- Email: [email protected]
The dynamics of collinear facilitation: Fast but sustained
Vision Research 48 (2008) 2715–2722
Contents lists available at ScienceDirect
Vision Research journal homepage: www.elsevier.com/locate/visres
The dynamics of collinear facilitation: Fast but sustained Pi-Chun Huang *, Robert F. Hess McGill Vision Research, Department of Ophthalmology, McGill University, Montreal, Quebec, Canada
a r t i c l e
i n f o
Article history: Received 3 January 2008 Received in revised form 4 September 2008
Keywords: Facilitation Dynamics Contrast Temporal summation Conduction velocity Feedback
a b s t r a c t The mechanism by which flanking Gabors facilitate the detection of a central test Gabor is not well understood. Since a knowledge of the dynamics of this effect will help constrain the class of possible model, we conducted three different but interrelated experiments designed to assess different aspects of the dynamics associated with this facilitation. In experiment 1, collinear facilitation was measured at different onset times of a test target for flanks whose contrast was sinusoidally-modulated at 1 Hz. In experiment 2, the order between test target and flanks was investigated by varying the SOA, both stimuli being presented for 50 ms. Experiment 3 assessed temporal summation with and without the flanks. The results obtained do not support either a single channel masking explanation which predicts transient dynamics or an explanation-based solely on the conduction of facilitatory impulses from flanks to target via long-range horizontal connections which predicts transient but delayed dynamics. The results suggest that the dynamics of facilitation are fast but sustained. We propose two underlying mechanisms, a rapid signal to initiate facilitation across large retinal distances, based on feedback from higher centers and a sustained facilitative response based on the temporal integration of locally-responsive, lower-level mechanisms. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction It is well-known that the detectability of a luminance-defined stimulus can be modulated by its neighboring elements and if the neighboring elements form a common global alignment, the detection threshold of the test target decreases, a phenomenon called collinear facilitation (Polat, 1999; Polat & Sagi, 1993; Woods, Nugent, & Peli, 2002). A typical experimental paradigm is to compare the detection threshold of a central test target in the absence and presence of two high contrast collinear flanks. Collinear facilitation has been investigated by manipulating the relationships between test target and flanks for different stimulus parameters, such as spatial frequency, orientation, phase, chromaticity, second-order stimuli, locations of the flanks and separations between test target and flanks, dichoptic viewing and stereoscoptic viewing (Huang & Hess, 2007; Huang, Hess, & Dakin, 2006; Huang, Mullen, & Hess, 2007; Polat, 1999; Polat & Sagi, 1993; Solomon & Morgan, 2000; Williams & Hess, 1998; Woods et al., 2002). However, the temporal properties have not been extensively investigated (Cass & Spehar, 2005; Polat & Sagi, 2006) and may provide valuable insight into the nature of the underlying neural interaction. Psychophysical investigations have suggested potential mechanisms to explain collinear facilitation, a number of which could
* Corresponding author. Present address: Department of Psychology, National Taiwan University, Taiwan. Fax: +1 514 843 1692. E-mail address: [email protected] (P.-C. Huang). 0042-6989/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.visres.2008.09.013
potentially be distinguished in terms of their temporal dynamics. One obvious explanation involves masking/facilitation within a single spatially-elongated mechanism (Solomon, Watson, & Morgan, 1999; Williams & Hess, 1998; Woods et al., 2002). We assume that the receptive field properties of such a mechanism reflect the purely feedforward connections that define the classical receptive field with minimal influence from long-range connections. One would expect the facilitation from flanks to be fast and short acting, being limited to the 50 ms time frame that has already been established for superimposed masking (Georgeson & Georgeson, 1987). Another previously proposed explanation involves the different neural mechanisms with their interactions occurring via long-range cortical afferents (Hirsch & Gilbert, 1991; Ts’o, Gilbert, & Wiesel, 1986; Weliky, Kandler, Fitzpatrick, & Katz, 1995) for which one would expect the facilitation to be solely feedforward in nature with a delayed time scale that matched the slow conducting velocity of these long-range cortical connections. The little temporal information we have at present implicates the role of different neural populations with their interaction occurring over a long time scale, consistent with the influence of long-range transmission assumed to be in the direction from flanks to target (Cass & Spehar, 2005; Polat & Sagi, 2006; Polat, Sterkin, & Yehezkel, 2007). Three experiments were conducted to assess this proposal. In experiment 1, collinear facilitation was measured as the flanks were sinusoidally counter-phasing at 1 Hz (for one temporal period) and the test target was presented (80 ms temporal pulse) at various time offsets. In experiment 2, the order of presentation of
2716
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
the test target and flanks was investigated by varying the stimulus onset asynchrony (SOA) between test target and flanks, both of which were presented for 50 ms. In experiment 3, temporal summation was investigated by comparing temporal summation in the presence and absence of the collinear flanks. Each of these experiments has predictions for models based on either (1) single channel masking where the interaction occurs within the same neuron or neural population or (2) multi-channel interaction involving an interaction between the responses of two different neuronal populations responding to test and flanks respectively via long-range interactions. For example, if collinear facilitation was due to masking within a single neuron or neural population its dynamics would be expected to be fast (Georgeson & Georgeson, 1987) and as a consequence its magnitude should be maximal very shortly after the flanks have been presented. On the other hand, if the facilitation is due to interactions between neurons responding to the flank and neurons detecting the test target via long-range fibers then the effects would be strictly feedforward and delayed with dynamics that covary with the flank to test distance with a time scale that corresponds to the slow conduction velocity of these fibers (Cass & Spehar, 2005). Our results do not support either the single channel masking model (Solomon et al., 1999; Williams & Hess, 1998) or an explanation-based solely on the conduction velocity of long-range horizontal connections (Cass & Spehar, 2005; Polat & Sagi, 2006; Polat et al., 2007). 2. Methods 2.1. Observers Six subjects (PCH, LHY, WW, CTC, YN and PL) with normal or corrected-to-normal vision participated in the experiments. PCH, LHY and WW participated in experiment 1; PCH, YN and PL participated in experiment 2; PCH and CTC participated in experiment 3. 2.2. Apparatus Stimuli were displayed on a Sony Trinitron monitor driven by a VSG 2/5 graphics board (Cambridge Research Systems) with 15 bits contrast resolution, housed in a Pentium PC computer. The frame rate of the display was 120 Hz and the resolution was 1024 768 pixels. The monitor was gamma corrected by software with lookup tables using luminance measurements obtained from a United Detector Technology Optometer (UDT S370) fitted with a 265 photometric sensor. The monitor was viewed in a dimly lit room. The mean luminance of the display was 71 cd/m2. 2.3. Stimuli The target was Gabor patches that were defined by the following equation:
Lðx; yÞ ¼ L0 þ L0 C cos½2pfs ðx x0 Þ þ ss exp½ðx2 þ y2 Þ=2r2s ð1Þ where L0 is the mean luminance, C is the contrast of the Gabor in Michelson contrast units, fs is the spatial frequency of the carrier, rs is the spatial standard deviation of the Gaussian envelope, ss is the phase of the carriers with respect to the center of a Gaussian window. The spatial frequency used was 2 cycles/°, and space constant (r) was 0.3° at a viewing distance of 200 cm. Consequently, the bandwidth (full width at half height) of the Gabor was 0.93 octaves. The Michelson contrast of the flanks was set to 0.6 in experiment 1 and 0.5 in experimentS 2 and 3 (the 0.5 limit was due to the temporal interlacing technique used). The target was presented to the fovea and the orientation of the target and flanks was vertical.
Three target–flank separations were used: 3, 4.5 and 6k in experiment 2. The target–flank separation of 3k was used in experiments 1 and 3. The absolute spatial phase of the target and flanks was in sine phase to reduce the luminance cues. In order to reduce the afterimage of the flanks in experiments 2 and 3, flanks were sandwiched by the 2D binary noise, element size of which is 2*2 pixels (4.2*4.2 arcmin) and the Michelson contrast of the noise was 0.5. 2.4. Procedure A temporal, two-alternative forced-choice (2AFC) paradigm, with auditory feedback, was used to measure the target contrastdetection threshold. Subjects were required to choose which of two intervals contained the target. Subjects were pre-cued to trial onset by an audible tone. The successive presentations were separated by a homogeneous field (1000 ms). A pair of nonius lines were presented around the target position throughout the experiments to reduce the positional uncertainty when no flanks were presented. In experiment 1, a brief target stimulus was presented at different times after the onset of flanks that were contrast reversing at 1 Hz for 1 s. The initial onset times of the test target relative to that of flanking Gabors were 210, 335, 460, 585 and 710 ms with a presentation time of 80 ms, which represented a compromise between temporal integration and time resolution relative to the flanks. The onset times were also blocked together with the single Gabor detection condition. The flank contrast was modulated sinusoidally at 1 Hz. In experiment 2, the temporal order between target and flanks was investigated by varying the SOA between target and flanks, both of which were presented briefly for 50 ms. In order to reduce any afterimage from the flanks, the flanks were temporally sandwiched between 2D noise that was presented at flank location only. The SOA was set to 0, ±50, ±100, ±150, ±200 and ±250 ms. The detection threshold of target alone was measured in the same block in which the flanks were presented. For the target alone condition, the white noise was only presented at the flank location. The method of constant stimuli was used. The SOA at each target–flank separation was blocked in the same run and each separation was chosen randomly. In experiment 3, we investigated the temporal summation for target detection using a two-pulse paradigm in which the flanks can occur before, at or after presentation of one of the two targets. The general procedure was the same as experiment 2, except that one of the two subthreshold test targets was presented (1) simultaneously with the flanks, (2) 50 ms before the flanks and (3) 50 ms after the flanks. SOAs between the two subthreshold test targets were varied and they were blocked together. The target detection threshold in the presence and absence of flanks was measured for different run. A constant stimuli method was used. A bootstrapping procedure (Wichmann & Hill, 2001a, 2001b) was used to derive the confidence interval associated with the Weibull fit, which was defined b as 0:5 þ 0:5 ð1 expððx=aÞ ÞÞ with a, b free parameters. The detection threshold (TH) was determined at 75% correct level. A collinear facilitation index is defined according to the equation:
Threshold elevation ¼ log10 ðTHwith flanks =THwithout flanks Þ
ð2Þ
The significance of the collinear facilitation was based on 95% confidence derived from the bootstrapping methods.
