Sunteți pe pagina 1din 15

ARTIFICIAL NEURAL NETWORKS IN MEDICINE: AN APPLICATION

OF BREAST CANCER DIAGNOSIS

Name of the University

NAME

NO:

Faculty

Degree: PhD Thesis Proposal

Referee:

Supervisor:

NAME

NO:

29th March 2017


Abstract Diagnosis of breast cancer has been slow in most developing countries due

to scarcity of cancer medical units in major hospitals. The use of various Artificial
Intelligence methods for medical diagnosis of diseases has recently become widespread.

In this paper, Back Propagation Rule will used for training the model using two types of
validations. Furthermore, in this paper, it will be expected that NN model can classify
the data with excellent accuracy and this will lead to an automated medical diagnosis

system for the breast cancer.

i
Contents

1 Introduction 1

1.1 Statement of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 1


1.2 Justification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4.1 Broad Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4.2 Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 3


1.5 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Literature Review 4

2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.1 Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Empirical Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Research Method and Specification 7

Bibliography 10

ii
1 Introduction

Artificial Neural Network (ANNs) is a branch of Artificial Intelligence that has been

accepted as a new technology in computer science Ganesan et al. (2010). ANNs are
broadly used in science and technology with applications in various branches of chemistry,
physics, and biology Amato et al. (2013). Further, ANNs Neural Networks are currently

a revered research area in medicine, more so in the fields of radiology, urology, cardiology,
and oncology and among others. However, it also has extensive application in many areas

such as education, business, engineering, and manufacturing.

Breast cancer is cancer of the 21st century as it affects today more women in the world.

Breast cancer, similar to other types of cancer, begin with faster and uncontrolled out-
growth and replication of parts of breast tissue that is dependent on its potential harm,
can be categorized into benign and malignant types Khosravanian and Ayat (2016). Be-

nign tumors represent an unnatural outgrowth but rarely lead to a patients death. Con-
sequently, the fight against cancer is far-off from completed. Medicine improvement on

all fronts to expand the care of patients and defeat this disease of the century. For of
this, it is indispensable that several disciplines continue to make their involvement and

particularly prediction based on Artificial Intelligence (AI). Therefore, ANNs as a branch


of AI can be a powerful tool for distributed diagnosis.

Cancer is a class of diseases in which a group of cells display uncontrolled growth and

invasion and sometimes metastasis. Breast cancer is a type of cancer that starts in the
breast, usually in the inner lining of the milk ducts or lobules HariKumar et al. (2009).

1.1 Statement of the Problem

The primary risk factors for breast cancer have often been identified as sex, age, hormone,
high-fat diet, alcohol intake, environmental factor, and genetic abnormality. These risk

factors are hardly taken into account during the cancer diagnosis. However, research by
Jirtle and Skinner (2007) these factors are rarely considered during diagnosis of breast

cancer. Knowledge of Neural Networks have shown that if the elements are modeled using

1
ANN, it can be used to is cancer diagnosis especially, breast cancer diagnosis.

1.2 Justification

Cancer diagnosis specifically involves invasive and noninvasive methods which are ethic-
ally subjective based on personalities. These methods may make people shy away from

diagnosis hence there is a need for an assurance for people on the use of majorly other
methods to diagnose cancer. This project is relevant since it will restore hope to people

on cancer diagnosis since in most cases, like biopsy, cancer cannot be detected initially.
This forces the patients to undergo the same process which many regards as an ordeal.

Existing diagnostics methods of breast cancer such as the experience of biopsy can in-
crease the threat of breast cancer Zollinger and Mihatsch (2012). Other diagnosis such
as mammography which is associated with a small amount of radiation that is harmful

to human health is also common. Breast Magnetic Resonance Imaging (MRI), which
is an imaging test creates detailed pictures of the inside of the breasts is also common

Heywang-Kobrunner et al. (2011). Although breast MRI is not regularly used for breast
cancer screening, it often helps detect breast cancer in its earliest stage for some women

who are at high risk for developing the disease. The increased sensitivity associated with
MRI may cause areas of the breast that do not have cancer to appear unnatural, giving
rise heightened count of false-positive test results. This may lead to unnecessary biopsies

and increased anxiety for many women.

