Ứng dụng của mạng Neural trong chẩn đoán bệnh viêm gan

Thuy Thi Hong Truong et al Journal of SCIENCE and TECHNOLOGY 127(13): 81 - 86

81

NEURAL NETWORK APPLICATION TO THE DIAGNOSIS OF HEPATITIS

Thuy Thi Hong Truong*

, Nga Thi Hong Do

University of Medicine and Pharmacy – TNU

ABSTRACT

Neural network can be used in many different problems that existed in the relationship of input

and output. One of the biggest advantages of neural network is that it is able to solve the problems

which have no algorithm or too complicated algorithms. Problem diagnosis in medicine is a good

example. This paper has studied the application of neural networks in medical diagnosis, some

problems of building a decision support system diagnosis and testing neural networkʼs application

to hepatitis diagnosis based on training with Wiscosin hepatitis data, to provide an overview of

possible applications of powerful information technology in the field of medicine.

Keyworks: Neural Network, Medical diagnosis, Decision support system, Training, Wiscosin

hepatitis data.

INTRODUCING THE NEURAL

NETWORK MODEL IN DIAGNOSIS*

Most physicians have diagnosed diseases

based on the knowledge accumulated in

colleges, training courses, ... However, the

medical knowledge is often quickly out of

date, in order to diagnose, doctors must have

enough experience (for about 10-20 working

years). Besides, doctors also have difficulty in

diagnosing rare diseases or emerging

diseases. The solution can be used to bring

the benefits of computers to improve the

diagnostic capabilities such as: using the data

collected from experienced colleagues, they

have been using the experience gathered

owing to connecting to the world,... [4] so far,

neural networks have achieved many

remarkable achievements when applying to

many different areas of medicine such as

disease diagnosis, medical image analysis,

biomedical analysis,...

Neural network applications in medical

diagnosis support is, in fact, essentially

solving the problem of classification of

medical statistics. The biostatistics data are

often given in tabular format in which each

line is a record, each column is a symptom

and a column is to determine the diagnosis.

The input of the neural network will be the

first symptom and the output will be the

diagnosis.

* Email: [email protected]

SEVERAL ISSUES FOR DEVELOPING

DIAGNOSIS SUPPORT SYSTEMS

To build networks, which are capable of

decision support and have diagnostic

accuracy and high performance, we have to

solve some of the following issues:

Data preprocessing

This phase plays an important role in the

process of building systems by data sets in the

individual studies, which are often too small

to produce reliable results. Besides, the data

entry process and measurement errors also

appear due to typing error or not checking the

boundary conditions of the variables...

Therefore, we need to better define the

characteristic variables, remove or fix

incorrect data to build reliable systems.

According to experts, the number of neural in

hidden layer affects the generalization ability

of the network. If a neural network has a

small number of hidden layers, that will not

be defined, the structure of the network to

perform adequately in the training data, in this

case, it is considered to be unsuitable

(underfitting). Conversely, if the number of

hidden layer neural network is too big, the

network can not clearly define the decision

boundary in the vector space which is

affected by the level set by the properties of

the training data network. In this case, it is

considered to be too joints (overfitting). The