Independent component analysis with mixture density model and its application to localize the brain alpha activity in fMRI and EEG

Recently, independent component analysis (ICA) has been introduced to solve the blind source separation problem. In the original and extended versions of ICA, nonlinearity functions are fixed to have specific forms such as supergaussian or subgaussian, limiting their performance. In this paper, we utilized ICA with mixture density model such that any assumption about the source density is not required, thus better separation is possible by matching flexible parametric nonlinearity to any kind of density of sources. Through simulation studies, the algorithm was validated and its better performance was demonstrated in comparison to other versions of ICA. Then mixture density ICA was applied to functional magnetic resonance imaging (fMRI) and electroencephalogram (EEG) data to localize the independent sources for alpha activity. We found that there is a strong spatial correlation between the sources in fMRI and EEG, proving the usefulness of our approach in its application to source separation problem in biomedical signal processing.