A mutual information approach to data integration for alzheimer's disease patients

Abstract

Clinical data alignment plays a critical role in identifying important features for significant experiments. A central problem is data fusion i.e., how to correctly integrate data provided by different labs. This integration is done in order to increase ability of inferring target classes of controls and patients. Our paper proposes an approach based both on a information theoretic perspective, generally used in a feature construction problem [3] and on the approximated solution for a mathematical programming task (i.e. the weighted bipartite matching problem [6]). Numerical evaluations with two competitive approaches show the improved performance of the proposed method. For this evaluation we used data sets from plasma / ethylenediaminetetraacetic acid (EDTA) of controls and Alzheimer patients collected in three different hospitals. © 2009 Springer Berlin Heidelberg.