Independent component analysis by convex divergence minimization: applications to brain fMRI analysis

Abstract

A class of independent component analysis (ICA) algorithms using a minimization of the convex divergence, called the f-ICA, is presented. This algorithm is a super class of the minimum mutual information ICA and our own /spl alpha/-ICA. The following properties are obtained: 1) the f-ICA can be implemented by both momentum and turbo methods, and their combination is also possible; 2) the formerly presented /spl alpha/-ICA can claim an equivalent form to the f-ICA if the design parameter /spl alpha/ is chosen appropriately; 3) the f-ICA is much faster than the minimum mutual information ICA; and 4) additional complexity required to the divergence ICA is light, and thus this algorithm is applicable to a large amount of data via conventional personal computers. Detection of human brain areas that strongly respond to moving objects is reported in this paper.