Download Artificial Neural Network for the Diagnosis of Thyroid Disease using

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Optogenetics wikipedia , lookup

Machine learning wikipedia , lookup

Network science wikipedia , lookup

K-nearest neighbors algorithm wikipedia , lookup

Central pattern generator wikipedia , lookup

Backpropagation wikipedia , lookup

Pattern recognition wikipedia , lookup

Artificial intelligence wikipedia , lookup

Natural computing wikipedia , lookup

Artificial neural network wikipedia , lookup

Types of artificial neural networks wikipedia , lookup

Transcript
Artificial Neural Network for the Diagnosis of
Thyroid Disease using Backpropagation
Algorithm
Jasveer Singh
Dept. Of CS/IT, Graphic Era University, Dehradun, Uttrakhand, India
[email protected]
Abstract
In this paper an artificial neural network approach is
developed using a back propagation algorithm in order
to diagnose thyroid problems. This research is based on
clinical observations,medical diagnosis and the expert’s
knowldge.We have selected the 10 major risk factors of
the thyroid disease.The data is obtained from different
hospitals in order to diagnosed disease. The data is
separated into inputs and targets. The targets for the
neural network will be identified with 1's as infected
and will be identified with 0's as non-infected.The
networks are simulated using MATLAB. The back
propagation learning algorithm is used to train the
neural
network to met the performance. The
conclusion is that back propagation neural network can
be used as an effective tool for accurately diagnosing the
thyroid problem.
Key words: Artificial Neural Networks, Backpropagation
Networks, Thyroid disease, medical diagnosis.
1. INTRODUCTION
Artificial Neural Network is a branch of Artificial
intelligence, has been accepted as a new technology
in computer science. Neural Networks are currently a
'hot' research area in medical field. The major
problem in medical field is to diagnose disease[1].
Neural networks are used to increase the accuracy
and objectivity of medical diagnosis[2]. In this paper
thyroid hormonal problems are diagnosed using an
artificial neural network approach. A group of
causing factors of a thyroid disease or problem is
given as inputs. There are 21 factors of thyroid
disease but for the purpose of research few major
causes are selected as the input data.We have selected
10 major risk factors of thyroid disease. Based on the
given input factors a decision will be arrived. The
decision making can be done using an artificial
neural network approach. The back propagation
learning algorithm is used to train the neural
network[3]. The efficiency of the network is
analyzed. It is concluded to enhance the algorithm
for still better results. Back propagation is one of the
gradient descent learning algorithm applied in feed
forward neural networks. The output may be a binary
result or classification having multiple results. BPN
assigns weights to the input layer neurons, Calculates
the weight of hidden layer neurons and from hidden
layer neurons the output values are calculated[3]. The
errors from the output are to be adjusted and again
the process is repeated for another set of weights.
Therefore BPN can otherwise be called as Error Back
Propagation neural networks(EBP).
2. INTRODUCTION OF THYROID DISEASE
The thyroid gland is one of the most important organs
in the body as thyroid hormones are responsible for
controlling metabolism. Thyroid is a butterfly shaped
gland found just below the Adam’s apple of our neck.
It's brownish red, with left and right halves (called
lobes) that look like a butterfly's wings. It's light like
a butterfly, too, and usually weighs less than an
ounce. The thyroid gland works like a tiny factory
that uses iodine (mostly from the diet in foods such
as seafood and salt) to produce thyroid hormones.
These hormones help to regulate the body's
metabolism and effects processes, such as growth and
other important functions of the body. The thyroid
gland produces two active thyroid hormones,
levothyroxine (abbreviated T4) and triiodothyronine
(abbreviated T3). These hormones are important in
the manufacture of proteins, in the regulation of body
temperature, and in overall energy production and
regulation [4].Thyroid disease occurs when the
thyroid gland doesn't supply the proper amount of
hormones needed by the body. If the thyroid is
overactive, it releases too much thyroid hormone into
the bloodstream, resulting in hyperthyroidism.
("Hyper" is from the Greek, meaning "over" or
"above.") Hyperthyroidism causes the body to use up
energy more quickly than it should, and chemical
activity (like metabolism) in the cells speeds
up[5].An underactive thyroid produces too little
