Reverse correlation for studying context integration in border ownership assignment

Neural signals in area V2 carry information about the assignment of edges to figures (border ownership) indicating mechanisms that integrate image context information far beyond the classical receptive fields. Remote stimulus features seem to modulate the neural responses, but do not evoke responses by themselves. To elucidate the spatial integration structure of these mechanisms we devised a nonlinear reverse correlation method. As in previous studies of border ownership coding (Zhou et al., J Neurosci 20:6594, 2000), one edge of a rectangular figure was placed in the receptive field under study, and the border ownership signal was defined as the difference between the responses to a given local edge when this edge was part of a figure on one side of the receptive field or the other (standard test). We then presented the same displays with additively superimposed dynamic binary noise (e.g., 16 by 15 pixels, figure occupying 5 by 5) and computed the spike-triggered average (STA) of the noise for either location of figure. We expected that each noise pixel would enhance or reduce the visibility of the figure according on its location and contrast. For example, a noise pixel on the inside of the boundary of a bright figure would enhance the visibility of the boundary at that location if the noise was adding light, but reduce its visibility if the noise was subtracting light. Because border ownership modulation consists in an enhancement of responses for one figure location, or a reduction for the opposite location, we expected that the difference between the STAs for the two figure locations would reveal the features of the figure-ground display that contribute to border ownership modulation. The linear portion of the receptive fields would cancel in the difference. In some cells, the differential STA revealed critical features such as the corners at the end of the stimulating edge, or the remote edges of the figure. Sometimes a feature had a facilitatory influence on one side and a suppressive influence in the corresponding location on the other side. To quantify this we computed the average weights of the remote edges and of the corners in the differential STAs for each neuron. However, as yet, we did not find a positive correlation between these values and the border ownership modulation index of the cells. Thus, it is uncertain if the method can reveal the spatial structure of the border ownership mechanism.