Abstract

Collinear proximal flankers can facilitate the detection of a low-contrast target or generate false-alarm target detection in the absence of a target. Although these effects are known to involve subthreshold neuronal interactions beyond the classical receptive field, the underlying neuronal mechanisms are not fully understood. Here, we used voltage-sensitive dye imaging that emphasizes subthreshold population activity, at high spatial and temporal resolution and imaged the visual cortex of fixating monkeys while they were presented with a low-contrast Gabor target, embedded within collinear or orthogonal flankers. We found that neuronal activity at the target site in area primary visual cortex increased and response latency decreased due to spatial spread of activation from the flankers’ site. This increased activity was smaller than expected by a linear summation. The presentation of flankers alone induced strong spatial filling-in at the target site. Importantly, the increased neuronal activity at the target site was synchronized over time, both locally and with neuronal population at the flanker's site. This onset synchronization was higher for collinear than for orthogonal flankers. We further show that synchrony is a superior code over amplitude, for discriminating collinear from orthogonal pattern. These results suggest that population synchrony can serve as a code to discriminate contextual effects.

Introduction

A fundamental ability of the human brain is integration of visual features into coherent objects and contours (Field et al. 1993; Fitzpatrick 2000; Field and Hayes 2004; Kourtzi et al. 2003; Mandon and Kreiter 2005). Contour integration follows the Gestalt law of “good continuation” (Wertheimer 1923), by which discrete contour elements positioned and oriented along a smooth path are readily grouped together, forming a continuous visual contour that is globally salient and “pops out” from its background (Field et al. 1993; Hess et al. 2003; Kourtzi et al. 2003; Kubota 2004).

The smallest fragment of contour that obeys the “good continuation” Gestalt law is a collinear element. When a short line segment or Gabor target is flanked by an additional collinear element, the target is detected at much lower contrasts than when it is presented alone (Polat and Sagi 1993, 1994a, 1994b; Kapadia et al. 1995; Bonneh and Sagi 1998; Gilbert et al. 2000; Solomon and Morgan 2000; Woods et al. 2002; Polat and Sagi 2007). Electrophysiological studies have shown that collinear stimuli can facilitate the neuron's firing rate to low-contrast target presented at the receptive field (RF). This facilitation is reduced when the elements’ separation distance is increased or by changing the orientation of the elements (Polat et al. 1998; Fitzpatrick 2000; Kasamatsu et al. 2001; Mizobe et al. 2001; Bauer and Heinze 2002; Chisum et al. 2003).

Accumulating studies suggest that the long horizontal connections in the primary visual cortex (V1) play a major role in collinear effects and contour integration (Rockland and Lund 1982; Ts'o et al. 1986; Ts'o and Gilbert 1988; Malach et al. 1993; Callaway 1998; Hess and Field 1999; Fitzpatrick 2000; Gilbert et al. 2000; Stettler et al. 2002; Chisum et al. 2003; Field and Hayes 2004). These lateral interactions can be further modulated by either feed-forward or feedback connections from neurons along the visual pathway (Field and Hayes 2004; Li et al. 2006). In addition, horizontal connections were suggested to operate at the subthreshold levels, modifying rather than driving the responses of the affected neurons (Fitzpatrick 2000; Chisum et al. 2003). Therefore, using voltage-sensitive dye imaging (VSDI), which is highly sensitive to synaptic potentials and subthreshold population dynamics, our first goal was to directly measure these influences. Our second goal was to determine the role of early visual areas in collinear effects by using VSDI, currently the best technique for simultaneous high spatial and temporal resolution of neuronal populations in behaving monkeys. In this study, we performed VSDI of the V1 and extrastriate cortex (V2) of 2 monkeys trained on a simple fixation task while they were presented with collinear or orthogonal patterns of Gabors. We found that the flankers can increase neuronal activation and decrease response latency at the target's site similar to the effects induced by increasing the target's contrast. This increased activation was smaller than expected by a linear summation and also dependent on the distance separation between the target and flankers. In addition, the presentation of flankers alone generated activation at the target site, which may account for the high rate of false-alarm target detection reported for this visual stimulus (Polat and Sagi 2007). Importantly, we found that collinear stimuli induced both faster dynamics and higher onset synchronization in area V1 when compared with orthogonal condition. However, onset synchronization was superior in discriminating between collinear and orthogonal conditions.

Materials and Methods

Behavioral Task and Visual Stimuli

Two monkeys (Maccaca Fascicularis, males, 7 and 9 kg) were trained on a passive fixation task. The trial started when a small gray fixation point (0.05–0.1°) appeared on a screen and after a variable interval (3000–4000 ms), a visual stimulus was turned on for 200 ms. We used only a brief presentation of the visual stimulus to avoid possible effects of eye movements. The monkey had to maintain fixation within a small fixation window for an additional interval of 2000 ms. The monkey was rewarded if the trial was successfully completed. Only trials in which the monkey maintained fixation within ±0.5° during the visual stimulus presentation were analyzed.

Visual stimuli were presented on a 21-in. Mitsubishi monitor at 85 Hz, placed 100 cm from the monkey's eyes. To measure retinotopic maps and study the spatial spread of activity induced by local stimuli, we used high-contrast bars: 0.1 × 0.25° white bar on a black screen (Fig. 1A). To study collinear effects, we used an array of 3 Gabor stimuli on an isoluminant screen (Fig. 2A).

Figure 1.

Spatiotemporal patterns of cortical activation to high-contrast bars. (A) Schematic presentation of visual stimuli, from top to bottom panel: The “upper bar” (0.1 × 0.25°; 100% contrast) positioned 1° below the horizontal meridian and 1.5° from the vertical meridian. The “lower-bar”: displaced 1° lower from the upper bar (0.1° × 0.25°; 100% contrast); “Both bars” The yellow dot represents the fixation point. (B) Spatiotemporal activation patterns evoked by the presentation of upper bar, lower-bar or both. A sequence of frames, 10 ms apart, depicts the VSDI onset response over the exposed visual cortex (V1 and V2). Time zero represents the onset of the visual stimulus. Scale bar is 1 mm. (C) Image of the blood vessels and cortical activation maps averaged over 100–200 ms after visual stimulus onset. The contour lines show a 2D Gaussian fit for the averaged activation maps. The curves denote 80%, 50%, and 20% amplitude of the fitted Gaussian. Abbreviations: A: anterior; P: posterior; M: medial; and L: lateral. (D,E) Time course of response and spatial profile over upper- and lower-bar cortical sites. Error bars are ±1SEM (standard error of the mean). (D) Response amplitude at the upper-bar site, following the presentation of the visual stimuli in A, averaged over all pixels within the top 10% of the fitted Gaussian. Visual stimuli: Upper bar: blue line (n = 16); Lower-bar: purple line (n = 13); Both bars: green trace (n = 20). (E) Mean spatial profile crossing through the upper- and lower-bar center of activation site depicted in Figure 1C, right panel. The lower traces represent cortical activation before visual stimulus onset. Abbreviations U: Upper-bar site; L: “Lower-bar site”; Line colors as in D.

Figure 1.

Spatiotemporal patterns of cortical activation to high-contrast bars. (A) Schematic presentation of visual stimuli, from top to bottom panel: The “upper bar” (0.1 × 0.25°; 100% contrast) positioned 1° below the horizontal meridian and 1.5° from the vertical meridian. The “lower-bar”: displaced 1° lower from the upper bar (0.1° × 0.25°; 100% contrast); “Both bars” The yellow dot represents the fixation point. (B) Spatiotemporal activation patterns evoked by the presentation of upper bar, lower-bar or both. A sequence of frames, 10 ms apart, depicts the VSDI onset response over the exposed visual cortex (V1 and V2). Time zero represents the onset of the visual stimulus. Scale bar is 1 mm. (C) Image of the blood vessels and cortical activation maps averaged over 100–200 ms after visual stimulus onset. The contour lines show a 2D Gaussian fit for the averaged activation maps. The curves denote 80%, 50%, and 20% amplitude of the fitted Gaussian. Abbreviations: A: anterior; P: posterior; M: medial; and L: lateral. (D,E) Time course of response and spatial profile over upper- and lower-bar cortical sites. Error bars are ±1SEM (standard error of the mean). (D) Response amplitude at the upper-bar site, following the presentation of the visual stimuli in A, averaged over all pixels within the top 10% of the fitted Gaussian. Visual stimuli: Upper bar: blue line (n = 16); Lower-bar: purple line (n = 13); Both bars: green trace (n = 20). (E) Mean spatial profile crossing through the upper- and lower-bar center of activation site depicted in Figure 1C, right panel. The lower traces represent cortical activation before visual stimulus onset. Abbreviations U: Upper-bar site; L: “Lower-bar site”; Line colors as in D.

Figure 2.

Gabor visual stimuli and flanker effects on neuronal activation at the target site. (A) Illustration of the Gabor stimuli and their parameters that were used in this study, from left to right panel: low-contrast target (16% contrast, “target condition”); target embedded in high-contrast collinear flankers (64% contrast, collinear condition). Separation distance between flanker and target is 0.75° (3λ). High-contrast collinear flankers without target (64% contrast, “collinear flanker-alone condition”). Low-contrast target with high-contrast orthogonal flankers (16% contrast of the target; 64% contrast of the flankers; orthogonal condition). (B) Time course of the VSDI signal at the target site for the different stimuli conditions (Gabor's parameters are as in A). “Target condition”: dotted line (n = 82 trials). “Collinear condition”: solid line (n = 83 trials). “Collinear flankers-alone condition”: no target appears (n = 79 trials): dashed line. The evoked response amplitude was normalized for each single trial by the average maximal amplitude of the collinear condition. Linear summation (LS): gray trace—the summation of evoked activity to the target and flanker-alone condition. Visual stimuli onset at t =0. Data were averaged across 3 recording sessions. (C) Contrast curve. Time course of the VSDI signal to increasing contrast of full-field square moving gratings. The time course was averaged over pixels in V1 and over trials. Increasing the contrast decreased the time to max half amplitude and increased the maximal response amplitude. (D) The difference between the linear summation model and the collinear condition presented in B. Bin size is 30 ms. Error bars are ±1SEM in B–D.