3. Results and discussion 3.1. Experiment 1: effect of test target onset time In experiment 1, collinear facilitation was measured for a target presented at various time onsets with respect to flanks that were sinusoidally-modulated at 1 Hz. The target and flank separation
2717
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
A
B
1
4
2
5
3
3
4
5
1 2
Threshold Elevation [ log (THwithflanks/THwithoutflanks)]
was 3k and the orientation of the target and flanks was vertical. The flanks were modulated in +sine phase and sine phase (see temporal profile in Fig. 1A). The target was presented briefly and the onset time varied during the temporal cycle of the flanks. The results from three subjects were averaged, collapsed across spatial phase configuration (i.e. +sine and sine) and plotted according to the relative onset of the test stimulus, as indicated by the numbers in Fig. 1A. The results show that although slightly more facilitation occurred when the target and flanks were in the first epoch of the same spatial phase, a comparable amount of facilitation occurred even in the spatial out-of-phase configuration (i.e. epochs 4 & 5). The facilitation effect found in the out-of-phase configuration (epochs 4 & 5) is likely to have been carried over from when the high contrast flanks were in the in-phase configuration because there is ample evidence that facilitation is minimal or non-existent when static flank and test stimuli are spatially out of phase (Solomon et al., 1999; Williams & Hess, 1998; but see Zenger & Sagi, 1996). However, if there is facilitation in the sine modulation it might suggest that facilitation can occur prior to the initiation of the flanks. In order to assess this particular issue regarding the importance of the order of presentation (i.e. flank vs. test), the results from this experiment were reanalyzed separately for +sine and sine phase modulation of the flanks and these results are shown separately in Fig. 2B along with predictions for the two current models; the single channel masking model and the multi-channel, long-range conduction model in which facilitatory impulses are assumed to travel from flanks to target via long-range connections. The predictions for the single channel masking model are based on three established features, first that any masking effects are restricted to a time window of 50 ms (Georgeson & Georgeson, 1987) and therefore to obtain a prediction the temporal modulation of the flanks have been convolved with a Gaussian of this width. Second, simultaneous flank facilitation does not occur when the flanks and test are 180° out of phase, therefore facilitation is only expected for one half cycle of the flank modulation (solid curve for +sine modulation of the flanks and dashed curve for sine modulation of the flanks). Third, facilitation saturates at relatively low contrasts of the flanks (Woods et al., 2002). The long-range model predictions are based on the aforementioned phase dependency and contrast saturation of facilitation. Furthermore, assuming a conduction velocity of 0.1 m/s (i.e. long-range
connections, a delay is expected that will depend on the test to flank separation (1.5° in this case). As before (Fig. 1), a slightly stronger collinear facilitation was found for the first epoch, however facilitation was present at all time epochs tested and similar for the +sine (Fig. 2C: filled squares) and sine (Fig. 2C: open squares) modulations of the flanks. This suggests two things, first a long-lasting facilitatory field is produced from the in-phase flanks and second comparable facilitation can occur when the target precedes the flanks. Neither of these effects would be expected from either single channel masking or multi-channel interactions-based solely on the conduction velocity of long-range connections assumed to travel in the direction from flanks to target (Fig. 2B). 3.2. Experiment 2: the order of presentation of the test target and flanks The order of presentation of the test target and flanks were investigated further in this experiment. The amount of facilitation was plotted against the SOA between flanks and test in Fig. 3 for each subject. An asymmetrical distribution was found. To derive the location of peak facilitation the results were fitted by a Gaussian function and the derived peak location measures are plotted in Fig. 4. Best facilitation occurred when the test target preceded the flanks (30 ms for average data at 3k separation). When the SOA between test target and flanks was larger than 150 ms, no significant facilitation was found. Second, as expected, the amount of collinear facilitation reduced as the target–flank separation increased and the position of peak facilitation scales with the test to flank distance. Fig. 4 shows a comparison between the position of peak facilitation as a function of the test to flank distance in units of mm of cortex (see Eq. 3) and the results of Cass and Spehar (2005). If one assumes that the change in dynamics with test to flank distance is completely accounted for by conduction time, the derived velocities, although showing some variability across subjects (0.18–1.67 m/s), the average result of 0.32 m/s is not that different from that derived by Cass and Spehar of 0.23 m/s. However, the one fundamental difference between the present results and those of Cass and Spehar (2005) is that the peak facilitation occurred when the test target preceded the flanks (hence the negative values in Fig. 4 for our results) whereas the obvious interpretation of Cass and Spehars’ model require it to occur when flanks precede the test stimulus (hence the positive values in Fig. 4 for the Cass
0.1 0.0 -0.1 -0.2 LHY PCH WW AVG
-0.3 -0.4 1
2
3
4
5
Epoch #
Fig. 1. Results for experiment 1. (A) The temporal profile of the stimuli. The solid wave represents the flanks modulated in +sine phase and the dash wave represents the flanks were modulated in sine phase. The marked epochs represent the onset time of the target. The number denoting each epoch indicates it has the same temporal and spatial relationship for the +sine and sine configuration. (B) Experimental results for three subjects and the average data is shown in red and the error bars represent standard errors among subjects. (For interpretation of the references to color in this figure legend the reader is referred to the web version of the paper.)
2718
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
B 0.2
0.2
4‘
2
5‘
3
3‘
4
1‘
5
Threshold Elevation
1
Prediction: Single Channel Model
1
0.1
Threshold Elevation
A
2 1‘
2‘
0.0 -0.1 -0.2 -0.3
2
0
2‘
0.0 -0.1 -0.2 -0.3
200
400
600
0
800 1000
200
time (ms)
Threshold Elevation
C 0.2
0.2
LHY
3
-0.4
-0.4
2‘
Prediction: Long-Range Model
0.1
0.2
PCH
400
600
0.2
0.1
0.1
0.1
0.0
0.0
0.0
0.0
-0.1
-0.1
-0.1
-0.1
-0.2
-0.2
-0.2
-0.2
-0.3
-0.3
-0.3
-0.3
-0.4
-0.4 1
2
3
Epoch #
4
5
1
2
3
Epoch
4
#
5
AVG
WW
0.1
-0.4
800 1000
time (ms)
-0.4 1
2
3
Epoch
4
#
5
1
2
3
Epoch
4
5
#
Fig. 2. Results for experiment 1. (A) The temporal profile of the stimuli. (B) Predictions based on the two current models. (C) The collinear facilitation was plotted against the epoch number as shown in (A) for each individual as well as averaged data. The results for the +sine modulation are shown as filled squares while those for the sine modulation are shown as open squares.
and Spehar results but see general discussion for a more general interpretation of this model based on Cass’s earlier work (Cass, 2005) and that of Cass, Spehar, Alais, and Arrighi (2005)). The present results showing essentially negative values are inconsistent with an explanation based on the conduction of facilitatory signals from flanks to test stimulus (Cass & Spehar, 2005).
