Experiments with Neural Networks as Prognostics Engines for Patient Physiological System Health Management

Abstract

Prognostics and prediction of patients‟ short term physiological health status is of critical importance in medicine because it affords medical interventions that prevent escalating medical complications. Accurate prediction of the patients‟ health status offers many benefits including faster recovery, lower medical costs and better clinical outcomes. This study proposes a prognostics engine to predict patient physiological status. The prognostics engine builds models from historical clinical data using neural network as its computational kernel. This study compared accuracy of various neural network models. Given the diversity of clinical data and disease conditions, no single model is ideal for all medical cases. Certain algorithms are more accurate than others depending on the type, amount and diversity of possible outcomes. Thus multiple neural network algorithms are necessary to build a generalizable prognostics engine. The study proposes using an oracle, an overseer program to select the most accurate predictive model that is most suited for a particular medical prediction among several neural network options.