1. INTRODUCTION

It is well established that simple reaction times (RTs) to visual stimuli reflect early visual processing mechanisms. The input layers of the primary visual cortex are dominated by the extraction of contrast and there is a long-established and substantial literature in support of the idea that visual signals operate along anatomically and physiologically distinct pathways named parvocellular (P), magnocellular (M), and koniocellular (K) after the layers of the lateral geniculate nucleus (LGN) through which they project. Lee [1] offers an excellent review of this concept. Briefly, ${\rm{M}}$ neurons specialize in extracting luminance contrast and mediate flicker. ${\rm{P}}$ neurons have poor sensitivity to achromatic contrast but are responsible for processing red–green color and high spatial frequency information. ${\rm{K}}$ neurones subserve the short wavelength pathway. In fact, retinal neurons are best identified in physiological experiments by their response to luminance contrast. Of relevance to the present study are the observations that ${\rm{M}}$ neurons respond transiently and saturate at relatively low contrast whereas ${\rm{P}}$ neurons respond linearly and exhibit low sensitivity [2,3]. The main objective of this work is to show that comparing RTs generated by fast onsets with those from slow or ramped onsets can reveal the activity of transient and sustained activity.

From the early days of measuring RTs, their link with contrast has been particularly important in confirming the existence of these pathways in vivo and revealing how their characteristics manifest in suprathreshold visual experiments [4]. As indicated in that study, the importance of suprathreshold studies cannot be underestimated because, while visual mechanisms are often defined by their threshold behavior, this is far-removed from so-called real-world vision. In general, there is an exponential increase in the RT as the stimulus intensity is reduced [5]. In the contrast domain Piéron’s law is expressed as

This relationship is extremely robust across a wide range of spatial frequency, luminance, and contrast, as reported by Plainis and Murray [6]. However, under some conditions, a conspicuous inflexion is evident at around 10% contrast. See Harwerth and Levi [7] for an early interpretation of this phenomenon. As the contrast is increased from close-to-threshold values, the RT gradually decreases, as expected. However, above approximately 10% contrast, the time course of the function changes, indicating an increase in the underlying sensitivity for the range of higher contrasts. As detailed below, this observation is exactly consistent with detection of the stimulus being mediated by a high sensitivity transient mechanism at low contrasts and a poorer sensitivity, sustained mechanism at higher contrasts.

Evidence for the way that RTs reflect the properties of the afferent neurons of the early visual pathway was presented for a wide range of spatial frequency and contrast by Murray and Plainis [8], who plotted RT versus the reciprocal of contrast. They showed that the slope of the corresponding RT versus 1/C relationship could be interpreted as a Naka–Rushton function. By plotting RTs against 1/C for low spatial frequencies, they were able to accentuate the occurrence of the bimodality and convert the slopes of the resultant two functions to different values of contrast gain, which matched those published for the ${\rm{P}}$ and ${\rm{M}}$ pathways based on single unit studies [9].

Most RT experiments use abruptly presented stimuli, but Tolhurst [10] showed that this introduces a bias in favor of transientness; that is, the sudden appearance of the stimulus activates transient response mechanisms. He demonstrated that when a stimulus is mediated by a transient mechanism the near-threshold RT histograms cluster around the timing of the onset and offset in a probabilistic manner. A slow onset stimulus, on the other hand, produces near-threshold RT histograms whose profiles reflect the duration of the stimulus. In other words, the temporal profile of the stimulus is key. If a stimulus is mediated wholly by sustained mechanisms then the RT versus contrast function is monophasic regardless of whether the onset is fast or slow.

In pursuit of the notion that suprathreshold contrast is important to understand real-world vision, Parry [11] developed the ideas in a novel way by proposing that slow and fast RTs should be compared for a wide range of contrasts. He argued that using a pure color stimulus which, according to general thinking, is wholly mediated by sustained mechanisms (i.e., it contains no transients) should generate a monophasic RT versus the contrast function for both fast and slow onsets. He tested this idea by recording RTs to a low spatial frequency, chromatic isoluminant grating. The data confirmed that the bimodality was absent for the “pure” chromatic grating; there was no discontinuity for either ramped or fast onset RT contrast functions.