thyroid hormone, resulting in hypothyroidism.
("Hypo" means "under" or "below.") When the
amount of hormone released into the bloodstream is
below normal, the body uses up energy more slowly,
and chemical activity (metabolism) in the cells slows
down.Although they are two different conditions, in
both hypothyroidism and hyperthyroidism the thyroid
can become larger than normal[6].
Collecting more and more information regarding
Thyroid disease from the doctors, it has been found
that the key risk factors for thyroid disease include
many factors like Fatigue and lack of energy,age,
Poor memory, forgetfulness , Baldness, hair thinning
, personal and so on .But for the purpose of research
10 major causes are selected as the input data.The
following Table1 causes are selected as input
variables.
where n is the number of things to choose from, and
you choose r of them
Figure1: combinations of input variables
Table 1 : Major causes of thyroid problem
3. DATA SET PREPARATION
The data used in this investigation of a thyroid
disease database is colected from doctors.The main
idea of this data set is to construct the neural network
model, which will perform the presumptive diagnosis
of thyroid disease. Initial queries by doctor as
symptoms and some required test details of patients
have been considered as main attributes.The above
Table 1 presents the patient symptom data which are
considered as diagnosis variables.To prepare the
results there was need of data sets.Considering above
ten factors as input variables, possible combinations
are drawn using permutation mathematical formula.
An ordered arrangement of r distinct objects is called
a permutation.
The formula is written:
In this research we have total 1024 combinations of
input variables see figure 1. In which 0’s represents
the low and 1’s represents high .The data is separated
into inputs and targets. The targets for the neural
network will be identified with 1’s as infected and
will be identified with 0’s as non-infected.To train
the inputs sets simulation is done in matlab.
4. MODEL BUILDING
Artificial Neural Network model could perform
"intelligent" tasks similar to those performed by the
human brain. Artificial neural network models offer a
completely different approach to problem solving and
they are sometimes called the sixth generation of
computing. Artificial neural networks are developed
based on brain structure. Like the brain, artificial
neural networks can recognize patterns, manage data
and learn [7]. They are made by artificial neurons
(figure 1), which implement the essence of biological
neurons. The backpropagation
artificial neural
network consists of three groups, or layers, of units: a
layer of "input" units is connected to a layer of
"hidden" units, which is connected to a layer of
"output" units. The activity of the input units
represents the raw information that is fed into the
network. The activity of each hidden unit is
determined by the activities of the input units and the
weights on the connections between the input and the
hidden units. The behavior of the output units
depends on the activity of the hidden units and the
weights between the hidden and output units.
4.1 Training
The ANN models is trained with BP algorithm with
momentum, “traingdm”, which is essentially a
gradient steepest descent method. The model starts
with small random real numbers as the starting
weights. At each training cycle, the error is
calculated, and the weights are changed in the
direction that minimizes the error. The error surface
has as many dimensions as the number of weights,
and all the weights obey this basic principle.This
process of changing the weights or updating the
weights is called training. A The most important
training parameters are: learning rate, momentum,
training epochs and initial weights.
Learning rate and momentum control the speed and
efficiency of the training process. Learning rate is the
rate, at which the network adjusts its weights during
training, hence primarily affects the training speed. A
high learning coefficient provides faster convergence
but training process becomes unstable and divergent
oscillations may occur. With a small learning rate,
training time is increased, but the probability of
reaching the global minimum is increased.
Momentum is a training parameter used to reduce
training time of the BP algorithm and to enhance the
stability of the training. A high momentum reduces
the risk of the network being stuck in local minimum,
but it increases the risk of skipping over the solution.
Using a small value for momentum will lead to
prolonged training. The training epochs of the
training cycle is the number of times the training data