Figure 2.

Gabor visual stimuli and flanker effects on neuronal activation at the target site. (A) Illustration of the Gabor stimuli and their parameters that were used in this study, from left to right panel: low-contrast target (16% contrast, “target condition”); target embedded in high-contrast collinear flankers (64% contrast, collinear condition). Separation distance between flanker and target is 0.75° (3λ). High-contrast collinear flankers without target (64% contrast, “collinear flanker-alone condition”). Low-contrast target with high-contrast orthogonal flankers (16% contrast of the target; 64% contrast of the flankers; orthogonal condition). (B) Time course of the VSDI signal at the target site for the different stimuli conditions (Gabor's parameters are as in A). “Target condition”: dotted line (n = 82 trials). “Collinear condition”: solid line (n = 83 trials). “Collinear flankers-alone condition”: no target appears (n = 79 trials): dashed line. The evoked response amplitude was normalized for each single trial by the average maximal amplitude of the collinear condition. Linear summation (LS): gray trace—the summation of evoked activity to the target and flanker-alone condition. Visual stimuli onset at t =0. Data were averaged across 3 recording sessions. (C) Contrast curve. Time course of the VSDI signal to increasing contrast of full-field square moving gratings. The time course was averaged over pixels in V1 and over trials. Increasing the contrast decreased the time to max half amplitude and increased the maximal response amplitude. (D) The difference between the linear summation model and the collinear condition presented in B. Bin size is 30 ms. Error bars are ±1SEM in B–D.

Gabor's configuration (spatial arrangement and proximity) and parameters (λ, σ, contrast) were set to duplicate human psychophysical studies (Polat and Sagi 1993, 1994a; Woods et al. 2002). In the following section, we elaborate on the visual stimuli. We tagged the central Gabor element as the “target”and the 2 other surrounding Gabors as “flankers.” Note that the term target refers to the spatial arrangement of the Gabor (central location within the array), which is analogous to the target Gabor detected in psychophysical studies. Unlike psychophysical studies, we did not imply a special “behavioral” role for the target, as the monkeys were passively fixating.

We used the following Gabor parameters: Contrast-16% (target) or 64% (flankers); Orientation (θ)-90° for the target. The flankers’ orientation was 90° (collinear condition) or 0o (orthogonal condition), wavelength (λ)-0.25°; σ-0.125°. In most recording sessions, the distance between the target and flankers was 0.75° (3λ) and the number of flankers was 1 or 3 from each side of the target. Visual stimuli conditions (Fig. 2A) were as follows: 1) The “target condition”: a single low contrast (16%) Gabor. 2) The “collinear condition”: a low-contrast Gabor with high-contrast (64%) collinear flankers. 3) The “collinear flankers-alone condition”: collinear flankers (without any target). 4) The “orthogonal condition”: a low-contrast Gabor with high-contrast orthogonal flankers. 5) The “orthogonal flankers-alone condition”: orthogonal flankers (without any target). 6) The blank condition: only the fixation point appeared without any visual stimulus; this condition was used to remove the heartbeat artifact (see Basic analysis of VSDI signals).

The retinal eccentricity of the central Gabor (e.g., the target) was varied between different recording sessions (to verify the results over different retinotopic locations): 0.9–2° below the horizontal meridian and 0.5–1° from the vertical meridian. Psychophysical studies showed that target detection was facilitated within these eccentricities (Shani and Sagi 2005).

Eye position was monitored by an infrared eye tracker (Dr Bouis Device, Kalsruhe, Germany), sampled at 1 kHz and recorded at 250 Hz. Two linked computers control the visual stimulation, data acquisition, and monkey's behavior (CORTEX software package). The system is equipped with a PCI-DAS 1602/12 card to control the behavioral task and data acquisition. The protocol of data acquisition in VSDI has been described in detail elsewhere (Shtoyerman et al. 2000; Slovin et al. 2002). Single trials were saved on separate data files to enable single-trial analysis.

VSDI Imaging

The surgical procedure and voltage-sensitive dye (VSD) staining have been reported in detail elsewhere (Grinvald et al. 1999; Shtoyerman et al. 2000; Arieli et al. 2002; Slovin et al. 2002). All experimental procedures were according to the NIH guidelines, approved by the Animal Care and Use Guidelines Committee of Bar-Ilan University, and supervised by the Israeli authorities for animal experiments. Antibiotics and analgesics were applied before and after surgical procedures and adequate measurements were taken to minimize pain or discomfort. Briefly, craniotomy was performed under full anesthetization and aseptic conditions, the dura mater was removed, exposing the visual cortex. To perform long-term VSDI, a transparent artificial dura was implanted over the cortical surface. The anterior border of the imaged area is typically 3–6 mm anterior to the Lunate sulcus. The center of the imaged V1 area is 1–3° below the horizontal meridian and 1–2° from the vertical meridian. We stained the cortex with VSDs: RH-1691 or RH-1838 supplied by Optical Imaging, Israel.

VSDI was performed using the Micam Ultima system based on a sensitive fast camera, which offers a resolution of 104 (100 × 100) pixels and up to a 10-kHz sampling rate. For the current study, we used sampling rate of 10 ms/frame, spatial resolution of 10 000 pixels where each pixel sums the activity from an area of 170 μm × 170 μm occupying few hundreds of neurons. The exposed cortex is illuminated using an epi-illumination box with an appropriate excitation filter (peak transmission 630 nm, width at half height 10 nm) and a dichroic mirror (DRLP 650), both from Omega Optical, Brattleboro, VT. To collect the fluorescence and reject stray excitation light, we placed a barrier postfilter above the dichronic mirror (RG 665, Schott, Mainz, Germany).

Analysis of the Imaging Data and Statistical Analysis

Data analysis was done over a total of 16 recording sessions in 2 hemispheres of 2 adult monkeys. In each recording session, we analyzed only correct trials that were carefully checked for any eye movements. Only trials with tight fixation were chosen for further analysis, and trials from different conditions were analyzed separately. All statistical analyses and calculations were done using Matlab software (Ver. 2006b, The MathWorks, Inc.).

Basic Analysis of VSDI Signals

The basic analysis consists of several steps: 1) To avoid high shot-noise level and obtain reliable signal to noise ratio of the VSDI signal, we set a fluorescence threshold level for each pixel. The specific threshold value depends on the camera well capacity, dark noise level, and staining quality. In our experimental setup, we found that pixels at 15% of maximal illumination level are informative for high-quality analysis. 2) To correct for the nonhomogeneous illumination pattern and because the optical response is proportional to the illumination level, all pixels are normalized to their DC level, which is the average fluorescence level of a given pixel over the first few frames, before stimulus onset. 3) To remove slow drifts of the imaged data (the photo bleaching effect), a linear trend is adjusted to each of the pixels’ signals and then subtracted. 4) Finally, to remove the heartbeat effect, the average blank condition is subtracted from each stimulus-evoked condition. The above procedures eliminate most of the noise resulting from heartbeat, respiration, and fixation point effects. Using these steps, we can calculate the single-condition maps (e.g., the visually stimulated condition divided by the blank condition), which represent the retinotopic activation in the visual areas (Fig. 1B).

VSDI Signal Analysis at the Target Site

To study the VSDI signal at the target site, we fitted a 2D Gaussian to the mean VSDI activation map in area V1 evoked by the target condition or by the bar visual stimuli (Fig. 1C). In general, elliptical 2D-Gaussian can be expressed with the following equation (1): 

(1)
graphic
a, b, and c are defined as 
graphic
where x,y are the Gaussian matrix indices denoting the spatial coordinates of the VSDI activation pixels. A is the maximal Gaussian amplitude (also the maximal amplitude of the activation patch). Cx, Cy are the indices of the Gaussian center; σx, σy are the Gaussian widths on x and y dimensions. θ is the angle of the Gaussian. We used automatic algorithm to fit 2D Gaussian to the VSDI signal at the cortical sites.

The time course of the evoked VSDI response at the target site was calculated by averaging over all pixels within the top 10% of the fitted Gaussian (region of interest, ROI target). We compared the time course of the VSDI signal at the ROI target for the different visual conditions: target, flankers alone, collinear or orthogonal condition (Fig. 2B). Results were verified also for pixels within the top 15% and 20% of the fitted Gaussian. A similar analysis was used for the bar visual stimuli (Fig. 1D). In order to estimate latencies, curves were smoothed and time to half of the maximal amplitude was calculated. To study the spatial spread over V1 and between the different retinotopic sites of the target and flanker, we measured a spatial profile passing through the target and the flankers’ activation centers (Fig. S2). A similar analysis was used for the bar visual stimuli (Fig. 1E). Finally, to study the target increased activation as a function of the flanker separation distance, we varied the distance between flankers and target (Fig. 3).

Figure 3.