Parry [12] reported that, when the difference between slow and fast RTs was calculated for low spatial frequency achromatic gratings, the inherent high and low gain sections of the contrast range emerged. For the low contrast, transient-mediated, high-gain region of the RT versus contrast function, the fast and slow onset RT versus the contrast functions diverged. It was presumed that the fast onset RTs tapped the transient mechanism but that these were not activated by the slow onset stimuli. At high contrasts, the slow and fast RTs were parallel because both forms of stimulation were controlled by the sustained mechanism. This suggests that the method of subtracting slow and fast RTs should be highly sensitive to transient intrusions.

All the experiments reported above were conducted using sinusoidal gratings. The contrast was spatially modulated and the chromatic stimulus was defined as the point of equilibrium or isoluminant point between two superimposed red and green gratings presented in anti-phase. The technique of heterochromatic flicker photometry was used to determine isoluminance.

In this paper, these ideas are developed in two ways. First, we wanted to establish that the same principles can be applied when the stimuli are zero spatial frequency and temporally modulated. Second, to determine the extent to which the technique is sensitive to small changes in the relative strengths of transient and sustained components, it is necessary to use a temporally modulated stimulus that systematically varies between being purely chromatic and a purely achromatic. This can be achieved by varying the ratio between the redness and greenness in the stimulus.

By manipulating the red–green ratio in small increments, we are able to introduce non-chromatic (transient) activation into the stimulus in a controlled way. Here, manipulating the red–green ratio is merely a surrogate for other sources of contamination. One such example arises in our studies of cone isolation using silent substitution where variations in cone opsins due to polymorphisms are present. These may be sufficient to introduce errors in the measurements that rely on accurate computations to achieve silent substitution. This possibility was entertained in Huchzermeyer and Kremers [13], who investigated temporal contrast sensitivity functions for ${\rm{S}}$-cone and rod-isolating stimuli. They presented a table showing how theoretical shifts in opsin peak wavelengths introduced relatively low residual modulations in “silenced” photoreceptors. To summarize, the purpose of this work is to present a technique that could be used in any experimental setup where inherent transient components are likely to arise.

2. METHODS

A. Stimulus

Simple RTs were recorded to stimuli presented in a ColorDome ganzfeld (Diagnosys LLC, Lowell, MA, USA). This contains four classes of LEDs; however, for these studies, only red (${\lambda _{{\max}}}{:}\;{{635}}\;{\rm{nm}}$, half-width at half-height: 10 nm) and green (${\lambda _{{\max}}}\;{{514}}\;{{\pm}}\;{{20}}\;{\rm{nm}}$) LEDs were used. To be compatible with our previous and future cone-isolating studies on this device [14], the stimulus was a binocularly viewed quarter annulus, presented in the upper right visual field, with an external diameter of 10.8 cm and an internal diameter 2.65 cm, as shown in Fig. 1. Viewed at a distance of  30.6 cm, these equated to an external and internal retinal subtense of 10° and 2.5° radius. A small hole (0.2 mm) in the center of the black mask provided a fixation target. A chin rest was provided. The ganzfeld was controlled by an Espion ${{\rm{E}}^2}$ electrophysiology system (Diagnosys LLC, Lowell, MA, USA) connected to a PC running Windows XP. The LED luminance was calibrated using a spectrophotometer (PR650, Photo Research Inc., Chatsworth, CA, USA).

 

Fig. 1. Stimulus configuration. (a) Layout of the mask attached to the ganzfeld so that the stimulus was a quarter annulus of internal and external radii of 2.5° and 10°. (b) Temporal profile of the fast (upper graph) and slow (lower graph) stimuli depicting a green onset on a yellow background with red–green ratio of 0.5 and contrast 0.5 (6 dB). The first 1000 ms are illustrated.