SOAs when the target succeeds the flanks (Fig. 5, right column). This suggests that the facilitation is maximal at or around when the test and flanks coincide in time, similar to the conclusion arrived at in experiment 2. There is no indication of facilitation being maximal at some time after the flanks have been presented. 4. General discussion
3.3. Experiment 3: temporal summation of test target The results of measuring the temporal summation of flank facilitation are shown in Fig. 5. The stimulus arrangements are illustrated at the top of each column. One of the two subthreshold test stimuli in this typical two-pulse paradigm, is either aligned in time with the flanks (top, middle) or presented before (top, left) or after (top, right) the flanks. The SOA between two test stimuli was varied on a temporal grid of ±200 ms. The curves in each of the columns of Fig. 5 show a comparison for two subjects between the temporal summation associated with the detection of an isolated Gabor compared with that in the presence of the facilitatory flanks for each of the three previously described temporal-alignments (see top illustrations). The temporal summation function (i.e. the autocorrelation function of the filters response) for the isolated Gabor has the typical unipolar form, consistent with lowpass dynamics. The presence of the flanks does not substantially change the form of the summation apart from imparting an overall facilitation. The time points where this facilitation is statistically significant have been marked with an asterisk. What is evident, particularly in the results of subject PCH, is that the facilitation is symmetrically arranged when the test is temporally coincident with the flanks (Fig. 5, middle column), it is asymmetrically displaced to more positive SOAs when the test precedes the flanks (Fig. 5, left column) and asymmetrically displaced to more negative
The experiments reported here have revealed that a complex set of dynamics underlies collinear facilitation. The dynamics are certainly not as simple as either of the two current models of collinear facilitation would predict. First, the effects of the flanks are not just instantaneous, the results of experiment 1 demonstrate that they can be long lasting, suggesting a sustained facilitatory response, one that can be built up over time. Secondly, facilitation is maximal at or before flank presentation, not following flank presentation (experiments 1, 2 and 3), suggesting a rapidly initiated interaction across different parts of the visual field for mechanisms of the same orientation forming a global alignment. 4.1. Underlying mechanisms One hypothesis assumes that collinear facilitation is a special example of subthreshold summation within a single, elongated V1 neuron detecting both flank and test stimuli (Solomon et al., 1999; Williams & Hess, 1998). In such a case, one would expect the dynamics to be simple and predictable on the basis of what we currently know about simultaneous masking (Georgeson & Georgeson, 1987). For example, facilitation would be expected to follow the in-phase modulation of the flanks in experiment 1 (as per prediction in Fig. 1B), to be limited to a 50 ms time window, centered on the flank presentation
2719
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
Threshold Elevation
-200
200
0
0.2
0.2
0.1
0.1
0.0
0.0
-0.1
-0.1
-0.2
-0.3
-0.2 3λ 4.5 λ 6 λ
PCH
-200
-100
0
100
YN
-200
200
-100
0
100
200
Time lag between target and flanks (ms)
0.2
0.2
0.1
0.1
0.0
0.0
-0.1
-0.1
-0.2
3λ 4.5 λ 6λ
-0.3
Time lag between target and flanks (ms)
Threshold Elevation
time
-0.2 3λ 4.5 λ 6λ
-0.3 PB
-200
-100
0
100
200
Time lag between target and flanks (ms)
3λ 4.5 λ 6λ
-0.3 AVG
-200
-100
0
100
200
Time lag between target and flanks (ms)
Fig. 3. (Top) The spatio-temporal profile of the stimuli. (Bottom) Experimental results for each subject and the average data for three subjects for three different tests to flank distances. The data have been fitted by a Gaussian function and the peak values so derived are plotted in Fig. 4.
(Georgeson & Georgeson, 1987) and to have dynamics that are invariant with test to flank distance. These model predictions in experiment 1 were not supported, the dynamics were such that peak facilitation occurred before presentation of the flanks and the effects of facilitation were long lasting. Furthermore, the temporal summation for the target was less than 100 ms (experiment 3) but the temporal summation between target and the flanks (experiment 2) is longer than 150 ms, not supporting the subthreshold summation hypothesis. Another hypothesis states that collinear facilitation is the result of interactions between neurons in different locations within V1, via slowly conducting, long-range connections. This transmission of facilitatory activity is assumed to be from flanks to test, and hence stronger facilitation would be expected to occur when flanks precede the test by a period predicted by the flank to test distance and the slow conduction velocities of this fiber system (i.e. 0.1–0.2 m/s). However, the results of experiments 1, 2 and 3 that show maximal facilitation occurs at or prior to presentation of the flanks is incompatible with this explanation.
In the original experiment of Cass and Spehar (2005) since facilitation was maximal at longer presentation durations one is led to assume that this would be equivalent to a positive SOA dependence consistent with the transmission time from flanks to test. However, we show here, using a variety of different techniques that such SOA dependence does not occur. An alternative explanation for the duration dependency of flank facilitation could be couched in terms of temporal integration, an important issue not considered within Cass and Spehars’ model. If the strength of neural integration varies with test to flank distance (Cass & Alais, 2006), this could explain their results. The most challenging finding of the present study is that facilitation is maximal at or prior to flank presentation. A similar finding was recently reported by Cass and Alais (2006). Here too an explanation was advanced in terms of the much slower dynamics associated with neural integration (Gawne, Kjaer, & Richmond, 1996) for low contrast test stimuli compared to that of the higher contrast flanks. The neural effects set up by the initial presentation of the test stimulus may last until the later presentation of the flanks (see Polat et al.,
2720
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
Position of peak facilitation
2007 for discussion). Alternatively, it may be the consequence of an anticipatory response from higher visual areas. Whatever the explanation it suggests a rapid orientation specific interaction involving a relatively local patch of cortex. Feedback connections have been described from V2 to V1 that satisfy these criteria. Feedback connections from V2 to V1 result in a rapid (Girard, Hupe, & Bullier, 2001) excitation of cells in like-orientation columns in local regions of V1 and these could play a role in enhancing the detectability of collinear image features (Shmuel, Korman, Sterkin, Harel, Ullman, Malach, & Grinvald, 2005).
Cass & Spehar (2005)
100
PCH YN PB AVG
50
0.23 m/s Flanks precede Target
0.18 m/s 1.67 m/s 0
0.32 m/s Target precedes Flanks
0.31 m/s -50
4.2. Comparison with previous studies We do not support the conclusions of an earlier study by Polat and Sagi (2006) where facilitation was only found when the flanks preceded the target, or were presented simultaneously, but not when the target preceded the flanks. However, it should be pointed out that there are a number of possible experimental differences between our studies and theirs. Our experiment 1 design is similar (sustained presentation of flanks) but not identical (static verses dynamic) to their experiment 2 in which the target was presented at the beginning, the middle and the end of flank presentation. They argued that collinear facilitation only occurs when the target is presented sometime after the flanks. Our data from experiment 1 suggest that collinear facilitation occurs at all temporal phases. Our
-100 5
10
15
20
25
30
Separation (mm in cortex) Fig. 4. Estimated conduction velocities. The peak facilitation positions are plotted as a function of target–flank separation for each subject. The horizontal axis represents the cortical distance in mm between target and flanks (from center to center). The results of Cass and Spehar (2005) are included for reference. In order to estimate the striate distance associated with a given visual angle (E) relative to the R 17:3 dE ¼ 17:3 lnðE þ 0:75Þ (3) where y fovea, the equation below was used y ¼ Eþ0:75 is the mm across retinotopic striate surface and E is the degrees of visual angle (Cass & Spehar, 2005; Horton & Hoyt, 1991).
<==SO A==||
<==SO A==||
-100
0
*
*
100
*
*
*
<==SO A==||
-100
time *
*
100
0
***
*
*
3
*
-100
time
**
**
**
8 7
0.01
PCH -50ms Without Flanks With Flanks -200
-100
9 8 7
0
100
-200
200
5
4
4
3
3
Threshold
Threshold
5
2
0.01 9 8 7 6 5
-100
9 8 7
0
100
*
*
*
7 6 5
-200
-100
0
100
200
SOA between test targets (ms)
0
100
200
*
*
4
3
2
9 8
-100
SOA between test targets (ms) 5
*
2
0.01
0.01
CTC -50ms Without Flanks With Flanks
*
*
PCH 50ms Without Flanks With Flanks -200
200
SOA between test targets (ms)
SOA between test targets (ms)
time
2
0.01
PCH 0 ms Without Flanks With Flanks
Threshold
9
2
Threshold
Threshold
Threshold
0.01
*
100
3
3
2
0
9 8
CTC 0 ms
7 6
Without Flanks With Flanks -200
-100
5
0
100
200
SOA between test targets (ms)
CTC 50ms Without Flanks With Flanks -200
-100
0
100
200
SOA between test targets (ms)
Fig. 5. The temporal arrangement of the stimuli and threshold for two subjects in the absence and presence of the flanks. The arrangement above shows the time relationship between one of the two test targets and the flanks. The other test target in the two-pulse paradigm was displayed on a grid of ±200 ms.
