Vous êtes sur la page 1sur 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/273204884

Early detection of Heart Diseases using ANN (Time and Cost Efficient)

Conference Paper · December 2014

CITATIONS READS

0 181

3 authors:

Chandraveer Singh Rathore Aynur Unal


Ansal University, India École de Technologie Supérieure
1 PUBLICATION   0 CITATIONS    110 PUBLICATIONS   88 CITATIONS   

SEE PROFILE SEE PROFILE

Jay Rajasekera
Tokyo International University
46 PUBLICATIONS   311 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Mushroom Nano Fluidics View project

A Social Welfare Economics Proposal Through Bargaining Theory: a Win-Win-Win Papakonstantinidis Model Approach Inserting Overall Arbitrator Player to the Local
Development Game View project

All content following this page was uploaded by Chandraveer Singh Rathore on 08 March 2015.

The user has requested enhancement of the downloaded file.


Early detection of Heart Diseases using ANN
(Time and Cost Efficient)

Prof. Dr. Aynur Unal


Chandraveer Singh Rathore* Founder of DIGITAL MONOZUKURI
School of Engineering and Technology http://dm.amidstky.com/ California
Ansal University (CA), USA
Gurgaon, India aynurunal@stanfordalumni.org
chandraveer@icloud.com

Antimdev Mishra
Shivaen Katial Assistant Professor
School of Engineering and Technology School of Engineering and Technology
Ansal University Ansal Unversity Gurgaon, India
Gurgaon, India antimdevmishra@ansaluniversity.edu.in
shivam.katial@gmail.com

Prof. Jay Rajasekera


Professor
International University of Japan
http://www.iuj.ac.jp/faculty/jrr/
jrr@iuj.ac.jp
Saurabh Juneja
School of Engineering and Technology
Ansal University Gurgaon,
India
saurabhjuneja7676@gmail.com

Abstract — Artificial Neural Networks (ANN) have estimated count of 17.3 million people died from heart
the ability to exploit the resistance for imprecision and attacks in 2008 (30% of all global deaths). And, the
ambiguity in real-world- problems, and their robustness number of people who die from heart attacks will increase
and uniformity; ANN and their techniques have become to reach 23.3 million by 2030. In present times, only rich
increasingly important for modelling and development in people have access to early detection of heart diseases.
many areas of science and engineering. This paper
presents an analytical overview of the most popular WHAT IS ANN?
ANNs. After an overview of ANN, the paper discusses Neural networks and artificial intelligence (AI) a r e
global expansion for ANN training. The paper then suitable for vague problems, such as recognition,
discusses the techniques and means for detecting heart prognosis, control and classification. A neural network
diseases at an early stage at almost negligible costs and consists of processing units, called ‘neurons’ or ‘nodes’.
lesser time. It bears a passing similarity to actual biological neurons.
A neuron is interconnected in the network by
Keywords: ANNs, heart diseases, early detection, cheaper costs, unidirectional connections with different strengths and
time efficient, backpropagation, review weights. In earlier times, designs were positioned on
perceptrons, which are capable of altering elements which
computed their entire inputs and applied a linear transfer
I. INTRODUCTION function to generate an output. In present times,
In today’s fast life, unhealthy diet, obesity, high blood networks use either binary or continuous activations, or
pressure and physical inactivity have increased the risk linear and non-linear functions.
of heart attacks. According to a WHO report, an
For a network to be trained using a supervised learning It will be dealt in some depth here as many of the
system a ‘training data set’ of sample inputs and their applications use this concept. Using the back
corresponding desired outputs are required. The network propagation algorithm, it is possible to train networks
weights, which will finally store the learnt patterns, are containing hidden layers and so implement complex
initially set to small random values. During learning, the non-linear mappings between the input and output
example inputs are conferred to the network and the domains, as opposed to the simple linear mappings
resultant and desired outputs are set side by side. possible with perceptron networks. It requires a known and
desired output for each input value for the
In general, an ANN can be viewed as a system that calculation of the loss function gradient. It is usually
generates a desired response to an input stimulus. The considered as an administered learning method although
pattern of connectivity in an ANN (that is the strengths it is also used in some unsupervised networks
of the connections between various processing units) like autoencoders. It also requires that the stimulating
defines the causal relations between the network's function used by the artificial neurons or nodes be
processors, and is therefore corresponding to a program differentiable. The delta rule (a learning rule for updating
in a conventional computer. However, in contrast to a weights) is categorized into multi- layered feed forward
conventional computer, the ANN is not given a step by networks, which is made possible by using the chain rule
step procedure to perform some desired task. Instead, the to iteratively compute gradients with each layer.
network is taught to do the task. In loose words, ANN is As referred from the Neuroph studio, the aim is
a biomimicry of a human brain. prediction and classification of severe and chronic
diseases. This will give an idea of the problem and the
'disease ratio' which will reduce a doctor's effort and
II. LITERATURE SURVEY would take much lesser time and costs.
Biological neurons are architecturally more complex
than the simplistic description. They are somewhat more It is an attempt to simulate within specialized hardware
complex than the existing artificial neurons that are built or sophisticated software, the multiple layers of simple
into today’s artificial neural networks. processing elements called neurons. Each neuron is
Processing units are normally considered as being linked to its neighbours with varying coefficients of
corresponding to neurons, and are assumed to work in connectivity that represent the strengths of these
parallel. The nature of a single processing unit in an ANN connections. Learning can be done by adjusting these
can be represented as follows: strengths to cause the overall network to output