Effect of flankers’ separation distance. (A,B) Effect of flankers’ distance on target increased activation. (A) Mean amplitude response averaged over pixels within the top 10% of the fitted Gaussian (ROI target) as a function of flanker distance. Visual stimuli included 100%-contrast target and 16%-contrast target for comparison with the increased activation effect of the flankers. Flankers’ distance varied from 0.75°, 1.25°, and 1.75° (flanker's center to target's center). (B) Maximal response amplitude for visual stimuli in A. Error bars are ±1SEM in A,B.

Figure 3.

Effect of flankers’ separation distance. (A,B) Effect of flankers’ distance on target increased activation. (A) Mean amplitude response averaged over pixels within the top 10% of the fitted Gaussian (ROI target) as a function of flanker distance. Visual stimuli included 100%-contrast target and 16%-contrast target for comparison with the increased activation effect of the flankers. Flankers’ distance varied from 0.75°, 1.25°, and 1.75° (flanker's center to target's center). (B) Maximal response amplitude for visual stimuli in A. Error bars are ±1SEM in A,B.

VSDI Signal Analysis for the Collinear and Orthogonal Conditions: Amplitude and Onset Synchrony

Cortical responses to collinear and orthogonal conditions were compared using single-condition maps and differential maps that were calculated by subtracting the orthogonal condition maps from the collinear condition maps. To study synchronization of neuronal population activity in the time domain, we calculated the mean onset synchronization maps using the Pearson correlation coefficient (CC). Specifically, we used equations (24) to calculate the mean onset synchronization maps: 

(2)
graphic
Equation (2) depicts the CC matrix, CC(x,y), between pixel i and the entire pixels in the map. The i index refers to a pixel located in the ROI target. The j index refers to all pixels in the VSDI map (9999 pixels), excluding pixel i. Cov stands for covariance and CC is the zero lag of the normalized covariance function. The CC was calculated over the rising phase of the evoked response (0–150 ms after visual stimulus onset), within a single trial. The result of equation (2) is a single matrix of correlation between pixel i and the rest of the pixels in the map. 
(3)
graphic
Our next step was to average the CC(i,j) for all pixels in the ROI target. Equation (3) shows single trial CC matrix, which was calculated by averaging over all CC matrices of the pixels at the ROI target (n). 
(4)
graphic
The final step was to average CCtrial over all trials (eq. 4). The resulting correlation map (CCmap) was defined as the mean onset synchronization map or averaged CC maps (Fig. 4D).

Figure 4.

Collinear versus orthogonal: Onset synchronization is more informative than maximal amplitude. (A–C) Maximal amplitude maps, time course, and pixel amplitude distribution in collinear and orthogonal condition. (A) Single-condition maps averaged over 150–200 ms from stimulus onset for the low-contrast target, collinear and orthogonal condition. The maps show patches of activation in area V1 (bottom part of the maps): target (left patch) and flanker (right patch) in the collinear and orthogonal condition. The upper elongated patch shows the activation in V2. Scale bar is 1 mm. The central Gabor was located 0.9° below the horizontal meridian and 1° from the vertical meridian. (B) Time course of activation following visual stimulus onset, calculated by averaging over all pixels within the top 10% of the fitted Gaussian (ROI target). Error bars are ±1SEM. (C) Distribution of the average maximal amplitude over pixels in ROI target. (D) Mean spatial correlation maps of all pixels within the top 10% of the fitted Gaussian (ROI target). The correlation was calculated for 0–150 ms after visual stimulus onset. Note the correlation patch of the target (left white patch) and the correlation patch of the flanker (right white patch) in the collinear conditions (middle panel). The same patches have lower correlation values in the orthogonal condition (right panel). Scale bar as in A. (E) Distribution of the correlation values for the pixels chosen in B. (F) Sliding window of correlation (70-ms time window), calculated over pixels chosen the same in B. (G) Average single-condition map obtained at 150 ms after the onset of the target. The target site is denoted by a white square. (H) Upper panel: Sequence of single-condition maps for the area depicted in G, 10 ms apart, following the presentation of the target condition. Lower panel: Differential maps: the orthogonal condition was subtracted from the collinear conditions for the target site, as depicted in G. Each frame is 10 ms apart. Scale bar is 1mm.

Figure 4.

Collinear versus orthogonal: Onset synchronization is more informative than maximal amplitude. (A–C) Maximal amplitude maps, time course, and pixel amplitude distribution in collinear and orthogonal condition. (A) Single-condition maps averaged over 150–200 ms from stimulus onset for the low-contrast target, collinear and orthogonal condition. The maps show patches of activation in area V1 (bottom part of the maps): target (left patch) and flanker (right patch) in the collinear and orthogonal condition. The upper elongated patch shows the activation in V2. Scale bar is 1 mm. The central Gabor was located 0.9° below the horizontal meridian and 1° from the vertical meridian. (B) Time course of activation following visual stimulus onset, calculated by averaging over all pixels within the top 10% of the fitted Gaussian (ROI target). Error bars are ±1SEM. (C) Distribution of the average maximal amplitude over pixels in ROI target. (D) Mean spatial correlation maps of all pixels within the top 10% of the fitted Gaussian (ROI target). The correlation was calculated for 0–150 ms after visual stimulus onset. Note the correlation patch of the target (left white patch) and the correlation patch of the flanker (right white patch) in the collinear conditions (middle panel). The same patches have lower correlation values in the orthogonal condition (right panel). Scale bar as in A. (E) Distribution of the correlation values for the pixels chosen in B. (F) Sliding window of correlation (70-ms time window), calculated over pixels chosen the same in B. (G) Average single-condition map obtained at 150 ms after the onset of the target. The target site is denoted by a white square. (H) Upper panel: Sequence of single-condition maps for the area depicted in G, 10 ms apart, following the presentation of the target condition. Lower panel: Differential maps: the orthogonal condition was subtracted from the collinear conditions for the target site, as depicted in G. Each frame is 10 ms apart. Scale bar is 1mm.

To quantify the difference between the collinear and orthogonal VSDI response (Fig. 6A), we studied 3 neuronal parameters that were measured over pixels at the ROI target: 1) “Maximal amplitude,” 2) “Maximal separation amplitude.” and 3) “Optimal correlation.” The time of peak VSDI amplitude was found (TMax) and averaged within a time window of TMax ±10 ms to obtain the “maximal amplitude value.” The “maximal separation amplitude” was defined as the amplitude of the VSDI signal (averaged within a time window of ±10 ms) at the time of maximal first derivative of the VSDI signal measured for the target condition. To calculate the optimal correlation, we studied onset synchronization as function of time by calculating the CCmap over a 0- to 200-ms period using a sliding window width of 70 ms (Fig. 5B). The highest correlation for each condition was defined as the “optimal correlation.”

Figure 5.

Average time course of response and correlation for the collinear and orthogonal conditions. Average of all recording sessions for the collinear and orthogonal conditions. (A) Mean time course of response averaged over pixels within top 10% of the Gaussian at each session separately and over trials in 8 recording sessions. The evoked response amplitude was normalized at single trial level to the maximal amplitude of the collinear response. Error bars are ±1SEM. (B) Mean sliding window correlation (70-ms time window), calculated over pixels chosen the same as in A, averaged over 8 recording sessions.

Figure 5.

Average time course of response and correlation for the collinear and orthogonal conditions. Average of all recording sessions for the collinear and orthogonal conditions. (A) Mean time course of response averaged over pixels within top 10% of the Gaussian at each session separately and over trials in 8 recording sessions. The evoked response amplitude was normalized at single trial level to the maximal amplitude of the collinear response. Error bars are ±1SEM. (B) Mean sliding window correlation (70-ms time window), calculated over pixels chosen the same as in A, averaged over 8 recording sessions.

Figure 6.

Onset synchronization is superior to amplitude in discriminating the collinear from the orthogonal condition. (A) Distributions of the pixels’ value for the collinear and orthogonal conditions, from a single recording session. Left panel: the optimal correlation calculated in the 70-ms sliding window (see Fig. 5B). Middle panel: maximal separation amplitude. Right panel: maximal amplitude. (B) ROC curve calculated for the collinear and orthogonal neurometers pixel value distributions averaged over all recording sessions.

Figure 6.

Onset synchronization is superior to amplitude in discriminating the collinear from the orthogonal condition. (A) Distributions of the pixels’ value for the collinear and orthogonal conditions, from a single recording session. Left panel: the optimal correlation calculated in the 70-ms sliding window (see Fig. 5B). Middle panel: maximal separation amplitude. Right panel: maximal amplitude. (B) ROC curve calculated for the collinear and orthogonal neurometers pixel value distributions averaged over all recording sessions.

Finally, to discriminate between the collinear and orthogonal conditions, we performed Receiver Operating Characteristic (ROC) over the ROI-target pixel value distribution for the different parameters: “maximal amplitude,” maximal separation amplitude, and “optimal correlation.” We also calculated the statistic d′, which is the difference between the 2 means divided by the square root of the distribution mean variance difference (Fig. 6).

In addition, to subtract the contribution of the visual stimulus onset, we used 2 approaches: We subtracted the shuffled correlation from the raw correlation or subtracted the averaged visual stimulus response from the VSDI response for each trial and pixel and then calculated the correlation. We defined these correlations as visual stimulus subtracted correlation.

Results

Spatial Spread of Neuronal Activation and Retinotopic Mapping Induced by Local Stimuli

Our first step was to obtain retinotopic maps in the V1, at high signal-to-noise ratio, using small high-contrast visual stimuli. Specifically, we wanted to study the spatial spread of neural activation between 2 adjacent cortical sites, evoked by the local stimuli. Therefore, the monkeys were presented with local high-contrast (100%) visual stimuli: an upper bar (size: 0.25° × 0.1°; position: 1° below the horizontal meridian and 1.5° from the vertical meridian) or a lower-bar displaced 1° below the upper bar (size: 0.25° × 0.1°; position: 2° below the horizontal meridian and 1.5° from the vertical meridian) or both (Fig. 1A). Using VSDI, we directly measured the spatiotemporal activation patterns evoked by these visual stimuli. As expected, the activation maps included 2 separate patches in V1 that correspond to the cortical retinotopic sites of the upper and lower-bar stimuli (Fig. 1B).

