Gc. Deangelis et al., SPATIOTEMPORAL ORGANIZATION OF SIMPLE-CELL RECEPTIVE-FIELDS IN THE CATS STRIATE CORTEX .2. LINEARITY OF TEMPORAL AND SPATIAL SUMMATION, Journal of neurophysiology, 69(4), 1993, pp. 1118-1135
1. We have tested the hypothesis that simple cells in the cat's visual
cortex perform a linear spatiotemporal filtering of the visual image.
To conduct this study we note that a visual neuron behaves linearly i
f the responses to small, brief flashes of light are mathematically re
lated, via the Fourier transform, to the responses elicited by sinusoi
dal grating stimuli. 2. We have evaluated the linearity of temporal an
d spatial summation for 118 simple cells recorded from the striate cor
tex (area 17) of adult cats and kittens at ages 4 and 8 wk postnatal.
These neurons represent a subset of the population of cells for which
we have described the postnatal development of spatiotemporal receptiv
e-field structure in the preceding paper. Spatiotemporal receptive-fie
ld profiles are constructed, with the use of a reverse correlation tec
hnique, from the responses to random sequences of small bar stimuli th
at are brighter or darker than the background. Fourier analysis of spa
tiotemporal receptive-field profiles yields linear predictions of the
cells' spatial and temporal frequency tuning. These predicted response
s are compared with spatial and temporal frequency tuning curves measu
red by the use of drifting, sinusoidal-luminance grating stimuli. 3. F
or most simple cells, there is good agreement between spatial and temp
oral frequency tuning curves predicted from the receptive-field profil
e and those measured by the use of sinusoidal gratings. These results
suggest that both spatial and temporal summation within simple cells a
re approximately linear. There is a tendency for predicted tuning curv
es to be slightly broader than measured tuning curves, a finding that
is consistent with the effects of a threshold nonlinearity at the outp
ut of these neurons. In some cases, however, predicted tuning curves d
eviate from measured responses only at low spatial and temporal freque
ncies. This cannot be explained by a simple threshold nonlinearity. 4.
If linearity is assumed, it should be possible to predict the directi
on selectivity of simple cells from the structure of their spatiotempo
ral receptive-field profiles. For virtually all cells, linear predicti
ons correctly determine the preferred direction of motion of a visual
stimulus. However, the strength of the directional bias is typically u
nderestimated by a factor of about two on the basis of linear predicti
ons. Consideration of the expansive exponential nonlinearity revealed
in the contrast-response function permits a reconciliation of the disc
repancy between measured and predicted direction selectivity indexes.
5. Overall, these findings show that spatiotemporal receptive-field pr
ofiles obtained with the use of reverse correlation may be used to pre
dict a variety of response properties for simple cells. These results
are generally consistent with recent theoretical work in which simple
cells are modeled as the combination of a linear spatiotemporal filter
, an exponent nonlinearity, and a contrast normalization mechanism.