Knowledge-based system for prognosis of specific types of cancer using Elman neural network

Abstract

This paper focuses on the classification problem of high dimensional patterns and especially of socio-demographic cancer questionnaires. The purpose of this study is to define a predictive indicator of a published clinical study regarding the influence of Hormone Replacement Therapy (HRT) on the growth of cancers, including breast, ovarian, endometrial, and colon cancers. The proposed study, in the preparation stage, combines independent factors of this research using a Bayesian model in order to achieve a normalizing data linked by significant relevant properties of these factors. The specific goal is to determine a standard threshold value in which an independent self-organizing system will decide the correlation between the normalizing data of the preprocessing stage via a well-fitted, recurrent Elman neural network using a threshold factor which is called the distance value. A case study involving a dataset of published clinical research is used and the evaluated procedure is implemented by a well-fitted t-test control.Artificial neural networks (ANNs) and decision trees (DTs) have been used in cancer detection and diagnosis for nearly 20 years [1] . This paper focuses on the classification problem of unrelated characteristics, especially those of epidemiologic questionnaires. The baseline questionnaire selected for the present study [2] is based on socio-demographic features (age, marital status, race, and education), height and body weight, generative and obstetric history, family and personal medical history, cigarette smoking, and use of hormone replacement therapy (HRT). The anonymous questionnaire results are related to findings of Endometrial Thickness (ET) measurements, which were taken using transvaginal ultrasonic examination [2] .The preparation stage of data processing is the most significant procedure in stochastic and probabilistic decision systems [3] an important subclass of which is neural networks. Modeling has an enormous impact on the success of a wide range of such complex data analyses as data mining and feature-extraction procedures [3,4] , mainly because the quality of the input data in neural network models strongly influences the results of analyses [3,5] and the efficiency of their performance, as wrongly prepared data are likely to result in problematic analyses. The appropriate pre-processing of input