We found that shortly after upper- or lower-bar stimulus onset, VSDI signal increased at the upper-bar or lower-bar cortical sites, respectively, reflecting an increase in neuronal population activity (Fig. 1B). Within the next time frames, the local activation in V1 (induced by either bars) spread horizontally over the cortical surface for several mm, far beyond that expected by the magnification factor for this eccentricity (Grinvald et al. 1994; Slovin et al. 2002). This large spread is characteristic to the VSDI signal in vivo, which reflects population activity, emphasizing subthreshold synaptic potentials (Petersen and Sakmann 2001; Petersen et al. 2003; for review see Grinvald et al. 1999). We previously showed that the spreading velocity of such activation was within the range of propagation via horizontal connections (Slovin et al. 2002). When both bars were presented, 2 distinct patches of activation appeared over the imaged cortical surface (Fig. 1B), indicating, that we can measure and characterize the spatial spread of neuronal activation between 2 distinct retinotopic sites. We further used the obtained retinotoic maps of the bars visual stimuli to localize the Gabor target and its proximal flankers in the visual field.

To calculate the evoked time course of neuronal activity and its spatial extent, we fitted a 2D Gaussian to the activation patches on each site (Fig. 1C and see Materials and Methods). The time course of the response was calculated by averaging the VSDI signal over all pixels included within the top 10% of the fitted Gaussian. As expected, we found that presentation of the upper bar generated a short delay and high response activation “at the upper-bar site” (Fig. 1D, blue curve). However, presentation of the vertically displaced, high-contrast lower bar induced delayed activation with a much lower amplitude response “at the upper-bar site” (Fig. 1D, purple curve). This result suggests that neuronal activation induced at the lower-bar site, following lower-bar presentation, spread horizontally over cortical surface, arriving at the upper-bar site, where it induced low-amplitude delayed activation (time to half-maximal amplitude: upper-bar stimulus = 48.7 ms; lower-bar stimulus = 72.3 ms; upper and lower bar = 47.5 ms). Next, we calculated the spatial activation profile for the upper- and lower-bar conditions along a transaction line passing through the centers of the fitted Gaussians, at the time of maximal response amplitude (Fig. 1C,E). The spatial profile shows 2 distinct hills of activation, spanning about 12–13 mm for the upper-bar and lower-bar conditions. The spatial effect of the lower-bar at the upper-bar site is evident from the partial overlap between the activations profiles.

We hypothesized that the large spatial spread of activation between adjacent distinct sites (the 2 bar sites and subsequently the 2 Gabor sites corresponding to the target and flanker) is a fundamental feature of the cortical activity network. We further suggest that this feature plays an important role in collinear and filling-in effects.

Increased Neuronal Activity at the Low-Contrast Gabor Target Site When Adding High-Contrast Flankers

Our next objective was to measure the spatiotemporal patterns of neuronal population activity evoked by a low-contrast Gabor target and how this activation was influenced on adding high-contrast Gabor flankers. Previous psychophysical studies demonstrated facilitation of target detection, namely, decreased threshold detection of a low-contrast target when it was embedded within high-contrast flankers (Polat and Sagi 1993, 1994a, 1994b, 2007). Here, we used a set of visual stimuli similar to these studies (see Materials and Methods). Figure 2A (left to right panel) depicts the visual stimuli conditions. 1) The “target condition”: a low-contrast (16%) Gabor visual stimulus. 2) The collinear condition: a low-contrast Gabor with high-contrast (64%) collinear flankers. 3) The “collinear flankers-alone condition”: collinear flankers (without any target). 4) The “orthogonal condition”: a low-contrast Gabor with high-contrast orthogonal flankers and 5) The “orthogonal flankers-alone condition” (not shown in Fig. 2A) orthogonal flankers (without any target).

Presentation of a small low-contrast Gabor patch (target condition) induced a long latency and low-amplitude VSDI signal at the target retinotopic site in area V1, as expected. Figure 2B shows the normalized population response to the target condition, averaged over 3 recording sessions (population response was normalized to the maximal VSDI signal in the collinear condition; see also figure legend). The time to half-maximal amplitude was about 115 ms and reached a relatively low-amplitude level (40%). Importantly, adding high-contrast flankers affected neuronal activity at the target site as evident by 2 aspects: 1) The maximal amplitude response at the target site was more than doubled relative to the target condition and reached 85%. 2) The time to half-maximal amplitude at the target site decreased by 30% to about 85ms (time to half-maximal amplitude: target condition = 115.2 ms; collinear condition = 85.12 ms; and flanker condition = 92.9 ms). Similar results were obtained for all recording sessions as shown in Table 1.

Table 1

Increased activation of VSDI signal at the target site when adding flankers and comparison to target at 100% contrast

Recording session Maximal amplitude (ΔF/F, ×10−4)
 
Time to half-max amp (ms)
 
Target 16% contrast Target and flankers Target 100% contrast Target 16% contrast Target and flankers Target 100% contrast 
5.81 9.08 — 145.37 116.67 — 
9.07 16.11 — 118.42 82.22 — 
6.45 12.19 — 162.73 126.52 — 
2.43 10.06 13.51 141.82 119.17 81.11 
5.16 10.38 10.8 163.13 126 107.5 
8.29 12.57 16.57 140.8 115.77 73.81 
5.25 8.66 10.93 135.29 109.26 80 
10.55 16.29 20.66 135 88.64 75.88 
Mean ± SEM 6.62 ± 0.77 11.92 ± 1.12 14.49 ± 2.29 142.82 ± 6.76 110.53 ± 6.15 83.66 ± 7.65 
Recording session Maximal amplitude (ΔF/F, ×10−4)
 
Time to half-max amp (ms)
 
Target 16% contrast Target and flankers Target 100% contrast Target 16% contrast Target and flankers Target 100% contrast 
5.81 9.08 — 145.37 116.67 — 
9.07 16.11 — 118.42 82.22 — 
6.45 12.19 — 162.73 126.52 — 
2.43 10.06 13.51 141.82 119.17 81.11 
5.16 10.38 10.8 163.13 126 107.5 
8.29 12.57 16.57 140.8 115.77 73.81 
5.25 8.66 10.93 135.29 109.26 80 
10.55 16.29 20.66 135 88.64 75.88 
Mean ± SEM 6.62 ± 0.77 11.92 ± 1.12 14.49 ± 2.29 142.82 ± 6.76 110.53 ± 6.15 83.66 ± 7.65 

Note: Adding flankers increased response amplitude and decreased the time to half-maximal amplitude, at the target site. The P value was calculated using Student's paired t-test between two out of the three conditions for each neuronal parameter. Maximum amplitude: Ptarget 16%–target and flankers = 3.1 × 10−5, Ptarget and flankers–target 100% = 0.015 and Ptarget 16%–target 100% = 0.0019. Time to half of maximal amplitude: Ptarget 16%–target and flankers = 0.0072, Ptarget 16%–target 100%, = 9.6 × 10−6, Ptarget 16%–target 100% = 9.6 × 10−6.

Interestingly, the increased neuronal activity observed in the collinear condition is similar to the neuronal correlates of increasing stimulus contrast. Figure 2C shows that VSDI activation increases and latency decreases when increasing the contrast of full field visual stimuli. In addition, Table 1 shows that increasing the contrast level of a single Gabor patch from 16% to 100%, resulted in a higher amplitude and decreased response latency (measured as time to half-maximal amplitude). Taken together, these results suggest that adding high-contrast flankers increase neuronal population response at the target site in a similar manner to increasing the target contrast, that is, increased population activity and decreased latency. One interpretation is that these 2 neuronal parameters may underlie the facilitation of target detection observed in psychophysical experiments in human subjects.

Presentation of high-contrast flankers alone (without the target) induced a VSDI response that spread over the cortical surface from the flanker sites, arriving at the target site (Fig. 2B, dashed line). In the presence of the flankers alone, the activation at the target site increased dramatically from 40% to 73% and the time to half-max amplitude decreased from 115 to 93 ms (Fig. 2B). We defined this activation at the target site, as “filling-in” effect in the absence of a target. The filling-in at the target site might be an important mechanism to link separate small local visual stimuli or to generate a phantom perception of nonexisting visual stimuli. Interestingly, the amplitude of response at the target site was higher for the high-contrast flankers-alone condition than for the low-contrast target condition (Fig. 2B, dotted line). This result suggests that in the flankers-alone condition, the subjects may detect a “phantom target” and therefore generate false-alarm behavioral responses. Importantly, a recent paper by Polat and Sagi (2007) reported that human subjects had a high rate of false-alarm responses when presented with the flankers-alone condition. Our results further confirm their observations. Finally, increased neuronal activation at the target site occurred also for the orthogonal flankers-alone condition, although this effect was smaller than the collinear flankers-alone condition (Supplementary material, Fig. S1).

