Medical image analysis of chest images using deep multi-layered GMDH-type neural network and convolutional neural network

Abstract

In this study, hybrid deep neural network is organized using the deep multi-layered Group Method of Data Handling (GMDH)-type neural network and the Convolutional Neural Network (CNN) and it is applied to the medical image analysis of chest images. In the deep GMDH-type neural network, the hyper parameters such as number of hidden layers, type of the neural network and useful input variables, are automatically selected to minimize prediction error criterion defined as Akaike’s Information Criterion (AIC) or Prediction Sum of Squares (PSS) and the deep neural networks with the optimal complexity are automatically organized. This deep neural network algorithm is applied to medical image analysis of chest images, and the organs such as liver, heart and bone, are recognized and these regions are extracted accurately using the deep multi-layered GMDH-type neural networks.