2721
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
experiment 2 is similar though not identical to their experiment 1 in which flanks were presented transiently. Unlike Polat and Sagi (2006) we show that maximal facilitation occurs when the target precedes the flanks or when it coincides in time with the presentation of the flanks. In a recent paper alluded to above, Cass and Alais (2006) presented data using a rotating flank paradigm in which facilitatory effects were found both when the test followed the flanks and when the test preceded the flanks in time. They interpret their results in terms nerve conduction from flanks to target in the former and in terms of neural integration at low contrasts in the latter. We agree with them concerning the explanation for facilitation when the test precedes the flanks in time but disagree with them in the case when the test follows the flanks in time. It is worth noting that their interpretation concerning nervous conduction assumes a linear dependence on orientation that is invariant with test to flank distance over a range of ±50°. There is good reason to believe that this is not the case. Fig. 6 shows the results where the orientation of temporally coincident flanks is varied in a way similar to Cass and Alais (2006) for three different test to flank distances. Not only does the slope of the relationship between facilitation and test to flank orientation change with test to flank distance but also, at close distances, facilitation is replaced by inhibition. The intricacies of these orientation-dependent effects would need to be taken into account before any firm conclusions can be arrived at concerning the dynamics. We took the more direct approach and explicitly varied the timing of test and flank stimuli by varying the SOA of aligned targets of constant duration. We found that facilitation either coincided with the presentation of the flanks or preceded them. We feel that instead of two separate explanations for these two findings where facilitation is observed (test precedes flanks and flanks pre-
4.3. A more general long-range model? The long-range conduction model discussed so far was that of Cass and Spehar (2005) in which it is assumed (though never explicitly stated) that the facilitatory transmission travels from flanks to target via long-range fibers. Our findings are not consistent with this. Cass’s original work (Cass, 2005 and later Cass et al., 2005) however did show that when SOA was varied between flanks and target, greatest facilitation occurs when the target precedes the flanks, just as we show here using a variety of approaches. This finding compromises the most obvious interpretation of the model proposed by Cass and Spehar (2005) unless one assumes that the facilitatory impulses can travel in either direction (i.e. from flanks to target and from target to flanks). Cass realized this and proposed this more general interpretation (Cass, 2005 and later Cass et al., 2005). There are two objections one could raise to this extended interpretation. First, the task is to detect the target and not the flanks and so it is difficult to understand why the presentation of the target itself would initiate facilitation of its own detection. Second, one should see a comparable amount of facilitation when the flanks precede the target which we do not observe. In Cass’s original results (also see Cass et al., 2005), a large degree of suppression occurred when the flanks preceded the target and he hypothesized that this may have masked the expected facilitation under these conditions. He proposed either backward masking by the
λ
0.4
S Shape 45°
0.2
HT 1000ms S-Shape
0.0
2λ 4λ 6λ
-0.2 0
15
30
45
60
75
λ
0.4
Threshold Elevation
0.6
Threshold Elevation
cedes test), on grounds of parsimony a single explanation can be couched simply in terms of the feedforward effects of the neural integration associated with the processing of test and flankers, combined with fast feedback interactions between test and flanker locations.
S Shape 45°
0.2
PCH 1000ms S-Shape 0.0 2λ 4λ 6λ
-0.2
90
0
Orientation (Degrees)
λ
0.4
C Shape 45 °
0.2
HT 1000ms C-Shape 0.0 2λ 4λ 6λ
-0.2 15
30
45
60
75
Orientation (Degrees)
30
45
60
75
90
90
λ
0.4
Threshold Elevation
Threshold Elevation
0.6
0
15
Orientation (Degrees)
C Shape 45 °
0.2
PCH 1000ms C-Shape 0.0
2λ 4λ 6λ
-0.2 0
15
30
45
60
75
90
Orientation (Degrees)
Fig. 6. Threshold elevation as a function of the orientation of the flanks. Two different flank orientation configurations are examined: one in which the relative orientation of the flanks is the same (S shape), and one in which it is opposite (C Shape). Results for two subjects are shown each at three different test to flank distances.
2722
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
B
0.2
0.1
Threshold Elevation
Threshold Elevation
A
0.0
-0.1
-0.2
-0.3
3 λ (2 cpd) 3 λ (6 cpd)
PCH -200
-100
0
100
200
Time lag between target and flanks (ms)
0.2 EXP 2 Replication of Cass' Exp 0.1
0.0
-0.1
-0.2
-0.3
PCH -200
-100
0
100
200
Time lag between target and flanks (ms)
Fig. 7. (A) The collinear facilitation effect as a function of the SOA between target and flanks at different spatial frequencies. The black circles represent 2c/d (replotted from Fig. 3) and gray squares represent 6c/d. The results showed similar asymmetrical distribution for both spatial frequencies. (B) The threshold elevation as a function of SOA between target and flanks in the presence of noise (replotted from Fig. 3) and absence of noise (replication of Cass’ (2005) experiment) at the position of the flanks. Both showed similar results.
noise mask or motion artifacts as the basis for this suppression. We do not find that the particular spatial frequency (2c/d) that we used is crucial for our conclusions. Previously we have used lower spatial frequencies (i.e. 0.75c/d) and others have often used higher spatial frequencies. All previous studies have demonstrated robust facilitation at all of these different scales. To be completely sure, we have now replicated our main findings (i.e. no facilitation and minimal suppression when the flanks precede the test target; Fig. 3) at a higher spatial frequency (6c/d) and for a different arrangement of noise masks (i.e. noise masks that cover the whole display and occur only at the very beginning and end of the trial) to better compare with the experimental conditions used by Cass (2005) and Cass et al. (2005) and the results are shown in Fig. 7. We did not observe any measurable suppression when the flanks preceded the target and hence our finding of no facilitation under these conditions argues against the notion of there being facilitatory transmission from flanks to target. The results suggest that the dynamics of facilitation are fast but sustained. We propose two underlying mechanisms, a rapid signal to initiate facilitation across large retinal distances, based on feedback from higher centers and a sustained facilitative response based on the temporal integration of locally-responsive, lower-level mechanisms. References Cass, J. (2005). Spatial and temporal determinants of long-range contrast facilitation and supperssion. University of New South Wales, Kensington. PhD Thesis. Cass, J., & Alais, D. (2006). The mechanisms of collinear integration. Journal of Vision, 6(9), 915–922. Cass, J., Spehar, B., Alais, D., & Arrighi, R. (2005). Spatial and temporal determinants of contrast facilitation and suppression [abstract]. Journal of Vision, 5(8), 459a. Cass, J. R., & Spehar, B. (2005). Dynamics of collinear contrast facilitation are consistent with long-range horizontal striate transmission. Vision Research, 45(21), 2728–2739. Gawne, T. J., Kjaer, T. W., & Richmond, B. J. (1996). Latency: Another potential code for feature binding in striate cortex. Journal of Neurophysiology, 76(2), 1356–1360. Georgeson, M. A., & Georgeson, J. M. (1987). Facilitation and masking of briefly presented gratings: Time-course and contrast dependence. Vision Research, 27(3), 369–379.
Girard, P., Hupe, J. M., & Bullier, J. (2001). Feedforward and feedback connections between areas v1 and v2 of the monkey have similar rapid conduction velocities. Journal of Neurophysiology, 85(3), 1328–1331. Hirsch, J. A., & Gilbert, C. D. (1991). Synaptic physiology of horizontal connections in the cat’s visual cortex. Journal of Neuroscience, 11(6), 1800–1809. Horton, J. C., & Hoyt, W. F. (1991). The representation of the visual field in human striate cortex. A revision of the classic holmes map. Archives of Ophthalmology, 109(6), 816–824. Huang, P. C., & Hess, R. F. (2007). Collinear facilitation: Effect of additive and multiplicative external noise. Vision Research, 47(24), 3108–3119. Huang, P. C., Hess, R. F., & Dakin, S. C. (2006). Flank facilitation and contour integration: Different sites. Vision Research, 46(21), 3699–3706. Huang, P. C., Mullen, K. T., & Hess, R. F. (2007). Collinear facilitation in color vision. Journal of Vision, 7(11): 6, 1–14. Polat, U. (1999). Functional architecture of long-range perceptual interactions. Spatial Vision, 12(2), 143–162. Polat, U., & Sagi, D. (1993). Lateral interactions between spatial channels: Suppression and facilitation revealed by lateral masking experiments. Vision Research, 33(7), 993–999. Polat, U., & Sagi, D. (2006). Temporal asymmetry of collinear lateral interactions. Vision Research, 46(6-7), 953–960. Polat, U., Sterkin, A., & Yehezkel, O. (2007). Spatio-temporal low-level neural networks account for visual masking. Advances in Cognitive Psychology, 3(1–2), 153–165. Shmuel, A., Korman, M., Sterkin, A., Harel, M., Ullman, S., Malach, R., et al. (2005). Retinotopic axis specificity and selective clustering of feedback projections from v2 to v1 in the owl monkey. Journal of Neuroscience, 25(8), 2117–2131. Solomon, J. A., & Morgan, M. J. (2000). Facilitation from collinear flanks is cancelled by non-collinear flanks. Vision Research, 40(3), 279–286. Solomon, J. A., Watson, A. B., & Morgan, M. J. (1999). Transducer model produces facilitation from opposite-sign flanks. Vision Research, 39(5), 987–992. Ts’o, D. Y., Gilbert, C. D., & Wiesel, T. N. (1986). Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis. Journal of Neuroscience, 6(4), 1160–1170. Weliky, M., Kandler, K., Fitzpatrick, D., & Katz, L. C. (1995). Patterns of excitation and inhibition evoked by horizontal connections in visual cortex share a common relationship to orientation columns. Neuron, 15(3), 541–552. Wichmann, F. A., & Hill, N. J. (2001a). The psychometric function: I. Fitting, sampling, and goodness of fit. Perception & Psychophysics, 63(8), 1293–1313. Wichmann, F. A., & Hill, N. J. (2001b). The psychometric function: II. Bootstrapbased confidence intervals and sampling. Perception & Psychophysics, 63(8), 1314–1329. Williams, C. B., & Hess, R. F. (1998). Relationship between facilitation at threshold and suprathreshold contour integration. Journal of the Optical Society of America A, Optics, Image Science, and Vision, 15(8), 2046–2051. Woods, R. L., Nugent, A. K., & Peli, E. (2002). Lateral interactions: Size does matter. Vision Research, 42(6), 733–745. Zenger, B., & Sagi, D. (1996). Isolating excitatory and inhibitory nonlinear spatial interactions involved in contrast detection. Vision Research, 36(16), 2497–2513.