The Increased Activity at the Target Site Is Smaller Than a Linear Model

Next, we tested whether we could fit a linear model to the increased cortical activation at the target site, when presenting the target with flankers (Fig. 2B,D). We calculated the linear model by summing the evoked VSDI response to the target condition and to the collinear flankers-alone condition and compared it with the activation evoked by the collinear condition (target with collinear flankers.) The VSDI response at the target site was nearly equal to the linear summation during the initial response rise time, however, at later times, it exhibited a marked “suppression” relative to the linear model (Fig. 2B,D). A spatial profile along the retinotopic axis of the target and flanker sites revealed that the suppression effect was evident mainly at the target site, whereas at the flanker site, the evoked activity in the collinear condition fitted well the linear model (supplementary material, Fig. S2 A,B). We observed similar results in all recording sessions; in particular, the evoked VSDI response to the collinear condition was always lower than that expected by the linear summation.

These results are different from previous publications demonstrating supralinear effects of spiking activity within the RF located at the target site. This difference may arise from several reasons: characteristics of the VSDI signal and our experimental approach. Supra-additive effects were previously reported using extracellular recording of spiking activity (Kapadia et al. 1995; Polat et al. 1998; Ito and Gilbert 1999; Mizobe et al. 2001; for review see Gilbert et al. 2000). Some of these studies adjusted the flanker distance from the RF located at the target site to ensure that the flankers will not induce spiking activity at the target site. In this case, the flankers’ effect on the RF at the target site was always set to zero in terms of spiking activity but not in terms of subthreshold activity, which was not measured directly in these studies. Unlike extracellular recording of spiking activity, the VSDI technique is highly sensitive to subthreshold population activity (Petersen and Sakmann 2001; Petersen et al. 2003; for a review see Grinvald et al. 1999) and therefore could easily detect subthreshold spreading of activation from the flankers to the target site. Another possible explanation is related to the noncentered RF approach we used. In this case, many cells for which the target is not centered in their RF contribute to VSDI signal. In a previous work of Jancke et al. (1999), a similar approach was reported to induce sublinearity effects. Finally, the target contrast was 16%, previously shown to induce suppressive responses and therefore cause sublinearity effects (Polat et al. 1998).

Increasing the Flankers’ Distance Decreases the Increased Activation at the Target Site

To determine whether the effects of the flankers on the target depend on the spatial extent between the flankers and the target as reported in psychophysical studies, we varied the distance between the flankers and target (0.75°, 1.25°, and 1.75°) and measured the neuronal activity at the target site (Fig. 3A,B). We found that the increased activation at the target site was maximal at small distances of 0.75° (3λ) and reduced monotonically when the distance between the flankers and the target was increased. At the eccentricity of our imaging area, we found that already at a distance of 1.75° (and for the flankers’ contrast level we used) the increased activation at the target almost disappeared.

To test whether the filling-in effect was depended on the distance between the flankers, we varied the distance between the flankers-alone and the target position (0.75°, 1.25°, and 1.75°; note that the target did not appear) and measured the activation at the target site (Fig. S3 A,B). Here again we found that the increased activation at the target site was maximal at small distance of 0.75° (3λ) and reduced monotonically when the distance between the flankers and the target was increased. In summary, the increased activity at the target site decreased with increasing the separation distance between the flankers for both the collinear and the flankers-alone conditions. The activation decrease can be correlated to psychophysical studies showing that the probability of low-contrast target detection and false-alarm target detection decreased when increasing the flanker separation distance (Polat and Sagi 2007). When taking this into consideration, a possible interpretation of our results is that response amplitude at the target site is correlated with facilitated low-contrast target detection and false-alarm target detection.

Collinear versus Orthogonal Condition: Response Amplitude and Dynamics

Our next goal was to study the effects of the flankers’ orientation on neuronal response increased activation at the target site. To this end, we compared the VSDI signal of the collinear and orthogonal condition (e.g., the noncollinear condition; Fig. 2A). We measured the response amplitude at the target site by averaging over all pixels within the top 10% of the fitted Gaussian and found that it was increased (relative to the target condition) for both collinear and orthogonal conditions, reaching similar maximal amplitude values (Fig. 4A,B). To quantify this result, we calculated the distribution of the pixel maximal amplitude at the target site for each condition separately and found that the 2 distributions largely overlapped (Fig. 4C). Table 2 summarizes the maximal amplitude values measured for all recording sessions and shows that there was a small but no significant difference between the collinear and orthogonal condition. Finally, the averaged time course over all recording sessions of VSDI response for the collinear and orthogonal conditions is depicted in Figure 5A.

Table 2

Collinear and orthogonal conditions: correlation and amplitude

Recording session Target optimal correlation (CC)
 
Target maximal amplitude (×10−4)
 
Target maximal separation amplitude (×10-4)
 
Target flanker optimal correlation (CC)
 
Collinear Orthogonal Collinear Orthogonal Collinear Orthogonal Collinear Orthogonal 
0.37 0.26 9.08 9.39 5.58 3.92 0.31 0.22 
0.66 0.53 16.11 15.12 14.1 9.97 0.58 0.47 
0.31 0.25 12.19 11.38 6.91 4.51 0.28 0.19 
0.31 0.26 10.06 9.91 6.89 3.5 0.26 0.11 
0.29 0.17 10.38 9.39 6.74 3.98 0.23 0.11 
0.56 0.44 12.57 12.8 9.91 6.57 0.41 0.28 
0.26 0.13 8.66 8.01 6.83 4.89 0.26 0.15 
0.60 0.48 16.29 16.23 14.63 11.98 0.43 0.33 
Mean ± SEM 0.42 ± 0.06 0.26 ± 0.06 11.92 ± 1.08 11.53 ± 1.04 8.95 ± 1.34 6.16 ± 1.19 0.35 ± 0.05 0.23 ± 0.05 
P value 2.9 × 10−5 0.079 2.6 × 10−5 2.6 × 10−7 
Recording session Target optimal correlation (CC)
 
Target maximal amplitude (×10−4)
 
Target maximal separation amplitude (×10-4)
 
Target flanker optimal correlation (CC)
 
Collinear Orthogonal Collinear Orthogonal Collinear Orthogonal Collinear Orthogonal 
0.37 0.26 9.08 9.39 5.58 3.92 0.31 0.22 
0.66 0.53 16.11 15.12 14.1 9.97 0.58 0.47 
0.31 0.25 12.19 11.38 6.91 4.51 0.28 0.19 
0.31 0.26 10.06 9.91 6.89 3.5 0.26 0.11 
0.29 0.17 10.38 9.39 6.74 3.98 0.23 0.11 
0.56 0.44 12.57 12.8 9.91 6.57 0.41 0.28 
0.26 0.13 8.66 8.01 6.83 4.89 0.26 0.15 
0.60 0.48 16.29 16.23 14.63 11.98 0.43 0.33 
Mean ± SEM 0.42 ± 0.06 0.26 ± 0.06 11.92 ± 1.08 11.53 ± 1.04 8.95 ± 1.34 6.16 ± 1.19 0.35 ± 0.05 0.23 ± 0.05 
P value 2.9 × 10−5 0.079 2.6 × 10−5 2.6 × 10−7 

Note: The P value was calculated using Student's paired t-test.

Although the maximal amplitude was comparable between the 2 conditions, the activation dynamics was different: The VSDI response increased faster for the collinear condition than for the orthogonal condition (Fig. 4B is an example for one recording session; Fig. 5Aaveraged over all recording sessions; time to half-maximal amplitude: target = 117 ms; collinear = 86.32 ms; and orthogonal = 102.6 ms). To verify that this effect was restricted to the target cortical site and does not reflect some general nonlocalized effect in the entire imaged area, we calculated the differential map by subtracting the orthogonal maps from the collinear maps. The resulting “differential maps” are depicted in Figure 4H and show that neuronal activation at the target site emerged earlier in the collinear condition when compared with the orthogonal condition.

Although the maximal response amplitude measured at the target site showed a small but nonsignificant difference between the collinear and orthogonal conditions, their different activation dynamics suggested that response amplitude at earlier times can be used to discriminate between these 2 conditions. Therefore, we calculated the maximal separation amplitude (see Materials and Methods) for the collinear and orthogonal conditions. We found a significant difference: the collinear amplitude (8.95 × 10−4) was higher than the orthogonal amplitude (6.16 × 10−4, P < 5 × 10−5, Student's paired t-test and see also Table 2).

Collinear versus Orthogonal Condition: Onset Synchronization Can Discriminate Better Than Amplitude

The faster VSDI rise time for the collinear condition at the target site may imply that neuronal activity at the target site is more synchronized for the collinear condition. Therefore, to study whether onset synchronization can discriminate between the collinear and orthogonal condition, we calculated the average spatial correlation map for the neuronal population within the target site. In particular, we calculated the Pearson CC of the evoked VSDI response amplitude (0–150 ms after visual stimulus onset) between pixels at the target site and the rest of the pixels in the entire imaged area (see Materials and Methods). The obtained CC maps measure the onset synchronization between the pixels located at the target site and the rest of the imaged area. Therefore, pixels with high CC values indicate high onset synchronization to the target's pixels whereas pixels with low (around zero) CC values indicate low onset synchronization to the target's pixels.

