We present a new method for  separating  multiple task-related events
   from other physiological and physical  events  revealed by functional
   Magnetic Resonance Imaging (fMRI) signals.  The method separates fMRI
   signals into different events by minimizing the  spatial and temporal
   correlations between events.  The  method  was  used  to analyze fMRI
   data sets from subjects performing real or imagined motor tasks.  The
   method   successfully   separated   task-related   events  (e.g,  the
   activation  in  primary  motor  cortex  during  real  motion  and  in
   supplementary  motor  area  during  imagined  motion) from  unrelated
   events  (e.g.,  an  activation  in  auditory  area  and slow  changes
   probably associated with head drifts).

   (Neurocomputing. Vol~49, page 227-239. 2002)

   (Papers' Index of Dawei Dong)