Download Full Size | PDF

Stimuli were modulated in two ways: with a square-wave pulse of 380 ms duration (“fast” presentations), or with raised cosinusoidal rising and falling edges, so that the stimulus slowly incremented for 190 ms, remained at the set contrast for 190 ms, and then decayed again over 190 ms (“slow” presentations). Modulation was achieved with ${\lt}{{1}}\;{\rm{ms}}$ resolution by continually reading the system clock and setting the LED instantaneous luminance desired luminance for that post onset latency, based on a lookup table. The mean luminance was ${{80}}\;{\rm{cd}}\cdot{{\rm{m}}^{- 2}}$. Red and green primaries were combined in various red–green mean luminance ratios (RGRs); thus,

When RGR = 0.5, the stimulus was photometrically isoluminant, and made up of ${\rm{red}} = {{40}}\;{\rm{cd}}\cdot{{\rm{m}}^{- 2}}$ and ${\rm{green}} = {{40}}\;{\rm{cd}}\cdot{{\rm{m}}^{- 2}}$. The chromatic stimulus incremented in the green direction. When RGR = 0, the stimulus was a green luminance increment on a green background, and when RGR = 1, it was a red decrement on a red background. In control studies, we established that there was no effect if this strategy were reversed, and RTs were recorded to chromatic increments in the red direction when the extremes of RGR would be a green decrement and a red increment. We also investigated the effects of varying the red versus green contrast rather than using the more traditional luminance ratio. The main reason for not using this approach was to ensure our experimental manipulation was compatible with previous work [11]. Note that the contrast was defined as the Weber fraction [(Lmax-Lo)/(Lo)] and converted to dB where ${\rm{dB}} = - {{20}.\rm{LogC}}$. Thus, dB equates to the linear and logarithmic contrast in the following way: $0 \,{\rm{dB}} = 1 \,({\rm{linear}}) = 0 \,({\rm{log}})$; $6 \,{\rm{dB}} = 0.501, -0.3$; $12 \,{\rm{dB}} = 0.251, -0.6$; $18 \,{\rm{dB}} = 0.126, -0.9$ and so on. A 1 dB step is a 0.05 log unit change in contrast.

B. RTs

RTs were recorded with 2 ms accuracy to the onset of the stimulus, with the subject instructed to respond as quickly as possible as soon as a stimulus was detected. The subject responded by pressing the space bar, which triggered the software to instantaneously read the system clock and record the post-onset time. Once an RT had been recorded, there was an interstimulus interval of random duration, with an equal probability of falling anywhere between 1000 and 3000 ms. If no response was recorded, the timeout was 5000 ms, after which the next stimulus was presented. No sound cues were provided for presentation or response.

C. Isoluminance

The isoluminance of the red–green modulations was determined using heterochromatic flicker photometry, with the stimulus reversing sinusoidally between the redder and greener component at 16 Hz, with the contrast initially at 22 dB. The subject made repeated (at least five) settings by method of adjustment, identifying the RGR when the flicker was minimal. If the range of RGR with no flicker was too broad, then the contrast was increased; if the flicker was always well above threshold, the contrast was decreased. The adjustment of the red and green lights was made using the up/down arrow keys on the PC keyboard. The errors in these settings were extremely small, with 95% CIs of approximately 0.05.

D. Subjects

We were the subjects (males, aged 67 and 71) plus a third subject (male, aged 55) who was naive to the purpose of the study but was experienced in such RT studies. All three subjects had normal color vision according to the Ishihara color vision test, the Hardy–Rand–Rittler (HRR) Standard Pseudoisochromatic Test, 4th edition, and the Nagel anomaloscope.