Figure 4D shows the spatial CC maps for three conditions in a single recording session: the target, collinear and orthogonal conditions. The left panel in Figure 4D shows that neuronal onset synchronization at the target site has relatively small CC values for the low-contrast target condition (mean CC = 0.29), suggesting low onset synchronization level at the target site for the target condition. When adding high-contrast flankers, the onset synchronization at the target increased dramatically. Additional patches of high correlation appeared at the flanker site and in area V2. Importantly, the CC maps revealed a clear difference between the collinear and orthogonal condition: Onset synchronization values at the target site were higher for the collinear condition (Fig. 4D). To quantify this, we compared the pixels CC values’ distribution of the collinear and orthogonal conditions, at the target site (Fig. 4E; Table 2). The CC histograms showed a clear separation between the 2 conditions: The collinear condition had higher correlation values (CC = 0.77) than the orthogonal condition (CC = 0.63). To study the temporal details of the onset synchronization, we calculated the CC as a function of time, using a sliding window width of 70 ms. We found that the CC was highest at the initial phase of the response activation (Fig. 4Ean example from a single recording session; Fig. 5B—averaged over all recording sessions) and was higher for the collinear condition as compared with the orthogonal condition. We defined the maximal CC value as the optimal correlation and found that it was higher for the collinear condition across all recording sessions (Table 2 and Fig. 5). On average, the CC for the collinear condition was higher by 62% (CC = 0.42, n = 8) than the orthogonal condition (CC = 0.26, n = 8; P < 10−5). In contrast, the average maximal amplitude of the collinear condition was only slightly higher (by 3.4%, amplitude = 11.92 × 10−4) but not significantly, compared with the orthogonal condition (amplitude = 11.53, P = 0.079 and see Table 2).

To determine which of the 3 neuronal parameters (i.e., neurometers): maximal amplitude, maximal separation amplitude, and optimal correlation can discriminate better between the 2 conditions, we calculated d′ and performed ROC analysis over the pixel value distribution for these neurometers (Fig. 6). We found that onset synchronization was the best neurometer for discriminating (Fig. 6B) between the collinear and orthogonal condition (averaged d′ = 2.7), whereas the maximal separation amplitude had a lower d′ value (1.7) and the maximal amplitude had a very low d′ value (0.25).

Onset Synchronization between Pixels at the Target and Flanker

The increased onset synchronization among neuronal populations at the target site for the collinear condition suggests that onset synchronization may also be involved in linking the target and a collinear flanker to create a continuous line, as suggested by the good continuation Gestalt law. The spatial correlation maps show high CC values between the target's pixels and flanker's pixels (Fig. 4D, right white patch) for the collinear condition and lower values for the orthogonal condition. Importantly, we found that the average CC value was significantly higher for the collinear (CC = 0.35; P = 7.1×10−7 Student's paired t-test) than for the orthogonal condition (CC = 0.23) and see also Table 2.

Discussion

We used VSDI to image the V1 of passively fixating monkeys while presented with Gabor target and flankers at different configurations. Enhanced activity in terms of increased VSDI amplitude and decreased response latency at the target retinotopic site was observed when the low-contrast Gabor target was presented with high-contrast flankers (collinear or orthogonal flankers). The increased activation effect was sublinear and its time course depended on the target flanker separation distance. Presentation of high-contrast flankers alone (in the absence of a target) resulted in a strong filling-in effect, which was dependent on the target flanker separation distance. Although the collinear and orthogonal configuration–induced neuronal activity reached a similar maximal amplitude response, the onset synchronization of neuronal activity at the target location was higher for the collinear condition.

Finally, we showed that the onset synchronization code is superior to the amplitude code when discriminating the collinear from the orthogonal condition. We suggest that these effects may underlie the increased behavioral sensitivity to a low-contrast target and the occurrence of false-alarm target detection in the presence of flankers alone that were reported in psychophysical studies.

Influence of High-Contrast Flankers on Neuronal Population Activity at the Target Site

A salient result of our work is that neuronal activation at the low-contrast target site was strongly influenced in the presence of proximal collinear or orthogonal flankers. The influence was evident by 2 parameters: decreased latency (measured as time to half-maximal amplitude) and increased amplitude of response. Although neuronal activity induced by the low-contrast target began with a long delay, reaching a low-amplitude level, in the presence of high-contrast flankers, activation at the target began earlier, had a faster rise time, and reached higher response amplitude. Moreover, we showed that these influences are similar to the changes observed when increasing the visual stimulus’ contrast: a shortening of the delay and an increase in amplitude. This similarity raises the possibility that adding high-contrast flankers results in “effective” increase of the perceived target's contrast, leading to higher detection sensitivity (Polat and Sagi 1993, 1994a, 1994b, 2007). Previous neurophysiological studies of collinear patterns (Nelson and Frost 1985; Kapadia et al. 1995; Polat et al. 1998; Ito and Gilbert 1999; Gilbert et al. 2000; Kasamatsu et al. 2001; Mizobe et al. 2001; Bauer and Heinze 2002; Crook et al. 2002; Li and Gilbert 2002; Chisum et al. 2003; Khoe et al. 2004) showed that single unit responses were mixed: Sometimes, activity increased and sometimes it decreased; however, in our study, the VSDI signal showed only an increase in activity. Why should that be? A possible explanation is related to the imaging technique we used. As opposed to single unit activity measured from a microelectrode, the VSDI signal in each pixel measures the “summed activity” over a few hundred neurons. It does “not differentiate” between single units and their “individual responses.” Therefore, although the VSDI signal may show a total net increase of population activity, it is likely that some single neurons responded with increased firing rates, whereas other responded with decreased firing rate.

What is the cortical substrate that can mediate this increased activity? It has been suggested previously that the network of long-range horizontal connection has a central role in contextual effects within the V1 (Gilbert and Wiesel 1979, 1983, 1989; Rockland and Lund 1982, 1983; Martin and Whitteridge 1984; Nelson and Frost 1985; Ts'o et al. 1986; Kapadia et al. 1995; Chisum et al. 2003). Nevertheless, feedback from higher-order cortical areas with larger RFs cannot be ruled out as a contributing mechanism. Contribution from feed-forward thalamic spread is probably low, due to the magnification factor for this eccentricity and the small Gabor patches that we have used (Angelucci and Sainsbury 2006a, 2006b)

Although previous electrophysiological studies reported that the orthogonal condition did not induce facilitation or increased activation at the classical RF (Kapadia et al. 1995; Polat et al. 1998; Mizobe et al. 2001; Chisum et al. 2003), we found that population response was increased also for the orthogonal condition. This result can be explained by the fact that VSDI emphasizes subthreshold activity, which is not captured when recording spiking activity. In addition, we averaged the VSDI signal over a large region including multiple orientation columns. Therefore, the increased VSDI signal at the target site for the orthogonal flankers may reflect increased activation in the orthogonal orientation columns within the target site.

Finally, although we were using oriented stimuli, the activation maps depicted in Figure 2 do not demonstrate orientation domains. This observation can be accounted for by the nonoptimal experimental conditions for obtaining orientation maps using VSDI. First, the pixel resolution we have used was far from optimal to obtain orientation maps in the monkey (the typical orientation column width is ∼200–250 μm, whereas the pixel size we used was 170 × 170 μm). In addition, the contrast of the Gabor stimuli was much below 100%, which is an important factor for obtaining high signal-to-noise ratio maps.

Spread of Activation—Filling-in Effects and False-Alarm Responses

Filling-in is an important feature of a sensory system; it provides us with the ability to interpolate sensory attributes in the presence of incomplete information by using spatial and temporal contexts. Several perceptual effects are classified as filling-in phenomena (Pessoa et al. 1998), including contours (Dresp and Bonnet 1991), as well as area filling-in (Ramachandran et al. 1994; Pillow and Rubin 2002). Here, we showed that the presentation of flankers alone induced population activity at the target site, which was even higher than the activity induced by the low-contrast target alone. This result may suggest that subjects will report a high rate of false alarms in target detection during a standard Yes/No experimental paradigm to measure the detection performance of a low-contrast target in the presence of flankers. A recent paper by Polat and Sagi (2007) reported a high rate of false alarm (FA) target detection in subjects that were viewing only the flankers. Therefore, our imaging results confirm Polat and Sagi's behavioral observations. Moreover, the subjects reported a high rate of false alarms that were dependent on the flankers’ distances, with the highest probability when the distance of the flankers was 3λ. Our VSDI results confirmed this observation as well: The highest increased activation was observed for 3λ.

Behavioral Effects

Freeman et al. (2001) have shown that lateral interactions between targets and flankers depend on attention to the flankers. They concluded that thresholds, and thus neuronal activity, depend on the particular behavioral context within which the target/flankers triplets are presented. In our study, the monkeys were trained on a passive fixation task and were not required to detect the target or indicate its orientation. Thus, the direct behavioral effect of target detection within a collinear or orthogonal array was not assessed in our study.

However, the neuronal effects we observed in V1 suggest that VSDI can link cortical activation patterns to Gestalt laws and may suggest further link to behavior. A similar approach was demonstrated recently by Jancke et al. (2004) who used VSDI to demonstrate the putative contribution of subthreshold spread to perceptual correlates of line motion illusion in anesthetized cats.

The relation between our finding of increased activation at target site and the psychophysical facilitation found in terms of reduced contrast detection threshold (better sensitivity) is not clear because we found a response increment without a target (flankers-alone condition). We note that Polat and Sagi (2007) did not find a better sensitivity (d′) with flankers in a mixed-condition yes/no detection experiment, which they attributed to observers’ failure to adjust their criterion to overcome illusory percepts of nonexisting targets (false alarms). According to this analysis, the improvement in contrast sensitivity could be attributed to a decision stage that performs discrimination between some properties of neuronal responses in V1, altered by lateral interactions. In addition, the target contrast we have used (16%) was chosen to obtain a reasonable VSDI response to the Gabor target but this contrast can be above the threshold detection of the monkey. In this case, studying lateral facilitation as defined in behavioral studies in the VSDI signal is problematic. More work is needed in order to explore these properties in relation to behavior.

