R. Acharya et al. [11] presented an 11 layer deep CNN
model for CHF diagnosis. The model requires less
preprocessing of ECG signals and neither engineered
features or classification. The model achieved an accuracy of
98.97% for one of the datasets taken. The model helps
cardiologists by providing fast interpretation of ECG signals.
3. PROPOSED SYSTEM
The proposed system consists of four modules based on the
learning curve given in [1]. We have used various machine
learning techniques as well as deep learning techniques for
this purpose. Our architecture goes as below.
3.1 Detection of Heart Failure
This is our main module wherein we detect whether a
patient is heart failure prone or not. We have firstly used a
two-class boosted decision tree which is developed using a
dataset of 10801 patients which consist of parameters such
as AVGHEARTBEATSPERMIN, PALPITATIONSPERDAY,
CHOLESTEROL, BMI, HEARTFAILURE, AGE, SEX,
FAMILYHISTORY, SMOKERLAST5YRS and
EXERCISEMINPERWEEK. This algorithm gives us a
probability of to which extent the patient is heart prone. If
the probability is higher than or equal to 50%, we pass the
ECG recordings of the respective patients to the CNN layer.
The CNN algorithm is trained using FANATASIA dataset
which is a public dataset available on PhysioNetBank. The
dataset consists of 60000 recordings of various patients. The
dataset was split into 2 parts consisting of training and
testing data. On the trained module we then pass the ECG
recording which is reshaped into 3D for the CNN module.
This layer finally gives us a rough estimation with greater
accuracy if the patient is truly heart failure prone or not.
Fig -1 Confusion matrix for HF Detection

Fig -2 Loss and Accuracy Curves
3.2 Detection of Heart Failure Type
We have used SVM as our algorithm for this module. The
algorithm gives us an accuracy of 84%. The dataset was a
public dataset which has parameters like systolic pulmonary
artery pressure, diastolic pulmonary artery pressure and
heart rate etc. This module gives us a type estimation using
3 classes stable, rare and frequent. The dataset was split into
train test in the ratio 9:1. We used K Cross Validation for
testing which consists of 10 folds. Given below is the
accuracy vs fold curve:
Fig -3 Accuracy vs Fold Curve
3.3 Detection of HF Severity
This module detects the severity of the heart failure and
classifies the patients into classes from num0 to num4 with
0 being no HF, and 4 being the highest severity of HF.
Artificial neural network is used for its functioning. The
module achieved an accuracy of 88.3%.
Fig -4 Confusion matrix of HF Severity
4. FUTURE SCOPE
The proposed system was based solely on Heart Failure/
CHF, which is, one of the many types of heart diseases.
Similarly, various algorithms like CNN can be used to predict
if a person is prone to heart disease or not. And if so, should
be able classify the type of heart disease accurately (up to
some extent).
5. CONCLUSION
In this paper a useful system is proposed and developed
which will be able to help the doctors in evaluating the
medical condition of a patient and more specifically be able
to detect if a patient is prone to heart failure or not. And if
so, be able to accurately predict the type of heart failure and
the severity of it as well. For the purpose of detection of
heart failure, a boosted decision tree and the CNN module is
used which gives an estimation of the patient being prone to
heart failure. The SVM algorithm is used for detection of
heart failure type, and an accuracy of 84% is obtained. And
to measure the severity of heart failure, an artificial neural
network is used, which according to the measures, show
88.30% accuracy.
6. REFERENCES
[1] Evanthia E. Tripoliti, Theofilos G. Papadopoulos,
Georgia S. Karanasiou, Katerina K. Naka, Dimitrios I.
Fotiadis. “Heart Failure: Diagnosis, Severity
Estimation and Prediction of Adverse Events

Through Machine Learning Techniques”,
Computational and Structural Biotechnology Journal,
15 (2017) 26-47.
[2] Nazar Elfadil, Intisar Ibrahim. “Self Organising
Neural Network Approach for Identification of
Patients with Congestive Heart Failure”. In 2011
International Conference on Multimedia Computing
and Systems.
[3] Guiqiu Yang, Yinzi Ren. “A Heart failure diagnosis
model based on support vector machine”. In 2010
3rd International Conference on Biomedical
Engineering and Informatics.
[4] Mai Shouman, Tim Turner, Rob Stocker. “Using Data
Mining Techniques in Heart Disease Diagnosis and
Treatment”. In 2012 Japan-Egypt Conference on
Electronics, Communications and Computers.
[5] Kathleen H. Miao, Julia H. Miao. “Coronary Heart
Disease Diagnosis using Deep Neural Networks”,
(IJACSA) International Journal of Advanced Computer
Science and Applications, Vol. 9, No. 10, 2018.
[6] Chang—Sik Son, Yoon-Nyun Kim, Hyung-Seop Kim,
Hyong-Seob Park, Min-Soo Kim. “Decision making
model for early diagnosis of CHF using rough set
and decision tree approaches”, Journal of Biomedical
Informatics 45 (2012) 999-1008.
[7] Amparo Alonso-Betanzos, Veronica Bolon-Canedo,
Guy R. Heyndrickx and Peter L.M. Kerkhof.
“Exploring Guidelines for Classification of Major
Heart Failure Subtypes by using Machine Learning”,
Clinical Medicine Insights: Cardiology, 9s1,
CMC.S18746.
[8] Yalcin Isler. “Discrimination of Systolic and Diastolic
Dysfunctions using Multi-Layer Perceptron in Heart
Rate Variability Analysis”, Computers in Biology and
Medicine, 76, 113-119.
[9] Mufti Mahmud, Mohammed Shamim Kaiser, Amir
Hussain, Stefano Vassanelli. “Applications of Deep
Learning and Reinforcement Learning to Biological
Data”. In IEEE Transactions on Neural Networks and
Learning Systems.
[10] Ankit A. Bhurane, Manish Sharma, Ru San-Tan, U.
Rajendra Acharya. “An Efficient Detection of
Congestive Heart Failure using frequency localized
filter banks for the diagnosis with ECG Signals”. In
Cognitive Systems Research (2018).
[11] U Rajendra Acharya, Hamido Fujita, Shu Lih Oh,
Yuki Hagiwara, Jen Hong Tan, Muhammad Adam, Ru
San Tan, “Deep Convolutional Neural Network for
the Automated Diagnosis of Congestive Heart
Failure using ECG Signals”. In Applied Intelligence
(2018).

IRJET- A System to Detect Heart Failure using Deep Learning Techniques

More Related Content

What's hot (20)

Similar to IRJET- A System to Detect Heart Failure using Deep Learning Techniques (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- A System to Detect Heart Failure using Deep Learning Techniques