However, the use of various Artificial Intelligence methods for medical diagnosis of diseases

has recently become widespread Mandal and Banerjee (2015). In addition, AI systems
help physicians as a diagnosis assistant. Consequently, many ANN, Rough Set, Decision

Tree, Bayesian Network are very popular for this purpose. Mandal and Banerjee (2015)
in their paper on, Multilayer Feed Forward Neural Network can be used to detect cancer
from Microarray Data and UCI Machine Learning Data. In this paper, Back Propagation

Rule will used for training the model. Throughout that paper, two types of validations
shall be performed: cross-validation and new case testing for above two data-sets with

2
a different combination of hidden layers and corresponding nodes. In this paper and in

relation to Mandal and Banerjee (2015) paper, it will be expected that NN model can
classify the data with excellent accuracy and this will lead to an automated medical

diagnosis system for the particular disease.

1.3 Research Questions

The study will be guided by the following research questions:

1. What is the effect of primary risk factors of breast cancer diagnosis in ANN in AI?

2. What would be the result of cancer diagnosis if Feed-forward or backward propaga-

tion are used in ANN in AI?

1.4 Objectives

1.4.1 Broad Objectives

The main objective of the project is to use feed-forward neural network in back-propagation
to develop rule based neural networks in classification problems.

1.4.2 Specific Objectives

1. To explore how Neural Networks are used in Artificial Intelligence.

2. To transform cancer risk factors in cancer diagnostic aspects into suitable forms
and parameters for an adaptive system coding.

3. To train ANN to be used in cancer diagnosis based on patients data.

1.5 Scope

The study aims at reviewing risk factors associated with breast cancer, thereafter de-
veloped appropriate tool that can be used in cancer diagnosis and detection.

3
2 Literature Review

2.1 Overview

ANNs are useful in medicine for diagnosis and detection. They are capable of processing
incomplete and imprecise data and detect non-linear relations in the data Asadi et al.

(2012). Literature review on the neural network in medicine is vast Amato et al. (2013).
In this chapter, I will summarize some of the most significant developments in neural
networks.

The thrust behind artificial neural network research has come from a desire to improve
on the limitations of the modern serial computer and tap into this concept of parallel

processing. While a computer can make complex mathematical computations much faster
than the human brain, biological neurons can perform tasks, such as image processing

and speech recognition, with the speed and precision that serial computers cannot match
Widrow and Walach (2008). Fields such as artificial intelligence have attempted to adapt
to the concept of parallel processing, but this connection to biological neurons is mostly

lost in the area of modeling. Instead, modelers have focused on the ability of ANNs to
estimate highly nonlinear functions Resop (2006).

2.1.1 Neural Networks

A neural network is a robust computational data model that can capture and represent
complex input/output relationships Principe et al. (2005). ANNs on the other hand,

are relatively crude electronic systems of neurons based on the neural structure of the
brain.

The advantage and power of neural networks lie in their ableness in the representation

of both linear and non-linear relationships besides having the capability to learn rela-
tionships directly from the data being modeled Principe et al. (2005). When it comes to

modeling and non-linear characteristic data, conventional linear models are inadequate
Principe et al. (2005).

4
On the other hand, classification is one of the most active research and application areas

of neural networks Zhang (2000). The foundation of neural networks is the simulation of
biological neurons. Consequently, the neural network is information driven rather than

data driven. Also, neural networks offer a powerful method to explore, classify, and
identify patterns in data Principe et al. (2005).