convenient results. Softwares which were taken into


 First, the unit estimates the total signal being consideration were NetBeans, Weka and Neuroph Studio
sent to it by other processors in the network.
Java is a high level programming language. It has a
 Second, the unit implements an activation number of feature that make the language apt for use on
function to the total signal, in order to adopt a World Wide Web. It was also used in the compilation of
particular level of internal activity. the main program.
 Third, the unit sends a signal to other processors NetBeans is an open source integrated development
in the network; this signal is a function of the environment (IDE) for developing, basically with Java. I
unit's internal activity. The signal that one All Java application types are supported by NetBeans.
processor sends to another is transmitted through
a packed connection, which is generally Weka is an open source software which was developed
described as being corresponding to a synapse. by University of Wekato, New Zealand.

Neuroph is lightweight Java neural network framework


WHAT IS BACKPROPAGATION? to develop simple neural network designs. It is available
One example of supervised learning is backpropagation, under Apache 2.0 license and is free to use. Neuroph
a common method of practicing a neural net in which the Studio also supports image recognition, handwritten letter
basic system output is compared to the desired output, recognition and text character recognition.
and the system is fine-tuned until the difference between
the two is diminished.
And the rest of the data for No.7 to No.15 was taken
from Gautam Hospital, Adarsh Nagar, New Delhi. The
hospital requested not to disclose the identity of these
nine people; thus name shown here are fictitious.
IV. ANALYSIS
The ANN model in Figure 1.1 was created in Weka and the
program was compiled in NetBeans. The scale mentioned
here, was consulted from a hospital and a pathological lab.

Figure 1.1. Classifier that uses backpropagation to classify


seven attributes

III. EXPERIMENTATION
The aim is to minimize the effort of the doctors and
detect potential heart diseases efficiently and in a cost
efficient manner. The real time data was used to
monitor the attributes of the patients such as Age, Sex,
Haemoglobin(Hb) level, Total cholesterol, HDL
Cholesterol (Good Cholesterol), LDL Cholesterol (Bad
Cholesterol), Heart Beat Rate. These attributes were
compared with the optimal and ideal values so as to
monitor the gravity of the situation. Usually,
Haemoglobin(Hb) level is optimal between 13-18 mg/dl;
below or greater reading will result in different
symptoms.
As and when the values of the patients are entered in the Table 1.1. Real time data for 15 patients with 7 attributes
program, the program matches the values with the
optimal ranges and displays the result accordingly and
also specifies the Disease Ratio of the person according Scale:
to the variations in the respective attributes and specified
ranges. Table 1.1 is drawn to feature the attributes of the Haemoglobin(Hb) Level
patient. The optimal ranges have been specified. The Optimal Range- 13-18 mg/dl
output window was made using NetBeans IDE 8.0. The < 13 mg/dl - Low Hemoglobin Level
values are entered comparing attributes from Table 1.1. > 18 mg/dl - Above Optimal/ Borderline High Level
After entering the values, the comments block shows the
optimal ranges of the particular attribute and in the end Total Cholesterol
gives the disease ratio of the risk of the disease. Two Optimal Range- < 200 mg/dl
screenshots of the program have been accommodated for 200 - 239 mg/dl - Borderline High Level
convenience as Figure 1.2 and Figure 1.3. >= 240 mg/dl High Level

