A BFGS-ICA algorithm and application in localization of brain activities

The natural gradient and fixed-point algorithm are two of the most popular algorithms in independent component analysis (ICA). However, there still remain some problems to be solved in application to the processing of the functional magnetic resonance imaging (fMRI) data. Based on the BFGS quasi-Newton algorithm, this paper presents a novel BFGS-ICA algorithm framework in performing localization of brain activities with fMRI data. The new BFGS-ICA algorithm possesses properties of good convergence and immunity of initial point sensitivity. The convincing results of its application in fMRI show the potential of BFGS-ICA in detection of the brain activities.