2.2 Empirical Review

Many aspects of Artificial Intelligence have been proposed and developed to aid in dia-
gnosis of Neural Network. For instance, the Computer-Aided-Diagnosis has been sugges-

ted to be useful in medical prognosis, thus making the fuzzy logic and ANN as the basis for
the intelligent systems (Sahiner et al.; 1996; Kiyan and Yildirim; 2004). The numerous
instances where AI has been employed for breast cancer diagnosis have concluded that

any methods that adopt ANN are very vital in cancer diagnosis (Hazlina et al.; 2004;
Fadzilah and Azliza; 2004). Most of these papers used MATLAB, a high performance

and easy to use environment; for the diagnosis and classification of the breast cancer.

However, cancer diagnosis and therapeutic strategies boast of several methods as por-

trayed in many pieces of research conducted. Researchers concluded their research by


highlighting comparisons among many neural network applications included to provide
a global view of computational intelligence with neural networks in medical imaging

Jiang et al. (2010).

Use of various Artificial Intelligence methods for medical diagnosis of diseases has recently

become widespread Mandal and Banerjee (2015). According to Mandal and Banerjee
(2015) AI intelligent systems help physicians in diagnosis. Consequently, many ANN,

Rough Set, Decision Tree, Bayesian Network are very popular for this purpose. Mandal and Banerjee
(2015) concluded in their paper on Multilayer Feed Forward Neural Network that ANN
be used to detect cancer from Micro-array Data and UCI Machine Learning Data. In that

article Back Propagation Rule was used for training the model. Throughout that paper,
two types of validations were performed: cross-validation and new case testing for above

5
two data sets with a different combination of hidden layers and corresponding nodes. In

that paper, it was found that NN model can classify the data with excellent accuracy
and this will lead to automated medical diagnosis system for the particular disease.

In the above paper without further exhausting literature materials from the past and
present, most of the literature has concentrated cancer data analysis rather than the
diagnosis of the disease. Literature like (Khosravanian and Ayat; 2016; Al-Hiary et al.;

2012; Goodman; 1995; Sridhar et al.; 2014), their studies concentrated on the use of
ANN in diagnosis of various types of cancers. However, in all the studies highlighted, the

primary risk factors for breast cancer, which have been identified as sex, age, hormone,
high-fat diet, alcohol intake, environmental factor, and genetic abnormality, have not

been used in modeling cancer diagnosis. In this work, these risk factors will be utilized,
and the difference found in this study and other studies noted.

6
3 Research Method and Specification

The material for ANNs programm installations and knowledge will be based on existing

information from both the internet and books. Nonetheless, critical materials will be:
An appropriate software that can offer a platform for ANNs designing and training,
specifically MeMbrain and Java neural network framework Encog for constructing and

testing our different ANNs.

The primary risk factors for breast cancer have often been identified as sex, age, hormone,

high-fat diet, alcohol intake, environmental factor, and genetic abnormality. When train-
ing artificial neural networks, a set of input and output data and weights must be adjusted

to give the same output seen in training data. At the same time, ANN must not be made
too particular, since the interest in training ANN is not to give precise results for the
training but to give incorrect results for all other data. When this happens, ANN has

been over-fitted.

The training process of ANN can be seen as an optimization problem, where one wishes to

minimize the mean square error training data. This issue can be solved in many different
ways, ranging from standard optimization heuristics like simulated annealing, through

unique optimization techniques like genetic algorithms to specialized gradient descent


algorithms like Backpropagation. The most used algorithm is the Back-propagation al-
gorithm. However, the extent of adjustment of weight in each iteration is the major

limitation in the algorithm.

Breast cancer risk factors can be scaled to weighted functions and trained based on past

data of patients as illustrated in the Figure 1.

7
process.jpg

Figure 1: Flow chart of ANN training processes.

8
The data shall be presented as in Figure 2.

presentation.jpg

Figure 2: Data presentation

9
Bibliography

Al-Hiary, H., Alhadidi, B. and Braik, M. (2012). An implemented approach for poten-

tially breast cancer detection using extracted features and artificial neural networks,
Computing and Informatics 31(1): 225244.

