Independent component analysis of fMRI data: a model based approach for artifacts separation

Independent component analysis applied to functional magnetic resonance imaging is a promising technique for non invasive study of brain function. We examine the behavior of spatial ICA decomposition applying ICA to simulated data sets. We study the ICA performances in presence of movement correlated and uncorrelated with activation task, also taking into account the presence of rician distributed noise. We show that the presence of image artifacts due to simulated subject movement and MRI noise greatly affects the method ability to reveal the activation, especially in the presence of movement correlated with activation task. Spatial smoothing of data, before ICA, seems to overcome this problem, allowing us to retrieve the original sources also in the presence of both correlated movement and high noise level.