The Increased Activity at the Target Site Is Smaller Than a Linear Model

The increased activation at the target site fitted a linear model only for the early rise time of the VSDI evoked response (0–120 ms after stimulus onset). The VSDI signal measured in subsequent time frames was subadditive relative to the linear model (120–230 ms). Although previous studies showed supra-additive effects, namely, facilitation, at the target location (Kapadia et al. 1995; Polat et al. 1998; Ito and Gilbert 1999; Gilbert et al. 2000; Kasamatsu et al. 2001; Mizobe et al. 2001), our findings showed the opposite: highly subadditive effects. This difference can be accounted for by differences in the experimental design. As opposed to most previous studies, we used a noncentered RF approach: We imaged the entire population response induced by a Gabor patch or array. Thus, in the VSDI signal, many cells contribute to activity for which the stimulus is not centered in their RF. Indeed, our results are in accordance with Jancke et al. (1999) who studied the distribution of population activation (DPA) for noncentered RF approach and found that the DPA of composite stimuli (single square of light) deviates from the superposition of its components because of distance dependent early excitation and late inhibition. Specifically, they have found sublinear amplitudes of activity for stimuli presented at small distances.

Another possible explanation is related to the fact that VSDI signal emphasizes subthreshold activity. Supra-additive effects were previously reported using extracellular recording of spiking activity. Some of these studies adjusted the flankers’ distance from the RF located at the target site to ensure that the flankers will not induce spiking activity at the target site. In this case, the flankers’ effect on the RF at the target site was always set to zero in terms of spiking activity, but not in terms of subthreshold activity, which was not measured directly using this technique. Unlike extracellular recording of spiking activity, the VSDI technique is highly sensitive to subthreshold population activity and therefore could easily detect subthreshold spreading of activation from the flankers to the target site (Figs. 1 and 2). This sensitivity of the VSDI to subthreshold population activity may account for the sublinearity effect. It is also important to understand that selectively adjusting the flankers’ position for each recorded neuron, suggests that psychophysical judgments are based on a subpopulation of neurons (those that the flankers were positioned outside of their RF); however, it is not clear how the brain selects these neurons. A more reasonable hypothesis is that psychophysical judgments are based on the entire population response, as we measured using the noncentered RF approach.

Another possible explanation relates the target's contrast we used. Polat et al. (1998) pointed to suppressive effects of flankers on target when targets are of medium-high contrast, in the cat visual cortex. The proportion of inhibitory responses increased with target contrast arriving at 55% for 80% target contrast. The target contrast in our study was set to 16%, which according to Polat et al. (1998) included 28% of inhibitory responses (from total of modulated neurons). Although the accurate number might be different in the awake monkey, this can be another source for the sublinearity.

Finally, the VSDI signal sums the “total neuronal activity” over hundreds of neurons in each pixel. Polat et al. (1998) pointed that the proportions of inhibitory units increased with target contrast. Still, it was not clear what the total neuronal activity was? For example, if the number or intensity of excitatory responses increases with the contrast, the summed neuronal activity measured over the entire population may increase or keep on constant level. The VSDI signal does not differentiate between individual neural responses but rather sums the “activity” of all activated neurons, whether they are inhibitory neurons or excitatory neurons.

As Distance between the Flanker and Target Increases, Increased Activation and Filling-In Effects Decrease

In accordance with our results, previous electrophysiology studies showed that the enhanced activity at the target site decreases as the separation distance between the flankers and the target increases (Kapadia et al. 1995; Mizobe et al. 2001). Polat and Sagi (2007) found a monotonous decline in the probability of Hit and false alarms when increasing flanker separation distance. Our VSDI analysis confirmed their observation by revealing similar effects: indeed both increased activation at the target site and filling-in effects declined monotonically with the distance separation (Fig. 3). The greatest increased activation effect we found was in 3λ, in accordance with the psychophysical studies (Polat and Sagi 1993, 1994b, 2007; Williams and Hess 1998; Woods et al. 2002; Cass and Spehar 2005).

Onset Synchronization Is a Better Code to Discriminate between Collinear and Orthogonal Conditions

Previous studies have demonstrated excitatory interaction between single neurons in V1 using cross-correlation analysis (Ts'o et al. 1986; Ts'o and Gilbert 1988). The distribution and range of these interactions corresponded to the clustering and extent of horizontal connection. However, synchronization of neuronal population activity in V1 induced by collinear and orthogonal patterns was not studied thus far. In this study, we found that spatial onset synchronization between adjacent pixels can serve as a better neuronal code than maximal amplitude and the maximal separation amplitude for discriminating between collinear and orthogonal conditions. Moreover, we found that onset synchronization between target and flankers’ pixels was higher in the collinear visual stimuli than in the orthogonal visual stimuli. Our findings provide evidence that onset synchronization can serve as a code for binding or grouping separate collinear visual elements that obey the Gestalt law of good continuation. Finally, the visual stimulus subtracted correlations were much smaller from the raw correlation, indicating that most of the studied correlation was highly linked to the visual stimulus onset.

Chisum et al. (2003) have been using similar visual stimuli while performing optical imaging of intrinsic signals in the visual cortex of anesthetized Tree Shrew. They found that the intrinsic signal activation elicited by an array of collinear Gabor elements was significantly stronger then that elicited by noncollinear array. Indeed, the VSDI maximal separation amplitude was higher for the collinear than the noncollinear. However, we further suggested that temporal synchronization better discriminates between the collinear and orthogonal conditions.

To decode the visual content from the neuronal activity, it is reasonable to assume that the brain does not choose to use separately the amplitude or onset synchronization code. Rather, it is possible that both codes are being used simultaneously, and differences (in amplitude or correlation) dictate the output. Future experiments in contour integration will indicate whether the grouping effect can be extended from collinear elements to full contour integration.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/

Funding

Israel Science Foundation (grant 859/05 to H.S.); German Israeli Foundation (grant 931-237.1/2006 to H.S.).

We are grateful to Hadar Edelman and Ariel Gilad for helping with the experiments and Yossi Shohat for excellent animal care and training.

Conflict of Interest: None declared.