has been presented to the network. The BP algorithm
guarantees that total error in the training set will
continue to decrease as the number of training epochs
increases[8][9]. But, on the other side, excessive
training results in phenomenon called over-training
or memorization. The training set is used to teach the
network. Training continues as long as the network
continues improving on the validation set. After
repeating this process for a sufficiently large number
of training cycles the network will usually converge
to some state where the error of the calculations is
small. In this case one says that the network has
learned a certain target function.
Figure2: Neural Network Training
In this study, we use Matlab as a programming
language since Matlab is widely used in all areas of
applied mathematics in education and research.
Matlab stands for MATrix LABoratory and the
software is built up around vectors and matrices
4.2 Experimental Results
Neural network toolbox from Matlab is used to
evaluate the performance of the proposed networks.
A two-layer feed-forward network with 10 inputs and
10 sigmoid hidden neurons and linear output neurons
are created.
Figure 3: Feed-Forward back propagation network
Training set (p,t) where p is the input vector and t is
the target vector. The activation functions for the
hidden layer is tansig and purelin. The training
function is traingdm.The other Parameters are a
learning rate of 0.1 and a goal of 0.2,epoches are
15000. The performance function used is mean sumsquared error (MSE). The mean squared error (MSE)
is the average squared difference between outputs
and targets. Lower values are better while zero means
no error. After making several trains, it was found
that for learning rate of 0.01 at 4390 epochs lead to
fast convergence with minimum error . In this
experiment our performance goal is met and the
diagnostic accuracy is 99% with the training set data
see figures.
Figure 4:
training
Network performance function during
5. CONCLUSION
Neural network has been proven of their capabilities
in many domains such as medical application. ANNs
can be effectively used for diagnosis of Thyroid. The
results of applying the bachpropagation neural
networks in diagnosis of the disease based upon the
selected symptoms showed that artificial neural
networks are able to learn the patterns corresponding
to symptoms of the person. Artificial neural networks
showed significant results in thyroid diseases
diagnosis. This work provides a basis for diagnosis
as well and hence can be extended to simulate a
Medical Expert System for automated diagnosis of
various diseases. The performance of the algorithm
can be compared with a variety of techniques.The
experiment may be repeated for more data sets and
with all risk factors of the disease.
6. REFERENCES
[1] Qeethara Kadhim Al-Shayea “Artificial Neural
Networks in Medical Diagnosis”, IJCSI International
Journal of Computer Science Issues, Vol. 8, Issue 2,
March 2011,pp. 150-154.
[2] R. Dybowski and V. Gant, “Clinical Applications
of Artificial Neural Networks” , Cambridge
University Press, 2007.
[3] Yegnanarayana, B., 2003, Artificial Neural
Networks, Prentice Hall of India Pvt. Ltd.
[4]
http:\\www.buzzle.com
–thyroid
problemssymptoms
of
hyperthyroidism
and
hypothyroidism
Figure5 : Neural Network
4390 ,performance met.
Training State ,epoch
[5] Ozyýlmaz and T. Yýldýrým. “Diagnosis of
thyroid disease using artificial neural network
methods”. in: Proceedings of ICONIP’02 9th
international conference on neural information
processing (Singapore: Orchid Country Club, 2002)
2033–2036
[6] E.T. Wong and M.W. Steffes, A fundamental
approach to the diagnosis of diseases of the thyroid
gland, Clinical Laboratory Medicine 4 (1984) 655–
670.
[7]Anupam Shukla, Prabhdeep Kaur, Ritu
Tiwari,R.R.Janghel” Diagnosis of Thyroid Disorders
using
Artificial
Neural
Networks”,
IEEE
International Advance Computing Conference (IACC
2009,pp. 1016-1020.
Figure 6: Regression Graph
[8]Anitha Thakur,surekha Bhanot, S.N.Mishra,“Early
diagnosis of Ishemia Stroke using Neural Network”,
Proceedings of International Conference on Man
Machine systems(IcoMMS),11-13 october 2009,Batu
Ferringh,Penang,Malsiya ,2B10-1 -2B10-5.
[9] Qeethara Kadhim Al-Shayea† and Itedal S. H.
Bahia,” Urinary System Diseases Diagnosis Using
Artificial Neural Networks”, IJCSNS International
Journal of Computer Science and Network S 118
ecurity, VOL.10 No.7, July 2010,pp 118-122.
Author
I completed my M.Sc. (IT) from
Graphic Era University Dehrdun
and also pursuing M.Tech (IT)
also
from
Graphic
Era
University, Dehrdun. My area
Research is Wireless Network
Security and Neural Networks. I
have
published
two
International Research Papers
Conferences.
in Journals And