ausblenden:
Schlagwörter:
-
Zusammenfassung:
The class of ambiguous stimuli, such as the Necker cube or the Rubin’s vase, is a powerful tool to
study the neuronal correlates of consciousness; under the constant physical stimulation, the conscious
experience of the stimulus spontaneously flips back and forth over time. Provided with careful
and strict control measures, it is possible to train monkeys to report their percepts as they see
such an ambiguous stimulus, which allows us to record neuronal activity directly related to the conscious
experience [Leopold,Maier, and Logothetis 2003 Journal of Consciousness Studies]. Previous
electrophysiological studies with concurrent behavioral measurements during ambiguous
percepts concentrated on the trial-by-trial relationship between the activity of isolated single neurons
and monkeys’ reports. In these studies, either the stimulus or the recorded neuron was carefully
selected so that the two alternative conscious experiences maximally differentiate the spike
counts of the recorded neurons. Focusing on single neurons ignores a potentially information-rich
signal: the temporal correlation in the spikes of neighboring neurons. We were interested to learn
how and when the firing of many neurons began to reflect the conscious perception of an ambiguous
stimulus over time. Specifically, we wanted to learn the extent towhich we could track the development
over time of a neural correlate of consciousness. To address this issue, we trained two
monkeys to report their percepts while they were seeing an ambiguous structure-from-motion stimulus.
We recorded neuronal activity from the motion sensitive areaMT, with 8-10 microelectrodes,
each of which was independently mobilized by micro-drives. We used a decoding approach to
quantify howmonkeys’ reports are correlated with the activity of the simultaneously recordedmultiple
neurons over time. For the decoding analysis, we trained a pattern classifier (regularized least
square classifier) using 70\% of trials and decoded the percepts fromneural activity for a test set (the
rest of 30\% of trials). The output from the classifier in the test set was submitted to signal detection
analysis to obtain non-parametric bias free measure of decoding performance (area under the
curve, A’). We used spike counts with a bin size of 100 msec and combined the counts from multiple
neurons in each time bin in an optimal manner with linear weights estimated by the classifier.
The time resolved decoding performance was compared between ambiguous and unambiguous
conditions. The ambiguity was manipulated via binocular disparity. The decoding performance attained
with >100 neurons in both conditions were very accurate (A’>.9. Chance decoding performance
is A’=.5), although a significant difference emerged between the ambiguous and
unambiguous conditions over time. For the unambiguous condition the decoding performance was
very accurate from shortly after the stimulus onset (A’>.9) and remained high throughout the stimulus
presentation. In a stark contrast, the decoding performance for the ambiguous condition built
up gradually (almost linearly) over time, and reached at the peak (A’\verb=~=.9) at around.4-.8 sec after the
stimulus onset. Our results show that the initial neuronal activity evoked by the onset of a stimulus
reflects the physical properties of the input and thus is less correlated with conscious percept, while the later activity is increasingly reflective of the conscious percept of the animal.
