Some examples for solving clinical problems using neural networks

Abstract

In this paper neural networks are presented for solving some pharmaceutical problems. We have predicted and prevented patients with potential risk of post-Chemotherapy Emesis and potentially intoxicated patients treated with Digoxin. Neural networks have been also used for predicting Cyclosporine A concentration and Erythropoietin concentrations. Several neural networks (multilayer perceptron for classification tasks and Elman and FIR networks for prediction) and classical methods have been used. Results show how neural networks are very suitable tools for classification and prediction tasks, outperforming the classical methods1. In a neural approach it is not strictly necessary to assume a specific relationship between variables and no previous knowledge of the problem is either necessary. These features allow the user better generalization performance than using classical methods. Several software applications have been developed in order to improve clinical outcomes and reduce costs to the Health Care System. © Springer-Verlag Berlin Heidelberg 2001.