Recognition of X-ray Images by Using Revised GMDH-type Neural Networks

Abstract

In this paper, X-ray images of the stomach are recognized by using the revised GMDH (Group Method of Data Handling)-type neural network algorithm. The revised GMDH-type neural networks can automatically organize themselves by using the heuristic self-organization method, which was developed in the GMDH algorithm and is very similar to the Genetic Algorithm. The structural parameters such as the useful input variables, the number of the neurons in the hidden layers, optimum neuron architectures in the hidden layers and the number of feedback loop calculations are automatically selected so as to minimize an error criterion defined as AIC (Akaike’s information criterion). AIC can not be used in the conventional multi-layered neural networks using the back propagation method but AIC can be used in the GMDH-type neural network algorithm because the GMDH-type neural networks use the conventional linear regression analysis in order to estimate the connection weights. In this paper, it is shown that this neural network algorithm is very useful in the recognition of X-ray images.