Contents lists available at ScienceDirect
Vision Research journal homepage: www.elsevier.com/locate/visres
The dynamics of collinear facilitation: Fast but sustained Pi-Chun Huang *, Robert F. Hess McGill Vision Research, Department of Ophthalmology, McGill University, Montreal, Quebec, Canada
a r t i c l e
i n f o
Article history: Received 3 January 2008 Received in revised form 4 September 2008
Keywords: Facilitation Dynamics Contrast Temporal summation Conduction velocity Feedback
a b s t r a c t The mechanism by which flanking Gabors facilitate the detection of a central test Gabor is not well understood. Since a knowledge of the dynamics of this effect will help constrain the class of possible model, we conducted three different but interrelated experiments designed to assess different aspects of the dynamics associated with this facilitation. In experiment 1, collinear facilitation was measured at different onset times of a test target for flanks whose contrast was sinusoidally-modulated at 1 Hz. In experiment 2, the order between test target and flanks was investigated by varying the SOA, both stimuli being presented for 50 ms. Experiment 3 assessed temporal summation with and without the flanks. The results obtained do not support either a single channel masking explanation which predicts transient dynamics or an explanation-based solely on the conduction of facilitatory impulses from flanks to target via long-range horizontal connections which predicts transient but delayed dynamics. The results suggest that the dynamics of facilitation are fast but sustained. We propose two underlying mechanisms, a rapid signal to initiate facilitation across large retinal distances, based on feedback from higher centers and a sustained facilitative response based on the temporal integration of locally-responsive, lower-level mechanisms. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction It is well-known that the detectability of a luminance-defined stimulus can be modulated by its neighboring elements and if the neighboring elements form a common global alignment, the detection threshold of the test target decreases, a phenomenon called collinear facilitation (Polat, 1999; Polat & Sagi, 1993; Woods, Nugent, & Peli, 2002). A typical experimental paradigm is to compare the detection threshold of a central test target in the absence and presence of two high contrast collinear flanks. Collinear facilitation has been investigated by manipulating the relationships between test target and flanks for different stimulus parameters, such as spatial frequency, orientation, phase, chromaticity, second-order stimuli, locations of the flanks and separations between test target and flanks, dichoptic viewing and stereoscoptic viewing (Huang & Hess, 2007; Huang, Hess, & Dakin, 2006; Huang, Mullen, & Hess, 2007; Polat, 1999; Polat & Sagi, 1993; Solomon & Morgan, 2000; Williams & Hess, 1998; Woods et al., 2002). However, the temporal properties have not been extensively investigated (Cass & Spehar, 2005; Polat & Sagi, 2006) and may provide valuable insight into the nature of the underlying neural interaction. Psychophysical investigations have suggested potential mechanisms to explain collinear facilitation, a number of which could
* Corresponding author. Present address: Department of Psychology, National Taiwan University, Taiwan. Fax: +1 514 843 1692. E-mail address: [email protected] (P.-C. Huang). 0042-6989/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.visres.2008.09.013
potentially be distinguished in terms of their temporal dynamics. One obvious explanation involves masking/facilitation within a single spatially-elongated mechanism (Solomon, Watson, & Morgan, 1999; Williams & Hess, 1998; Woods et al., 2002). We assume that the receptive field properties of such a mechanism reflect the purely feedforward connections that define the classical receptive field with minimal influence from long-range connections. One would expect the facilitation from flanks to be fast and short acting, being limited to the 50 ms time frame that has already been established for superimposed masking (Georgeson & Georgeson, 1987). Another previously proposed explanation involves the different neural mechanisms with their interactions occurring via long-range cortical afferents (Hirsch & Gilbert, 1991; Ts’o, Gilbert, & Wiesel, 1986; Weliky, Kandler, Fitzpatrick, & Katz, 1995) for which one would expect the facilitation to be solely feedforward in nature with a delayed time scale that matched the slow conducting velocity of these long-range cortical connections. The little temporal information we have at present implicates the role of different neural populations with their interaction occurring over a long time scale, consistent with the influence of long-range transmission assumed to be in the direction from flanks to target (Cass & Spehar, 2005; Polat & Sagi, 2006; Polat, Sterkin, & Yehezkel, 2007). Three experiments were conducted to assess this proposal. In experiment 1, collinear facilitation was measured as the flanks were sinusoidally counter-phasing at 1 Hz (for one temporal period) and the test target was presented (80 ms temporal pulse) at various time offsets. In experiment 2, the order of presentation of
2716
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
the test target and flanks was investigated by varying the stimulus onset asynchrony (SOA) between test target and flanks, both of which were presented for 50 ms. In experiment 3, temporal summation was investigated by comparing temporal summation in the presence and absence of the collinear flanks. Each of these experiments has predictions for models based on either (1) single channel masking where the interaction occurs within the same neuron or neural population or (2) multi-channel interaction involving an interaction between the responses of two different neuronal populations responding to test and flanks respectively via long-range interactions. For example, if collinear facilitation was due to masking within a single neuron or neural population its dynamics would be expected to be fast (Georgeson & Georgeson, 1987) and as a consequence its magnitude should be maximal very shortly after the flanks have been presented. On the other hand, if the facilitation is due to interactions between neurons responding to the flank and neurons detecting the test target via long-range fibers then the effects would be strictly feedforward and delayed with dynamics that covary with the flank to test distance with a time scale that corresponds to the slow conduction velocity of these fibers (Cass & Spehar, 2005). Our results do not support either the single channel masking model (Solomon et al., 1999; Williams & Hess, 1998) or an explanation-based solely on the conduction velocity of long-range horizontal connections (Cass & Spehar, 2005; Polat & Sagi, 2006; Polat et al., 2007). 2. Methods 2.1. Observers Six subjects (PCH, LHY, WW, CTC, YN and PL) with normal or corrected-to-normal vision participated in the experiments. PCH, LHY and WW participated in experiment 1; PCH, YN and PL participated in experiment 2; PCH and CTC participated in experiment 3. 2.2. Apparatus Stimuli were displayed on a Sony Trinitron monitor driven by a VSG 2/5 graphics board (Cambridge Research Systems) with 15 bits contrast resolution, housed in a Pentium PC computer. The frame rate of the display was 120 Hz and the resolution was 1024 768 pixels. The monitor was gamma corrected by software with lookup tables using luminance measurements obtained from a United Detector Technology Optometer (UDT S370) fitted with a 265 photometric sensor. The monitor was viewed in a dimly lit room. The mean luminance of the display was 71 cd/m2. 2.3. Stimuli The target was Gabor patches that were defined by the following equation:
Lðx; yÞ ¼ L0 þ L0 C cos½2pfs ðx x0 Þ þ ss exp½ðx2 þ y2 Þ=2r2s ð1Þ where L0 is the mean luminance, C is the contrast of the Gabor in Michelson contrast units, fs is the spatial frequency of the carrier, rs is the spatial standard deviation of the Gaussian envelope, ss is the phase of the carriers with respect to the center of a Gaussian window. The spatial frequency used was 2 cycles/°, and space constant (r) was 0.3° at a viewing distance of 200 cm. Consequently, the bandwidth (full width at half height) of the Gabor was 0.93 octaves. The Michelson contrast of the flanks was set to 0.6 in experiment 1 and 0.5 in experimentS 2 and 3 (the 0.5 limit was due to the temporal interlacing technique used). The target was presented to the fovea and the orientation of the target and flanks was vertical.
Three target–flank separations were used: 3, 4.5 and 6k in experiment 2. The target–flank separation of 3k was used in experiments 1 and 3. The absolute spatial phase of the target and flanks was in sine phase to reduce the luminance cues. In order to reduce the afterimage of the flanks in experiments 2 and 3, flanks were sandwiched by the 2D binary noise, element size of which is 2*2 pixels (4.2*4.2 arcmin) and the Michelson contrast of the noise was 0.5. 2.4. Procedure A temporal, two-alternative forced-choice (2AFC) paradigm, with auditory feedback, was used to measure the target contrastdetection threshold. Subjects were required to choose which of two intervals contained the target. Subjects were pre-cued to trial onset by an audible tone. The successive presentations were separated by a homogeneous field (1000 ms). A pair of nonius lines were presented around the target position throughout the experiments to reduce the positional uncertainty when no flanks were presented. In experiment 1, a brief target stimulus was presented at different times after the onset of flanks that were contrast reversing at 1 Hz for 1 s. The initial onset times of the test target relative to that of flanking Gabors were 210, 335, 460, 585 and 710 ms with a presentation time of 80 ms, which represented a compromise between temporal integration and time resolution relative to the flanks. The onset times were also blocked together with the single Gabor detection condition. The flank contrast was modulated sinusoidally at 1 Hz. In experiment 2, the temporal order between target and flanks was investigated by varying the SOA between target and flanks, both of which were presented briefly for 50 ms. In order to reduce any afterimage from the flanks, the flanks were temporally sandwiched between 2D noise that was presented at flank location only. The SOA was set to 0, ±50, ±100, ±150, ±200 and ±250 ms. The detection threshold of target alone was measured in the same block in which the flanks were presented. For the target alone condition, the white noise was only presented at the flank location. The method of constant stimuli was used. The SOA at each target–flank separation was blocked in the same run and each separation was chosen randomly. In experiment 3, we investigated the temporal summation for target detection using a two-pulse paradigm in which the flanks can occur before, at or after presentation of one of the two targets. The general procedure was the same as experiment 2, except that one of the two subthreshold test targets was presented (1) simultaneously with the flanks, (2) 50 ms before the flanks and (3) 50 ms after the flanks. SOAs between the two subthreshold test targets were varied and they were blocked together. The target detection threshold in the presence and absence of flanks was measured for different run. A constant stimuli method was used. A bootstrapping procedure (Wichmann & Hill, 2001a, 2001b) was used to derive the confidence interval associated with the Weibull fit, which was defined b as 0:5 þ 0:5 ð1 expððx=aÞ ÞÞ with a, b free parameters. The detection threshold (TH) was determined at 75% correct level. A collinear facilitation index is defined according to the equation:
Threshold elevation ¼ log10 ðTHwith flanks =THwithout flanks Þ
ð2Þ
The significance of the collinear facilitation was based on 95% confidence derived from the bootstrapping methods.