Amato, F., Lopez, A., Pe


na-Mendez, E. M., Va
nhara, P., Hampl, A. and Havel, J. (2013).

Artificial neural networks in medical diagnosis.

Asadi, S., Hadavandi, E., Mehmanpazir, F. and Nakhostin, M. M. (2012). Hybridization

of evolutionary levenbergmarquardt neural networks and data pre-processing for stock


market prediction, Knowledge-Based Systems 35: 245258.

Fadzilah, S. and Azliza, M. (2004). Web-based neurofuzzy classification for breast cancer,

Proceedings Second International Conference on Artificial Intelligence in Engineering


&Technology, pp. 383387.

Ganesan, N., Venkatesh, K., Rama, M. and Palani, A. M. (2010). Application of neural
networks in diagnosing cancer disease using demographic data, International Journal

of Computer Applications (0975-8887) 1(26).

Goodman, P. H. (1995). Comparing the prediction accuracy of artificial neural networks


and other statistical models for breast cancer survival.

HariKumar, R., Vasanthi, N. and Balasubramani, M. (2009). Performance analysis of


artificial neural networks and statistical methods in classification of oral and breast can-

cer stages, International Journal of Soft Computing and Engineering (IJSCE) Volume
2.

Hazlina, H., Sameem, A., NurAishah, M. and Yip, C. (2004). Back propagation neural

network for the prognosis of breast cancer: comparison on different training algorithms,
Proc. 2nd. Int. Conf. Artificial Intelligence in Engineering & Technology (ICAIET),

pp. 445449.

10
Heywang-Kobrunner, S. H., Hacker, A. and Sedlacek, S. (2011). Advantages and disad-

vantages of mammography screening, Breast care 6(3): 199207.

Jiang, J., Trundle, P. and Ren, J. (2010). Medical image analysis with artificial neural
networks, Computerized Medical Imaging and Graphics 34(8): 617631.

Jirtle, R. L. and Skinner, M. K. (2007). Environmental epigenomics and disease suscept-


ibility, Nature reviews genetics 8(4): 253262.

Khosravanian, A. and Ayat, S. (2016). Diagnosing breast cancer type by using probabil-

istic neural network in decision support system, Knowledge Engineering 2(1): 7376.

Kiyan, T. and Yildirim, T. (2004). Breast cancer diagnosis using statistical neural net-

works, Journal of electrical & electronics engineering 4(2): 11491153.

Mandal, S. and Banerjee, I. (2015). Cancer classification using neural network, Interna-
tional Journal 172.

Principe, J., Lefebvre, C., Lynn, G., Fancourt, C. and Wooten, D. (2005). Neurosolutions
documentation, NeuroDimension, Inc.: Gainesville, FL .

Resop, J. P. (2006). A comparison of artificial neural networks and statistical regression

with biological resources applications, PhD thesis.

Sahiner, B., Chan, H.-P., Petrick, N., Wei, D., Helvie, M. A., Adler, D. D. and Good-
sitt, M. M. (1996). Classification of mass and normal breast tissue: a convolution
neural network classifier with spatial domain and texture images, IEEE transactions

on Medical Imaging 15(5): 598610.

Sridhar, B., Reddy, K. and Prasad, A. (2014). Detection of lesions in medical image
using an artificial neural networks and morphological filters., Computer Science &

Telecommunications 43(3).

Widrow, B. and Walach, E. (2008). Thirty years of adaptive neural networks: Perceptron,

madaline, and backpropagation, Adaptive Inverse Control: A Signal Processing Ap-


proach, Reissue Edition, IEEE .

11
Zhang, G. P. (2000). Neural networks for classification: a survey, IEEE Transactions on

Systems, Man, and Cybernetics, Part C (Applications and Reviews) 30(4): 451462.

Zollinger, H. U. and Mihatsch, M. J. (2012). Renal pathology in biopsy: light, electron

and immunofluorescent microscopy and clinical aspects, Springer Science & Business
Media.

12

S-ar putea să vă placă și