Scale-invariance of receptive field properties in primary visual cortex

Abstract

Background: Our visual system enables us to recognize visual objects across a wide range of spatial scales. The neural mechanisms underlying these abilities are still poorly understood. Size- or scale-independent representation of visual objects might be supported by processing in primary visual cortex (V1). Neurons in V1 are selective for spatial frequency and thus represent visual information in specific spatial wavebands. We tested whether different receptive field properties of neurons in V1 scale with preferred spatial wavelength. Specifically, we investigated the size of the area that enhances responses, i.e., the grating summation field, the size of the inhibitory surround, and the distance dependence of signal coupling, i.e., the linking field.

Results: We found that the sizes of both grating summation field and inhibitory surround increase with preferred spatial wavelength. For the summation field this increase, however, is not strictly linear. No evidence was found that size of the linking field depends on preferred spatial wavelength.

Conclusion: Our data show that some receptive field properties are related to preferred spatial wavelength. This speaks in favor of the hypothesis that processing in V1 supports scale-invariant aspects of visual performance. However, not all properties of receptive fields in V1 scale with preferred spatial wavelength. Spatial-wavelength independence of the linking field implies a constant spatial range of signal coupling between neurons with different preferred spatial wavelengths. This might be important for encoding extended broad-band visual features such as edges.

Figure 1

Stimulation protocol. While the monkey performed a fixation task, grating patches of 7 different spatial wavelengths and 6 different sizes were presented centered on each cRF. Orientations of stimuli were optimal for each recording site.

Figure 2

Mapping of size and wavelength preference. Determining preferred spatial wavelength and summation field size. Two example recording sites a and b. (A) Interpolated multiple unit activity to stimuli of different sizes and spatial wavelengths. (B) Distribution of 1000 global maxima calculated via the bootstrap method. (C) Mean (green line) and variance (broken line) of preferred spatial wavelength estimates as a function of stimulus size. The green diamond indicates the mean of the distribution from B.

Figure 3

Preferred Wavelength and ΣRF. Joint distribution of preferred spatial wavelength and summation field size (135 out of 152 recorded channels; two monkeys, 1 hemisphere each). Dark blue line: model of constant ΣRF size. Green line: model of ΣRF size scaled with preferred spatial wavelength. (A) Preferred spatial wavelength and absolute summation field size show a significant positive correlation for both monkeys. Pink and light blue line: fit of the model with additive term as described in Discussion. (B) Preferred wavelength and relative summation field size show a significant negative correlation for both monkeys. Pink and light blue line: marginal distributions of relative summation field size.

Figure 4

Preferred Wavelength and stimulus size. Dependence of preferred spatial wavelength on stimulus size. Results from monkey K (data from monkey B are similar). Green line indicates the preferred spatial frequency estimate at a given stimulus size. The gray area denotes the population mean of the individual variances. (A) Normalized preferred spatial wavelength is plotted with respect to absolute stimulus size. Variance of the estimate decreases as stimulus size increases. (B) Normalized preferred spatial wavelength is plotted with respect to normalized stimulus size. The blue line indicates the number of recording sites available for averaging at this particular normalized stimulus size. Variance of the estimate decreases as normalized stimulus size approaches 0, i.e., summation field size, but stays constant for larger stimuli. Significant deviations from preferred spatial wavelength are coded in red.

Figure 5

Preferred Wavelength and surround size. ΣRF and surround size measures from the DOG method. (A) Comparison of ΣRF size estimated with the bootstrap and the DOG method. The two measures are highly correlated (see text for details). (B) Joint distribution of preferred spatial wavelength and surround size (82 out of 152 recorded channels; two monkeys, each 1 hemisphere). Dark blue line: model of constant surround size. Preferred spatial wavelength and surround size show a significant positive correlation for both monkeys (see text for details). Pink and light blue line show the regression lines calculated via principal component analysis for monkey K and B, respectively. The black dots show the corresponding ΣRF sizes.

Figure 6

No systematic influence of preferred spatial wavelength on coupling strength. Dependence of coupling strength on distance. Pairs of recording sites were separated into three groups with short, medium and long preferred spatial wavelength (red, orange and green dots, see text for details). (A) If preferred spatial wavelength were to influence coupling strength, the groups would differ. However, no difference is observed. (B) Coupling strength is plotted with respect to distance normalized to mean preferred spatial wavelength. If linking field size were to scale linearly with preferred spatial wavelength, the groups would not differ. However, the three groups are clearly distinct. Results from the other animal and for different coupling measures are comparable. Taken together these results are clear evidence against a scaling of the linking field with preferred spatial wavelength.

Figure 7

No systematic effect of overlap on coupling strength. Two examples that illustrate the lack of a systematic effect of summation field overlap on signal coupling. (A) Cross-correlation of MUA, monkey K. (B) Cross-correlation of LFP, monkey K. Pairs of recording sites were divided in three groups with small, medium and large residual from the fit of the summation field overlap (red, orange and green dots, see text for details). Signal coupling showed a pronounced dependency on distance of the recording sites. The analysis of residuals revealed a significant negative correlation of overlap and cross-correlation strength of MUA and a positive correlation of overlap and coherence of LFP for monkey K. However, as is evident from the plots, these effects are unsystematic and small. Data from the coherence measure and from the other monkey also fail to reveal a systematic effect of overlap on coupling strength.