True Factor Analysis in Medical Imaging: Dealing with High-Dimensional Spaces

Abstract

This article presents a new method for discovering hidden patterns in high-dimensional dataset resulting from image registration. It is based on true factor analysis, a statistical model that aims to find clusters of correlated variables. Applied to medical imaging, factor analysis can potentially identify regions that have anatomic significance and lend insight to knowledge discovery and morphometric investigations related to pathologies. Existent factor analytic methods require the computation of the sample covariance matrix and are thus limited to low-dimensional variable spaces. The proposed algorithm is able to compute the coefficients of the model without the need of the covariance matrix, expanding its spectrum of applications. The method’s efficiency and effectiveness is demonstrated in a study of volumetric variability related to the Alzheimer’s disease.