Blind sensor characteristics estimation in a multi-sensor network applied to fMRI analysis

We propose an algorithm, based on blind source separation methods, for blindly estimating the sensor characteristics of a multi-sensor network, whose structure is also unknown. From the observed sensor outputs, the non-linearities are recovered using a well-known Gaussianization procedure. The underlying sources are then reconstructed using spatial decorrelation. Application of this robust algorithm to data sets acquired through functional magnetic resonance imaging (fMRI) lead to detecting a distinctive 'bump' of the BOLD (blood oxygenation level dependent) effect at larger activations.