Neural Networks versus Genetic Algorithms as Medical Classifiers - Foundations on Natural and Artificial Computation - page 392

Neural Networks versus Genetic Algorithms as
Medical Classifiers

Oscar Mar ın, Irene Perez, Daniel Ruiz, Antonio Soriano, and Joaquin D. Garc ıa

Department of Computer Technology. University of Alicante, PO 99,
03080 Alicante, Spain
{ omarin,iperez,druiz,soriano,jdgarcia } @dtic.ua.es

Abstract. In this article we want to assess the feasibility of using genetic
algorithms as classifiers that could be used in clinical decision support
systems, for urological diseases diagnosis in our case. The use of artificial
neural networks is more common in this field, and we have previously
tested their use with the same purpose. At the end of the document we
compare the obtained results using genetic algorithms and two different
artificial neural networks implementations. The obtained accuracy rates
show that genetic algorithms could be a useful tool to be used in the
clinical decision support systems field.

Keywords: neural networks, genetic algorithms, clinical decision sup-
port systems, urological disorders.

1 Introduction

Doctors often use as a basis to diagnose their expert knowledge, the responses
obtained from patients to questions about their illnesses and the results of rele-
vant clinical tests. This is a long and complex process which has no mathematical
precision and may be influenced by many external factors, which can impede the
experts to take the right decision about a final diagnosis and the possible treat-
ments to be performed. An example could be the case of a patient that may
suffer from various diseases, which makes him to give contradictory and unclear
answers to doctor's questions about his symptons. Other scenario could be the
case of a patient who has some clear ailments but the real cause of these is an
undetected problem; this fact would delay the start of an appropiate treatment
[1]. Also in situations of shortage of doctors, especially in rural areas, a clini-
cal decision support system could be used for prevention and early detection of
health problems [2].
In our case, we have used as basis of this article a database of tests performed
over patients with urological disorders. In this case, the knowledge about the ori-
gin of the detected dysfunctions depends largely on the experience of the experts,
and the research activity that they constantly perform. Specifically, in urology field
there are many disorders which exact diagnosis is dicult to do, due to the inter-
action with the neural system and the limited knowledge we have about it. Within
this environment, and based on machine learning theory, clinical decision support