The data for No.1 to No.6 in Table 1.1 was taken from HDL Cholesterol
Dr Lal Path Labs, Mukherjee Nagar, New Delhi. Optimal Range - 40 - 60 mg/dl
And the patients are
1. Harsh Grover LDL Cholesterol
2. Surendra Mohan Optimal Range - < 100 mg/dl
3. Sajal Katial 100 - 129 - Above Optimal
4. Sudha Rani 130 - 159 Borderline High
5. Namrata Sethi 160 - 189 High
6. Bharat Bhushan
planned and validated before actual testing started.
Accurate details of implemented neural network
topology were acknowledged. In this project, Artificial
Neural Network technique allowed us to successfully
collect information and compete with other complex
machines in order to obtain specific results. The
efficiency of the program was in the range of
90-92%. The goal of this project is reducing the costs and
increasing the efficient at the same time. Sometimes data
security is a major concern in communication systems but
with the help of Artificial Neural Networks we can
overcome this problem very quickly. Few demerits were
highlighted during the whole research. The whole
process is too complex and just cannot be understood by a
layman. In the future, we wish to make this process as easy
as testing sugar for diabetes.
VI. FUTURE SCOPE
Most practical applications of artificial neural networks
are based on a computational model involving the
propagation of continuous variables from one processing
Figure 1.2. JAVA model for Experiment 1 unit to the next. As far as future is concerned, it is believed
there are several areas yet to be explored by Artificial
Neural Networks or some other great models believed to
be developed soon which can give us good and more
effective results and error free reports which user expect
in today’s world. The most common use for neural
networks is to project what is most likely to happen. There
are many areas where prediction can help in setting
priorities. For example, the emergency room at a hospital
can be a hectic place, to know who needs the most critical
help can enable a more successful operation This method
will reduce the doctor’s effort to check the reports and
nature of the disease. It will take lesser time to figure
out the disease as compared to any conventional
report. It is cheaper and time efficient. It is believed
this method can be extended for an application of a
power system, automobile engineering, application
screening for jobs, control of a wireless car, monitoring of
stock market on which whole world’s economy depends
upon and much more.

ACKNOWLEDGEMENT
Figure 1.3. JAVA model for Experiment 2
We would specially like to thank Dr. Aynur Unal
(Stanford Alumni) for her immense contribution to this
V. CONCLUSION paper. She has been into the research field since decades
Much attention has been given to the operation of ANN and has published many technical papers. This paper
model to obtain results. Testing of model has ensured was completed under the undivided guidance of
that it was working properly and was systematically her.
REFERENCES Learning: A deterministic optimization approach”,
World Scientific Publishing Company, International
1) Simon S. Haykin, “Neural Networks: A comprehensive Journal of Bifurcation and Chaos, Vol. 16, No.7 (2006)
Foundation, Phi Learning Private Limited, Second Edition, 1929-1950
2008
5) Takayuki Yamada and Tetsuro Yabuto,” Remarks on
2) B. Yegnanarayana, “Artificial Neural Networks”, Neural Network Controller which Uses an Auto- tuning
First Edition, 2009 Method for Nonlinear Functions”, IEEE, 0-7803-
0559-0/92
3) Ming Zhang, “Artificial Higher Order Neural Networks
for Modeling and Simulation”, Information Science 6) Melek Yalcintas and Sedat Akkurt, “Artificial neural
Reference, 2012 networks applications in building energy productions and a
case study for tropical climates”, International Journal of
4) George D. Magoulas and Michael N. Vrahatis, Energy Research, Int. J. Energy Res.2005;29:891-901
“Adaptive Algorithms for Neural Network
Supervised

View publication stats

Vous aimerez peut-être aussi