Responses in lateral geniculate nucleus (LGN) exhibit size tuning (Jones and Sillito, 1991). This tuning is predicted by a simple model (Bonin et al. Soc Neurosci Abs, 2002). Model LGN neurons have a linear receptive field (RF, a difference of Gaussians), and receive a divisive signal from a contrast integration field. This field is a region around the RF that estimates local contrast by filtering the retinal image and computing variance. What are the origins of the contrast integration field?

We recorded responses of LGN neurons in anesthetized, paralyzed cats to sums of drifting sinusoidal gratings varying in contrast, diameter, orientation and spatiotemporal frequency.

The model provided good fits to the data. It explains: (1) size tuning; (2) contrast saturation; (3) masking caused by superimposed gratings (present in 19/25 cells). Moreover, it correctly predicts that size tuning increases with contrast, and is absent at low contrast. The contrast integration field is 2.4 ± 0.4 (s.e.m., N=28) times larger than the RF surround.

The contrast integration field is broadly selective for stimulus attributes:

(1) It depends weakly on orientation. In 13/19 cells masking was independent of orientation (p<0.05, ANOVA).

(2) It is selective for a broad range of spatial frequencies. Its selectivity was often lowpass, with high frequency cutoffs similar to those of the RF.

(3) It is selective for a broad range of temporal frequencies. High frequency cutoffs correlated with those of the RF (r=.76, N=25).

The contrast integration field explains size tuning and other properties of LGN neurons. Its origins might include cortical feedback (Murphy and Sillito, 1987), but its broad selectivity for orientation and spatiotemporal frequency seems more consistent with a retinal or thalamic origin (Shapley and Victor, 1978).