3. Results and discussion 3.1. Experiment 1: effect of test target onset time In experiment 1, collinear facilitation was measured for a target presented at various time onsets with respect to flanks that were sinusoidally-modulated at 1 Hz. The target and flank separation
2717
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
A
B
1
4
2
5
3
3
4
5
1 2
Threshold Elevation [ log (THwithflanks/THwithoutflanks)]
was 3k and the orientation of the target and flanks was vertical. The flanks were modulated in +sine phase and sine phase (see temporal profile in Fig. 1A). The target was presented briefly and the onset time varied during the temporal cycle of the flanks. The results from three subjects were averaged, collapsed across spatial phase configuration (i.e. +sine and sine) and plotted according to the relative onset of the test stimulus, as indicated by the numbers in Fig. 1A. The results show that although slightly more facilitation occurred when the target and flanks were in the first epoch of the same spatial phase, a comparable amount of facilitation occurred even in the spatial out-of-phase configuration (i.e. epochs 4 & 5). The facilitation effect found in the out-of-phase configuration (epochs 4 & 5) is likely to have been carried over from when the high contrast flanks were in the in-phase configuration because there is ample evidence that facilitation is minimal or non-existent when static flank and test stimuli are spatially out of phase (Solomon et al., 1999; Williams & Hess, 1998; but see Zenger & Sagi, 1996). However, if there is facilitation in the sine modulation it might suggest that facilitation can occur prior to the initiation of the flanks. In order to assess this particular issue regarding the importance of the order of presentation (i.e. flank vs. test), the results from this experiment were reanalyzed separately for +sine and sine phase modulation of the flanks and these results are shown separately in Fig. 2B along with predictions for the two current models; the single channel masking model and the multi-channel, long-range conduction model in which facilitatory impulses are assumed to travel from flanks to target via long-range connections. The predictions for the single channel masking model are based on three established features, first that any masking effects are restricted to a time window of 50 ms (Georgeson & Georgeson, 1987) and therefore to obtain a prediction the temporal modulation of the flanks have been convolved with a Gaussian of this width. Second, simultaneous flank facilitation does not occur when the flanks and test are 180° out of phase, therefore facilitation is only expected for one half cycle of the flank modulation (solid curve for +sine modulation of the flanks and dashed curve for sine modulation of the flanks). Third, facilitation saturates at relatively low contrasts of the flanks (Woods et al., 2002). The long-range model predictions are based on the aforementioned phase dependency and contrast saturation of facilitation. Furthermore, assuming a conduction velocity of 0.1 m/s (i.e. long-range
connections, a delay is expected that will depend on the test to flank separation (1.5° in this case). As before (Fig. 1), a slightly stronger collinear facilitation was found for the first epoch, however facilitation was present at all time epochs tested and similar for the +sine (Fig. 2C: filled squares) and sine (Fig. 2C: open squares) modulations of the flanks. This suggests two things, first a long-lasting facilitatory field is produced from the in-phase flanks and second comparable facilitation can occur when the target precedes the flanks. Neither of these effects would be expected from either single channel masking or multi-channel interactions-based solely on the conduction velocity of long-range connections assumed to travel in the direction from flanks to target (Fig. 2B). 3.2. Experiment 2: the order of presentation of the test target and flanks The order of presentation of the test target and flanks were investigated further in this experiment. The amount of facilitation was plotted against the SOA between flanks and test in Fig. 3 for each subject. An asymmetrical distribution was found. To derive the location of peak facilitation the results were fitted by a Gaussian function and the derived peak location measures are plotted in Fig. 4. Best facilitation occurred when the test target preceded the flanks (30 ms for average data at 3k separation). When the SOA between test target and flanks was larger than 150 ms, no significant facilitation was found. Second, as expected, the amount of collinear facilitation reduced as the target–flank separation increased and the position of peak facilitation scales with the test to flank distance. Fig. 4 shows a comparison between the position of peak facilitation as a function of the test to flank distance in units of mm of cortex (see Eq. 3) and the results of Cass and Spehar (2005). If one assumes that the change in dynamics with test to flank distance is completely accounted for by conduction time, the derived velocities, although showing some variability across subjects (0.18–1.67 m/s), the average result of 0.32 m/s is not that different from that derived by Cass and Spehar of 0.23 m/s. However, the one fundamental difference between the present results and those of Cass and Spehar (2005) is that the peak facilitation occurred when the test target preceded the flanks (hence the negative values in Fig. 4 for our results) whereas the obvious interpretation of Cass and Spehars’ model require it to occur when flanks precede the test stimulus (hence the positive values in Fig. 4 for the Cass
0.1 0.0 -0.1 -0.2 LHY PCH WW AVG
-0.3 -0.4 1
2
3
4
5
Epoch #
Fig. 1. Results for experiment 1. (A) The temporal profile of the stimuli. The solid wave represents the flanks modulated in +sine phase and the dash wave represents the flanks were modulated in sine phase. The marked epochs represent the onset time of the target. The number denoting each epoch indicates it has the same temporal and spatial relationship for the +sine and sine configuration. (B) Experimental results for three subjects and the average data is shown in red and the error bars represent standard errors among subjects. (For interpretation of the references to color in this figure legend the reader is referred to the web version of the paper.)
2718
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
B 0.2
0.2
4‘
2
5‘
3
3‘
4
1‘
5
Threshold Elevation
1
Prediction: Single Channel Model
1
0.1
Threshold Elevation
A
2 1‘
2‘
0.0 -0.1 -0.2 -0.3
2
0
2‘
0.0 -0.1 -0.2 -0.3
200
400
600
0
800 1000
200
time (ms)
Threshold Elevation
C 0.2
0.2
LHY
3
-0.4
-0.4
2‘
Prediction: Long-Range Model
0.1
0.2
PCH
400
600
0.2
0.1
0.1
0.1
0.0
0.0
0.0
0.0
-0.1
-0.1
-0.1
-0.1
-0.2
-0.2
-0.2
-0.2
-0.3
-0.3
-0.3
-0.3
-0.4
-0.4 1
2
3
Epoch #
4
5
1
2
3
Epoch
4
#
5
AVG
WW
0.1
-0.4
800 1000
time (ms)
-0.4 1
2
3
Epoch
4
#
5
1
2
3
Epoch
4
5
#
Fig. 2. Results for experiment 1. (A) The temporal profile of the stimuli. (B) Predictions based on the two current models. (C) The collinear facilitation was plotted against the epoch number as shown in (A) for each individual as well as averaged data. The results for the +sine modulation are shown as filled squares while those for the sine modulation are shown as open squares.
and Spehar results but see general discussion for a more general interpretation of this model based on Cass’s earlier work (Cass, 2005) and that of Cass, Spehar, Alais, and Arrighi (2005)). The present results showing essentially negative values are inconsistent with an explanation based on the conduction of facilitatory signals from flanks to test stimulus (Cass & Spehar, 2005).
