Identifying activated brain regions from fMRI data based on blind non-negative component separation technique

Identifying brain regions activated by a task or stimulus is important for exploring the information processing mechanism of brain. Independent Component Analysis (ICA) based method is one of the well-used approaches for functional magnetic resonance imaging (fMRI) data analysis. It firstly decomposes fMRI data into statistically independent components and then identifies the activated brain regions by selecting the voxels of the decomposed consistently task related (CTR) components, whose time course most closely match the task function. However, it is found that the independence assumption is sometime violated in practice and the ICA based method does not select for independence in decomposing fMRI data. In this work, we introduce a new method for activated brain region localization based on blind separation of mixtures of non-negative sources. The proposed method emphasizes the source non-negativity and sparsity, which is considered more realistic to fMRI data. Data analysis results in this paper show that the proposed method is more suitable for fMRI data analysis.