Most neuroimaging experiments aim to identify brain areas whose activation
correlates tightly with an aspect of the subjects' behavioral task. If the
logic of neuroimaging is correct, however, it should also be possible to reverse
this sequence of operations. Once we understand the function of a given brain
area or network of areas, it should be possible to use on-line activation
measurements to infer what kind of task the subject was performing. As a first
step in this direction, we report here that functional magnetic resonance
images (fMRI) of motor cortex have sufficient signal-to-noise ratio to accurately
predict the subject's manual behavior on a single-trial basis with close to
100% accuracy.
