Neural networks in visualization of multispectral medical images

In this paper we demonstrate how a neural network can be used for visualizing multispectral medical images. The network used is a feedforward network that uses an unsupervised modification of the backpropagation algorithm with Sammon's stress function as the objective function. The aim of using the network is to provide as much information as possible from a multispectral image in an image with fewer components without making any rigid classification into different categories. A nonlinear mapping is made from the original n-dimensional feature space onto an m-dimensional pixel space where m < n. The method is an alternative to linear transformations such as principal component analysis.