SOAs when the target succeeds the flanks (Fig. 5, right column). This suggests that the facilitation is maximal at or around when the test and flanks coincide in time, similar to the conclusion arrived at in experiment 2. There is no indication of facilitation being maximal at some time after the flanks have been presented. 4. General discussion
3.3. Experiment 3: temporal summation of test target The results of measuring the temporal summation of flank facilitation are shown in Fig. 5. The stimulus arrangements are illustrated at the top of each column. One of the two subthreshold test stimuli in this typical two-pulse paradigm, is either aligned in time with the flanks (top, middle) or presented before (top, left) or after (top, right) the flanks. The SOA between two test stimuli was varied on a temporal grid of ±200 ms. The curves in each of the columns of Fig. 5 show a comparison for two subjects between the temporal summation associated with the detection of an isolated Gabor compared with that in the presence of the facilitatory flanks for each of the three previously described temporal-alignments (see top illustrations). The temporal summation function (i.e. the autocorrelation function of the filters response) for the isolated Gabor has the typical unipolar form, consistent with lowpass dynamics. The presence of the flanks does not substantially change the form of the summation apart from imparting an overall facilitation. The time points where this facilitation is statistically significant have been marked with an asterisk. What is evident, particularly in the results of subject PCH, is that the facilitation is symmetrically arranged when the test is temporally coincident with the flanks (Fig. 5, middle column), it is asymmetrically displaced to more positive SOAs when the test precedes the flanks (Fig. 5, left column) and asymmetrically displaced to more negative
The experiments reported here have revealed that a complex set of dynamics underlies collinear facilitation. The dynamics are certainly not as simple as either of the two current models of collinear facilitation would predict. First, the effects of the flanks are not just instantaneous, the results of experiment 1 demonstrate that they can be long lasting, suggesting a sustained facilitatory response, one that can be built up over time. Secondly, facilitation is maximal at or before flank presentation, not following flank presentation (experiments 1, 2 and 3), suggesting a rapidly initiated interaction across different parts of the visual field for mechanisms of the same orientation forming a global alignment. 4.1. Underlying mechanisms One hypothesis assumes that collinear facilitation is a special example of subthreshold summation within a single, elongated V1 neuron detecting both flank and test stimuli (Solomon et al., 1999; Williams & Hess, 1998). In such a case, one would expect the dynamics to be simple and predictable on the basis of what we currently know about simultaneous masking (Georgeson & Georgeson, 1987). For example, facilitation would be expected to follow the in-phase modulation of the flanks in experiment 1 (as per prediction in Fig. 1B), to be limited to a 50 ms time window, centered on the flank presentation
2719
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
Threshold Elevation
-200
200
0
0.2
0.2
0.1
0.1
0.0
0.0
-0.1
-0.1
-0.2
-0.3
-0.2 3λ 4.5 λ 6 λ
PCH
-200
-100
0
100
YN
-200
200
-100
0
100
200
Time lag between target and flanks (ms)
0.2
0.2
0.1
0.1
0.0
0.0
-0.1
-0.1
-0.2
3λ 4.5 λ 6λ
-0.3
Time lag between target and flanks (ms)
Threshold Elevation
time
-0.2 3λ 4.5 λ 6λ
-0.3 PB
-200
-100
0
100
200
Time lag between target and flanks (ms)
3λ 4.5 λ 6λ
-0.3 AVG
-200
-100
0
100
200
Time lag between target and flanks (ms)
Fig. 3. (Top) The spatio-temporal profile of the stimuli. (Bottom) Experimental results for each subject and the average data for three subjects for three different tests to flank distances. The data have been fitted by a Gaussian function and the peak values so derived are plotted in Fig. 4.
(Georgeson & Georgeson, 1987) and to have dynamics that are invariant with test to flank distance. These model predictions in experiment 1 were not supported, the dynamics were such that peak facilitation occurred before presentation of the flanks and the effects of facilitation were long lasting. Furthermore, the temporal summation for the target was less than 100 ms (experiment 3) but the temporal summation between target and the flanks (experiment 2) is longer than 150 ms, not supporting the subthreshold summation hypothesis. Another hypothesis states that collinear facilitation is the result of interactions between neurons in different locations within V1, via slowly conducting, long-range connections. This transmission of facilitatory activity is assumed to be from flanks to test, and hence stronger facilitation would be expected to occur when flanks precede the test by a period predicted by the flank to test distance and the slow conduction velocities of this fiber system (i.e. 0.1–0.2 m/s). However, the results of experiments 1, 2 and 3 that show maximal facilitation occurs at or prior to presentation of the flanks is incompatible with this explanation.
In the original experiment of Cass and Spehar (2005) since facilitation was maximal at longer presentation durations one is led to assume that this would be equivalent to a positive SOA dependence consistent with the transmission time from flanks to test. However, we show here, using a variety of different techniques that such SOA dependence does not occur. An alternative explanation for the duration dependency of flank facilitation could be couched in terms of temporal integration, an important issue not considered within Cass and Spehars’ model. If the strength of neural integration varies with test to flank distance (Cass & Alais, 2006), this could explain their results. The most challenging finding of the present study is that facilitation is maximal at or prior to flank presentation. A similar finding was recently reported by Cass and Alais (2006). Here too an explanation was advanced in terms of the much slower dynamics associated with neural integration (Gawne, Kjaer, & Richmond, 1996) for low contrast test stimuli compared to that of the higher contrast flanks. The neural effects set up by the initial presentation of the test stimulus may last until the later presentation of the flanks (see Polat et al.,
2720
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
Position of peak facilitation
2007 for discussion). Alternatively, it may be the consequence of an anticipatory response from higher visual areas. Whatever the explanation it suggests a rapid orientation specific interaction involving a relatively local patch of cortex. Feedback connections have been described from V2 to V1 that satisfy these criteria. Feedback connections from V2 to V1 result in a rapid (Girard, Hupe, & Bullier, 2001) excitation of cells in like-orientation columns in local regions of V1 and these could play a role in enhancing the detectability of collinear image features (Shmuel, Korman, Sterkin, Harel, Ullman, Malach, & Grinvald, 2005).
Cass & Spehar (2005)
100
PCH YN PB AVG
50
0.23 m/s Flanks precede Target
0.18 m/s 1.67 m/s 0
0.32 m/s Target precedes Flanks
0.31 m/s -50
4.2. Comparison with previous studies We do not support the conclusions of an earlier study by Polat and Sagi (2006) where facilitation was only found when the flanks preceded the target, or were presented simultaneously, but not when the target preceded the flanks. However, it should be pointed out that there are a number of possible experimental differences between our studies and theirs. Our experiment 1 design is similar (sustained presentation of flanks) but not identical (static verses dynamic) to their experiment 2 in which the target was presented at the beginning, the middle and the end of flank presentation. They argued that collinear facilitation only occurs when the target is presented sometime after the flanks. Our data from experiment 1 suggest that collinear facilitation occurs at all temporal phases. Our
-100 5
10
15
20
25
30
Separation (mm in cortex) Fig. 4. Estimated conduction velocities. The peak facilitation positions are plotted as a function of target–flank separation for each subject. The horizontal axis represents the cortical distance in mm between target and flanks (from center to center). The results of Cass and Spehar (2005) are included for reference. In order to estimate the striate distance associated with a given visual angle (E) relative to the R 17:3 dE ¼ 17:3 lnðE þ 0:75Þ (3) where y fovea, the equation below was used y ¼ Eþ0:75 is the mm across retinotopic striate surface and E is the degrees of visual angle (Cass & Spehar, 2005; Horton & Hoyt, 1991).
<==SO A==||
<==SO A==||
-100
0
*
*
100
*
*
*
<==SO A==||
-100
time *
*
100
0
***
*
*
3
*
-100
time
**
**
**
8 7
0.01
PCH -50ms Without Flanks With Flanks -200
-100
9 8 7
0
100
-200
200
5
4
4
3
3
Threshold
Threshold
5
2
0.01 9 8 7 6 5
-100
9 8 7
0
100
*
*
*
7 6 5
-200
-100
0
100
200
SOA between test targets (ms)
0
100
200
*
*
4
3
2
9 8
-100
SOA between test targets (ms) 5
*
2
0.01
0.01
CTC -50ms Without Flanks With Flanks
*
*
PCH 50ms Without Flanks With Flanks -200
200
SOA between test targets (ms)
SOA between test targets (ms)
time
2
0.01
PCH 0 ms Without Flanks With Flanks
Threshold
9
2
Threshold
Threshold
Threshold
0.01
*
100
3
3
2
0
9 8
CTC 0 ms
7 6
Without Flanks With Flanks -200
-100
5
0
100
200
SOA between test targets (ms)
CTC 50ms Without Flanks With Flanks -200
-100
0
100
200
SOA between test targets (ms)
Fig. 5. The temporal arrangement of the stimuli and threshold for two subjects in the absence and presence of the flanks. The arrangement above shows the time relationship between one of the two test targets and the flanks. The other test target in the two-pulse paradigm was displayed on a grid of ±200 ms.
