Spatial and temporal properties related to direction selectivity
of both simple and complex type visual cortex neurons were assessed
by cross-correlation analysis of their responses to random ternary
white noise. This stimulus consisted of multiple randomly placed
bars, each colored white, black, or gray with equal probability,
which were rerandomized every 5–10 ms. A first-order
cross-correlation analysis of a neuron's spike train with
the spatiotemporal history of the stimulus provided an estimate
of the neuron's linear spatiotemporal filtering properties.
A nonlinear correlation analysis measured the amount of interaction
for pair-wise combinations of bars as a function of their relative
spatial and temporal separations. The spatiotemporal orientation
of each of these functions was quantified using a “motion
energy index” (MEI), which was compared to the
neurons' direction selectivity measured with drifting sinewave
gratings. Both first-order and nonlinear correlation plots usually
showed s–t orientation whose sign
was consistent with the neuron's direction preference;
however, in many cases the MEI for first-order analysis
was weak compared to that seen in the nonlinear interactions.
The structures of the nonlinear interaction functions were also
compared with predictions from a conventional model of direction
selectivity based on a simple spatiotemporally oriented linear
filter, followed by an intensive nonlinearity (“LN
model”). These comparisons showed that some neurons'
data agreed reasonably well with such a model, while others
agreed poorly or not at all. Simulations of an alternative model
which combines signals from idealized lagged and nonlagged
front-end linear filters produce noise correlation results more
like those seen in the neurophysiological data.
