|
||||||||||||||
|
|
|
||||||||||||
Inferring individual perceptual experience
from MEG: Robust statistics approach Andrey Zhdanov Infei Inferring person's perceptual experience
from accompanying brain signals is one
of the primary goals of functional neuroimaging. Brain-related features that
mark the subjective experience can be particularly useful in evaluating and
monitoring pathological mental states such as in psychiatric and neurological
disorders. Bistable
perceptual phenomena, in which perception alternates between several
competing interpretations while physical attributes of the stimulus remain
constant provides an ample model for studying the individual's experience.
Binocular Rivalry (BR) is a particular example of such phenomena in which
although two images are presented simultaneously to the subject's brain, the
subjective perception alternates sporadically every few seconds between the
presented images. We
introduce a robust statistical approach for inferring the individual's
perceptual experience of face dominating over house during BR experiment from
accompanying magnetoencephalographic (MEG) recordings. To this end we used
patterns of the MEG signals invoked by presenting each of the competing
images separately to each eye (control condition) to find optimal projections
of the sensory data, via a regularized Fisher Liner Discriminant modeling, in
which the regularizer is chosen in several ways. The Fisher Projection
Coefficients are used to create maps that indicate scalp regions which are
most informative in distinguishing between the images in the control
condition. We demonstrate the difference in maps obtained by different
regularizers (see figure 1) and show that projecting the signals obtained
during BR onto ‘optimal’ maps yields prediction accuracy for the perceptual
appearance of face higher than what might be expected by chance. |
|
|||||||||||||
|