2721
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
experiment 2 is similar though not identical to their experiment 1 in which flanks were presented transiently. Unlike Polat and Sagi (2006) we show that maximal facilitation occurs when the target precedes the flanks or when it coincides in time with the presentation of the flanks. In a recent paper alluded to above, Cass and Alais (2006) presented data using a rotating flank paradigm in which facilitatory effects were found both when the test followed the flanks and when the test preceded the flanks in time. They interpret their results in terms nerve conduction from flanks to target in the former and in terms of neural integration at low contrasts in the latter. We agree with them concerning the explanation for facilitation when the test precedes the flanks in time but disagree with them in the case when the test follows the flanks in time. It is worth noting that their interpretation concerning nervous conduction assumes a linear dependence on orientation that is invariant with test to flank distance over a range of ±50°. There is good reason to believe that this is not the case. Fig. 6 shows the results where the orientation of temporally coincident flanks is varied in a way similar to Cass and Alais (2006) for three different test to flank distances. Not only does the slope of the relationship between facilitation and test to flank orientation change with test to flank distance but also, at close distances, facilitation is replaced by inhibition. The intricacies of these orientation-dependent effects would need to be taken into account before any firm conclusions can be arrived at concerning the dynamics. We took the more direct approach and explicitly varied the timing of test and flank stimuli by varying the SOA of aligned targets of constant duration. We found that facilitation either coincided with the presentation of the flanks or preceded them. We feel that instead of two separate explanations for these two findings where facilitation is observed (test precedes flanks and flanks pre-
4.3. A more general long-range model? The long-range conduction model discussed so far was that of Cass and Spehar (2005) in which it is assumed (though never explicitly stated) that the facilitatory transmission travels from flanks to target via long-range fibers. Our findings are not consistent with this. Cass’s original work (Cass, 2005 and later Cass et al., 2005) however did show that when SOA was varied between flanks and target, greatest facilitation occurs when the target precedes the flanks, just as we show here using a variety of approaches. This finding compromises the most obvious interpretation of the model proposed by Cass and Spehar (2005) unless one assumes that the facilitatory impulses can travel in either direction (i.e. from flanks to target and from target to flanks). Cass realized this and proposed this more general interpretation (Cass, 2005 and later Cass et al., 2005). There are two objections one could raise to this extended interpretation. First, the task is to detect the target and not the flanks and so it is difficult to understand why the presentation of the target itself would initiate facilitation of its own detection. Second, one should see a comparable amount of facilitation when the flanks precede the target which we do not observe. In Cass’s original results (also see Cass et al., 2005), a large degree of suppression occurred when the flanks preceded the target and he hypothesized that this may have masked the expected facilitation under these conditions. He proposed either backward masking by the
λ
0.4
S Shape 45°
0.2
HT 1000ms S-Shape
0.0
2λ 4λ 6λ
-0.2 0
15
30
45
60
75
λ
0.4
Threshold Elevation
0.6
Threshold Elevation
cedes test), on grounds of parsimony a single explanation can be couched simply in terms of the feedforward effects of the neural integration associated with the processing of test and flankers, combined with fast feedback interactions between test and flanker locations.
S Shape 45°
0.2
PCH 1000ms S-Shape 0.0 2λ 4λ 6λ
-0.2
90
0
Orientation (Degrees)
λ
0.4
C Shape 45 °
0.2
HT 1000ms C-Shape 0.0 2λ 4λ 6λ
-0.2 15
30
45
60
75
Orientation (Degrees)
30
45
60
75
90
90
λ
0.4
Threshold Elevation
Threshold Elevation
0.6
0
15
Orientation (Degrees)
C Shape 45 °
0.2
PCH 1000ms C-Shape 0.0
2λ 4λ 6λ
-0.2 0
15
30
45
60
75
90
Orientation (Degrees)
Fig. 6. Threshold elevation as a function of the orientation of the flanks. Two different flank orientation configurations are examined: one in which the relative orientation of the flanks is the same (S shape), and one in which it is opposite (C Shape). Results for two subjects are shown each at three different test to flank distances.
2722
P.-C. Huang, R.F. Hess / Vision Research 48 (2008) 2715–2722
B
0.2
0.1
Threshold Elevation
Threshold Elevation
A
0.0
-0.1
-0.2
-0.3
3 λ (2 cpd) 3 λ (6 cpd)
PCH -200
-100
0
100
200
Time lag between target and flanks (ms)
0.2 EXP 2 Replication of Cass' Exp 0.1
0.0
-0.1
-0.2
-0.3
PCH -200
-100
0
100
200
Time lag between target and flanks (ms)
Fig. 7. (A) The collinear facilitation effect as a function of the SOA between target and flanks at different spatial frequencies. The black circles represent 2c/d (replotted from Fig. 3) and gray squares represent 6c/d. The results showed similar asymmetrical distribution for both spatial frequencies. (B) The threshold elevation as a function of SOA between target and flanks in the presence of noise (replotted from Fig. 3) and absence of noise (replication of Cass’ (2005) experiment) at the position of the flanks. Both showed similar results.
noise mask or motion artifacts as the basis for this suppression. We do not find that the particular spatial frequency (2c/d) that we used is crucial for our conclusions. Previously we have used lower spatial frequencies (i.e. 0.75c/d) and others have often used higher spatial frequencies. All previous studies have demonstrated robust facilitation at all of these different scales. To be completely sure, we have now replicated our main findings (i.e. no facilitation and minimal suppression when the flanks precede the test target; Fig. 3) at a higher spatial frequency (6c/d) and for a different arrangement of noise masks (i.e. noise masks that cover the whole display and occur only at the very beginning and end of the trial) to better compare with the experimental conditions used by Cass (2005) and Cass et al. (2005) and the results are shown in Fig. 7. We did not observe any measurable suppression when the flanks preceded the target and hence our finding of no facilitation under these conditions argues against the notion of there being facilitatory transmission from flanks to target. The results suggest that the dynamics of facilitation are fast but sustained. We propose two underlying mechanisms, a rapid signal to initiate facilitation across large retinal distances, based on feedback from higher centers and a sustained facilitative response based on the temporal integration of locally-responsive, lower-level mechanisms. References Cass, J. (2005). Spatial and temporal determinants of long-range contrast facilitation and supperssion. University of New South Wales, Kensington. PhD Thesis. Cass, J., & Alais, D. (2006). The mechanisms of collinear integration. Journal of Vision, 6(9), 915–922. Cass, J., Spehar, B., Alais, D., & Arrighi, R. (2005). Spatial and temporal determinants of contrast facilitation and suppression [abstract]. Journal of Vision, 5(8), 459a. Cass, J. R., & Spehar, B. (2005). Dynamics of collinear contrast facilitation are consistent with long-range horizontal striate transmission. Vision Research, 45(21), 2728–2739. Gawne, T. J., Kjaer, T. W., & Richmond, B. J. (1996). Latency: Another potential code for feature binding in striate cortex. Journal of Neurophysiology, 76(2), 1356–1360. Georgeson, M. A., & Georgeson, J. M. (1987). Facilitation and masking of briefly presented gratings: Time-course and contrast dependence. Vision Research, 27(3), 369–379.
Girard, P., Hupe, J. M., & Bullier, J. (2001). Feedforward and feedback connections between areas v1 and v2 of the monkey have similar rapid conduction velocities. Journal of Neurophysiology, 85(3), 1328–1331. Hirsch, J. A., & Gilbert, C. D. (1991). Synaptic physiology of horizontal connections in the cat’s visual cortex. Journal of Neuroscience, 11(6), 1800–1809. Horton, J. C., & Hoyt, W. F. (1991). The representation of the visual field in human striate cortex. A revision of the classic holmes map. Archives of Ophthalmology, 109(6), 816–824. Huang, P. C., & Hess, R. F. (2007). Collinear facilitation: Effect of additive and multiplicative external noise. Vision Research, 47(24), 3108–3119. Huang, P. C., Hess, R. F., & Dakin, S. C. (2006). Flank facilitation and contour integration: Different sites. Vision Research, 46(21), 3699–3706. Huang, P. C., Mullen, K. T., & Hess, R. F. (2007). Collinear facilitation in color vision. Journal of Vision, 7(11): 6, 1–14. Polat, U. (1999). Functional architecture of long-range perceptual interactions. Spatial Vision, 12(2), 143–162. Polat, U., & Sagi, D. (1993). Lateral interactions between spatial channels: Suppression and facilitation revealed by lateral masking experiments. Vision Research, 33(7), 993–999. Polat, U., & Sagi, D. (2006). Temporal asymmetry of collinear lateral interactions. Vision Research, 46(6-7), 953–960. Polat, U., Sterkin, A., & Yehezkel, O. (2007). Spatio-temporal low-level neural networks account for visual masking. Advances in Cognitive Psychology, 3(1–2), 153–165. Shmuel, A., Korman, M., Sterkin, A., Harel, M., Ullman, S., Malach, R., et al. (2005). Retinotopic axis specificity and selective clustering of feedback projections from v2 to v1 in the owl monkey. Journal of Neuroscience, 25(8), 2117–2131. Solomon, J. A., & Morgan, M. J. (2000). Facilitation from collinear flanks is cancelled by non-collinear flanks. Vision Research, 40(3), 279–286. Solomon, J. A., Watson, A. B., & Morgan, M. J. (1999). Transducer model produces facilitation from opposite-sign flanks. Vision Research, 39(5), 987–992. Ts’o, D. Y., Gilbert, C. D., & Wiesel, T. N. (1986). Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis. Journal of Neuroscience, 6(4), 1160–1170. Weliky, M., Kandler, K., Fitzpatrick, D., & Katz, L. C. (1995). Patterns of excitation and inhibition evoked by horizontal connections in visual cortex share a common relationship to orientation columns. Neuron, 15(3), 541–552. Wichmann, F. A., & Hill, N. J. (2001a). The psychometric function: I. Fitting, sampling, and goodness of fit. Perception & Psychophysics, 63(8), 1293–1313. Wichmann, F. A., & Hill, N. J. (2001b). The psychometric function: II. Bootstrapbased confidence intervals and sampling. Perception & Psychophysics, 63(8), 1314–1329. Williams, C. B., & Hess, R. F. (1998). Relationship between facilitation at threshold and suprathreshold contour integration. Journal of the Optical Society of America A, Optics, Image Science, and Vision, 15(8), 2046–2051. Woods, R. L., Nugent, A. K., & Peli, E. (2002). Lateral interactions: Size does matter. Vision Research, 42(6), 733–745. Zenger, B., & Sagi, D. (1996). Isolating excitatory and inhibitory nonlinear spatial interactions involved in contrast detection. Vision Research, 36(16), 2497–2513.