References

Angelucci
A
Sainsbury
K
Contribution of feedforward thalamic afferents and corticogeniculate feedback to the spatial summation area of macaque V1 and LGN
J Comp Neurol
 , 
2006
, vol. 
498
 (pg. 
330
-
351
)
Angelucci
A
Sainsbury
K
Contribution of feedforward thalamic afferents and corticogeniculate feedback to the spatial summation area of macaque V1 and LGN
J Comp Neurol
 , 
2006
, vol. 
498
 (pg. 
330
-
351
)
Arieli
A
Grinvald
A
Slovin
H
Dural substitute for long-term imaging of cortical activity in behaving monkeys and its clinical implications
J Neurosci Methods
 , 
2002
, vol. 
114
 (pg. 
119
-
133
)
Bauer
R
Heinze
S
Contour integration in striate cortex. Classic cell responses or cooperative selection?
Exp Brain Res
 , 
2002
, vol. 
147
 (pg. 
145
-
152
)
Bonneh
Y
Sagi
D
Effects of spatial configuration on contrast detection
Vision Res
 , 
1998
, vol. 
38
 (pg. 
3541
-
3553
)
Callaway
EM
Local circuits in primary visual cortex of the macaque monkey
Annu Rev Neurosci
 , 
1998
, vol. 
21
 (pg. 
47
-
74
)
Cass
JR
Spehar
B
Dynamics of cross- and iso-surround facilitation suggest distinct mechanisms
Vision Res
 , 
2005
, vol. 
45
 (pg. 
3060
-
3073
)
Chisum
HJ
Mooser
F
Fitzpatrick
D
Emergent properties of layer 2/3 neurons reflect the collinear arrangement of horizontal connections in tree shrew visual cortex
J Neurosci
 , 
2003
, vol. 
23
 (pg. 
2947
-
2960
)
Crook
JM
Engelmann
R
Lowel
S
GABA-inactivation attenuates colinear facilitation in cat primary visual cortex
Exp Brain Res
 , 
2002
, vol. 
143
 (pg. 
295
-
302
)
Dresp
B
Bonnet
C
Psychophysical evidence for low-level processing of illusory contours and surfaces in the Kanizsa square
Vision Res
 , 
1991
, vol. 
31
 (pg. 
1813
-
1817
)
Field
DJ
Hayes
A
Chalupa
LM
Werner
JS
Contour integration and the lateral connections of V1 neurons
The visual neurosciences
 , 
2004
1st ed
Cambridge (MA) and London
MIT Press
(pg. 
1069
-
1079
Chapter 70
Field
DJ
Hayes
A
Hess
RF
Contour integration by the human visual system: evidence for a local “association field”
Vision Res
 , 
1993
, vol. 
33
 (pg. 
173
-
193
)
Fitzpatrick
D
Seeing beyond the receptive field in primary visual cortex
Curr Opin Neurobiol
 , 
2000
, vol. 
10
 (pg. 
438
-
443
)
Freeman
E
Sagi
D
Driver
J
Lateral interactions between targets and flankers in low-level vision depend on attention to the flankers
Nat Neurosci
 , 
2001
, vol. 
4
 (pg. 
1032
-
1036
)
Gilbert
C
Ito
M
Kapadia
M
Westheimer
G
Interactions between attention, context and learning in primary visual cortex
Vision Res
 , 
2000
, vol. 
40
 (pg. 
1217
-
1226
)
Gilbert
CD
Wiesel
TN
Morphology and intracortical projections of functionally characterised neurones in the cat visual cortex
Nature
 , 
1979
, vol. 
280
 (pg. 
120
-
125
)
Gilbert
CD
Wiesel
TN
Clustered intrinsic connections in cat visual cortex
J Neurosci
 , 
1983
, vol. 
3
 (pg. 
1116
-
1133
)
Gilbert
CD
Wiesel
TN
Columnar specificity of intrinsic horizontal and corticocortical connections in cat visual cortex
J Neurosci
 , 
1989
, vol. 
9
 (pg. 
2432
-
2442
)
Grinvald
A
Lieke
EE
Frostig
RD
Hildesheim
R
Cortical point-spread function and long-range lateral interactions revealed by real-time optical imaging of macaque monkey primary visual cortex
J Neurosci
 , 
1994
, vol. 
14
 (pg. 
2545
-
2568
)
Grinvald
A
Shoham
D
Shmuel
A
Glaser
D
Vanzetta
I
Shtoyerman
E
Slovin
H
Sterkin
A
Wijnbergen
C
Hildesheim
R
, et al.  . 
Windhorst
U
Johansson
H
In-vivo optical imaging of cortical architecture and dynamics
Modern techniques in neuroscience research
 , 
1999
1st ed.
New York
Springer
(pg. 
893
-
969
)
Hess
R
Field
D
Integration of contours: new insights
Trends Cogn Sci
 , 
1999
, vol. 
3
 (pg. 
480
-
486
)
Hess
RF
Hayes
A
Field
DJ
Contour integration and cortical processing
J Physiol Paris
 , 
2003
, vol. 
97
 (pg. 
105
-
119
)
Ito
M
Gilbert
CD
Attention modulates contextual influences in the primary visual cortex of alert monkeys
Neuron
 , 
1999
, vol. 
22
 (pg. 
593
-
604
)
Jancke
D
Chavane
F
Naaman
S
Grinvald
A
Imaging cortical correlates of illusion in early visual cortex
Nature
 , 
2004
, vol. 
428
 (pg. 
423
-
426
)
Jancke
D
Erlhagen
W
Dinse
HR
Akhavan
AC
Giese
M
Steinhage
A
Schöner
G
Parametric population representation of retinal location: neuronal interaction dynamics in cat primary visual cortex
J Neurosci
 , 
1999
, vol. 
19
 (pg. 
9016
-
9028
)
Kapadia
MK
Ito
M
Gilbert
CD
Westheimer
G
Improvement in visual sensitivity by changes in local context: parallel studies in human observers and in V1 of alert monkeys
Neuron
 , 
1995
, vol. 
15
 (pg. 
843
-
856
)
Kasamatsu
T
Polat
U
Pettet
MW
Norcia
AM
Collinear facilitation promotes reliability of single-cell responses in cat striate cortex
Exp Brain Res
 , 
2001
, vol. 
138
 (pg. 
163
-
172
)
Khoe
W
Freeman
E
Woldorff
MG
Mangun
GR
Electrophysiological correlates of lateral interactions in human visual cortex
Vision Res
 , 
2004
, vol. 
44
 (pg. 
1659
-
1673
)
Kourtzi
Z
Tolias
AS
Altmann
CF
Augath
M
Logothetis
NK
Integration of local features into global shapes: monkey and human FMRI studies
Neuron
 , 
2003
, vol. 
37
 (pg. 
333
-
346
)
Kubota
T
Massively parallel networks for edge localization and contour integration–adaptable relaxation approach
Neural Netw
 , 
2004
, vol. 
17
 (pg. 
411
-
425
)
Li
W
Gilbert
CD
Global contour saliency and local colinear interactions
J Neurophysiol
 , 
2002
, vol. 
88
 (pg. 
2846
-
2856
)
Li
W
Piech
V
Gilbert
CD
Contour saliency in primary visual cortex
Neuron
 , 
2006
, vol. 
50
 (pg. 
951
-
962
)
Malach
R
Amir
Y
Harel
M
Grinvald
A
Relationship between intrinsic connections and functional architecture revealed by optical imaging and in vivo targeted biocytin injections in primate striate cortex
Proc Natl Acad Sci USA
 , 
1993
, vol. 
90
 (pg. 
10469
-
10473
)
Mandon
S
Kreiter
AK
Rapid contour integration in macaque monkeys
Vision Res
 , 
2005
, vol. 
45
 (pg. 
291
-
300
)
Martin
KA
Whitteridge
D
Form, function and intracortical projections of spiny neurones in the striate visual cortex of the cat
J Physiol
 , 
1984
, vol. 
353
 (pg. 
463
-
504
)
Mizobe
K
Polat
U
Pettet
MW
Kasamatsu
T
Facilitation and suppression of single striate-cell activity by spatially discrete pattern stimuli presented beyond the receptive field
Vis Neurosci
 , 
2001
, vol. 
18
 (pg. 
377
-
391
)
Nelson
JI
Frost
BJ
Intracortical facilitation among co-oriented, co-axially aligned simple cells in cat striate cortex
Exp Brain Res
 , 
1985
, vol. 
61
 (pg. 
54
-
61
)
Pessoa
L
Thompson
E
Noe
A
Finding out about filling-in: a guide to perceptual completion for visual science and the philosophy of perception
Behav Brain Sci
 , 
1998
, vol. 
21
 (pg. 
723
-
748
)
Petersen
CC
Grinvald
A
Sakmann
B
Spatiotemporal dynamics of sensory responses in layer 2/3 of rat barrel cortex measured in vivo by voltage-sensitive dye imaging combined with whole-cell voltage recordings and neuron reconstructions
J Neurosci
 , 
2003
, vol. 
23
 (pg. 
1298
-
1309
)
Petersen
CC
Sakmann
B
Functionally independent columns of rat somatosensory barrel cortex revealed with voltage-sensitive dye imaging
J Neurosci
 , 
2001
, vol. 
21
 (pg. 
8435
-
8446
)
Pillow
J
Rubin
N
Perceptual completion across the vertical meridian and the role of early visual cortex
Neuron
 , 
2002
, vol. 
33
 (pg. 
805
-
813
)
Polat
U
Mizobe
K
Pettet
MW
Kasamatsu
T
Norcia
AM
Collinear stimuli regulate visual responses depending on cell's contrast threshold
Nature
 , 
1998
, vol. 
391
 (pg. 
580
-
584
)
Polat
U
Sagi
D
Lateral interactions between spatial channels: suppression and facilitation revealed by lateral masking experiments
Vision Res
 , 
1993
, vol. 
33
 (pg. 
993
-
999
)
Polat
U
Sagi
D
Spatial interactions in human vision: from near to far via experience-dependent cascades of connections
Proc Natl Acad Sci USA
 , 
1994
, vol. 
91
 (pg. 
1206
-
1209
)
Polat
U
Sagi
D
The architecture of perceptual spatial interactions
Vision Res
 , 
1994
, vol. 
34
 (pg. 
73
-
78
)
Polat
U
Sagi
D
The relationship between the subjective and objective aspects of visual filling-in
Vision Res
 , 
2007
, vol. 
47
 (pg. 
2473
-
2481
)
Ramachandran
VS
Ruskin
D
Cobb
S
Rogers-Ramachandran
D
Tyler
CW
On the perception of illusory contours
Vision Res
 , 
1994
, vol. 
34
 (pg. 
3145
-
3152
)
Rockland
KS
Lund
JS
Widespread periodic intrinsic connections in the tree shrew visual cortex
Science
 , 
1982
, vol. 
19
 (pg. 
1532
-
1534
)
Rockland
KS
Lund
JS
Intrinsic laminar lattice connections in primate visual cortex
J Comp Neurol
 , 
1983
, vol. 
216
 (pg. 
303
-
318
)
Shani
R
Sagi
D
Eccentricity effects on lateral interactions
Vision Res
 , 
2005
, vol. 
45
 (pg. 
2009
-
2024
)
Shtoyerman
E
Arieli
A
Slovin
H
Vanzetta
I
Grinvald
A
Long-term optical imaging and spectroscopy reveal mechanisms underlying the intrinsic signal and stability of cortical maps in V1 of behaving monkeys
J Neurosci
 , 
2000
, vol. 
20
 (pg. 
8111
-
8121
)
Slovin
H
Arieli
A
Hildesheim
R
Grinvald
A
Long-term voltage-sensitive dye imaging reveals cortical dynamics in behaving monkeys
J Neurophysiol
 , 
2002
, vol. 
88
 (pg. 
3421
-
3438
)
Solomon
JA
Morgan
MJ
Facilitation from collinear flanks is cancelled by non-collinear flanks
Vision Res
 , 
2000
, vol. 
40
 (pg. 
279
-
286
)
Stettler
DD
Das
A
Bennett
J
Gilbert
CD
Lateral connectivity and contextual interactions in macaque primary visual cortex
Neuron
 , 
2002
, vol. 
36
 (pg. 
739
-
750
)
Ts'o
DY
Gilbert
CD
The organization of chromatic and spatial interactions in the primate striate cortex
J Neurosci
 , 
1988
, vol. 
8
 (pg. 
1712
-
1727
)
Ts'o
DY
Gilbert
CD
Wiesel
TN
Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis
J Neurosci
 , 
1986
, vol. 
6
 (pg. 
1160
-
1170
)
Wertheimer
M
Untersuchungen zur Lehre der Gestalt
Psychol Forsch
 , 
1923
, vol. 
4
 (pg. 
301
-
350
)
Williams
CB
Hess
RF
Relationship between facilitation at threshold and suprathreshold contour integration
J Opt Soc Am A Opt Image Sci Vis
 , 
1998
, vol. 
15
 (pg. 
2046
-
2051
)
Woods
RL
Nugent
AK
Peli
E
Lateral interactions: size does matter
Vision Res
 , 
2002
, vol. 
42
 (pg. 
733
-
745
)