Non independent strategies for blind source separation in functional MRI

This paper explores spatial decomposition of functional MR images by several extended blind source separation (BSS) techniques in order to detect the brain activation. The four techniques are based on different criteria and assumptions like mutual statistical independence or non-Gaussianity of sub-band sources, spatial linear predictability, spatial-temporal decorrelation and non-stationarity of the colored sources. We conclude that the detection of activation in fMRI experiments is possible in many cases by exploiting these characteristics of the sources extracted from raw data.