Trauma Care Classification

The objective of this Phase I project is to develop a processing architecture that accepts data from multiple inputs an provide likely trauma survival ratings. The processing architecture is based on a neural network configuration expanded to encode in a direct and unambiguous manner statistical information This creates a hybrid architecture that permits the best attributes of both domains to be utilized for the classification of data corresponding to trauma survival predictive variable. The approach is based on the realization that no single techniques is capable of solving by itself the more difficult aspects of the highly complex trauma survival classification problem. Thus, there exists a strong need to coherently assemble the best elements of different techniques so as to reenforce the positive contributions by each and to neutralize, through complemention , their deficiencies. The architecture chosen is Gaussian Potential Function Network (GPFN) consisting of Gaussian Potential Functions Units (GPFU) with some key parameters determined by the statistical properties of the input feature vectors. Other network structural parameters are determined through a "training" process that aims to yield a network output compatible to the object class the input vector belongs to.