E. Procedure

1. Chromatic and Achromatic RTs versus Contrast

After determining isoluminance by heterochromatic flicker photometry (HFP), a series of RTs were recorded to various levels of contrast. Each run was performed at a single contrast level, and comprised 32 fast and 32 slow stimuli, randomly mixed, which took approximately 3 minutes. Analysis was performed off-line using an Excel spreadsheet (Microsoft Corp., Redmond, WA, USA). RTs of less than 150 ms or greater than 1000 ms were excluded from the analysis. Mean and median RTs were calculated from the remaining data. To estimate and control for the impact of interstimulus interval (ISI), RT versus ISI was calculated, and a correction made for the slope of the resulting function, correcting all RTs to a theoretical ISI of 2000 ms.

 

Fig. 2. Mean RT (${{\pm 95}}\%$ CIs) versus achromatic contrast for three observers using fast and slow onset stimuli. There is a conspicuous divergence between the two functions with the difference between the two reaching a maximum at low contrasts.

Download Full Size | PDF

 

Fig. 3. Mean RT (${{\pm 95}}\%$ CIs) versus chromatic contrast for the three observers using fast and slow onsets. Overall, the functions are approximately parallel.

Download Full Size | PDF

For each contrast run, the experiment started at high contrast (6 dB for S1 and S2, 0 dB for S3) and each subsequent run was 3 dB lower in contrast, until the detection rate fell below 50%. Two contrast runs were performed, one at isoluminance and one with RGR = 0 (achromatic). A short low-contrast run was also performed with RGR = 1. Separate functions were derived from fast and slow presentations and the fast–slow difference was computed and plotted as a function of contrast.

2. Low Contrast RTs versus RGR

Using the RT versus C functions recorded in the first experiment, the lowest contrast to still produce detectability better than 90% across the range of RGR was identified and this contrast was used to record RTs to a range of red–green ratios from 0 to 1, in steps of 0.1 but with higher resolution around isoluminance. Typically, this entailed 13 runs. Other than the fact that contrast was constant and RGR varied, the analysis was as in the first experiment. The slow–fast RT difference (RTD) was computed and plotted as a function of RGR.

3. RESULTS

A. Fast versus Slow Onset RTs; Chromatic and Achromatic Contrasts

In Fig. 2, we present data from the three observers for achromatic temporal modulation. There is a constant delay of around 50 ms for the slow onset RTs over the high contrast region. As the contrast is decreased below a critical value (around 12 dB), the delay increases markedly, and the two functions diverge. Here and throughout, we have used the arithmetic mean RT. The data are essentially the same if the median is used.

All observers show this characteristic relationship between fast and slow onset RTs. The abrupt stimulus onset activates transient detectors. Hence, for high contrasts, RTs are relatively short. As the contrast is decreased, the slow onset RTs increase disproportionately, resulting in a divergence of the fast and slow functions.

The comparison of fast and slow onset RTs for chromatic modulations is shown in Fig. 3 for the same observers. Note that these RTs were recorded at the observers’ isoluminant point according to heterochromatic flicker photometry. Again, the slow onset RTs are longer than those from fast onsets. Although the delay varies between observers, the fast and slow RTs versus the chromatic contrast functions do not diverge as they do for achromatic contrasts and are basically parallel throughout the contrast range.

B. Difference Functions for Fast and Slow Onset RTs

Figure 4(a) shows the results averaged across the same three observers after computing the RTD for a range of achromatic contrasts. The data are means and 95% CI. The error bars depicting the inter-observer differences are larger for lower contrasts, as might be expected. As the contrast is increased, the RTD decreases exponentially, suggesting that high and low regions of contrast tap different visual mechanisms when fast onset stimulus is used. This exponential function essentially reflects the divergence between fast and slow onset RTs evident in Fig. 2.

Figure 4(b) shows that the RTD is approximately unchanged for chromatic contrasts, suggesting that, unlike for the achromatic condition, the RTD is largely independent of the chromatic contrast. The slow onset RTs are approximately 80 ms longer than the fast RTs for all contrasts and observers.

 

Fig. 4. Mean (${{\pm 95}}\%$ CI) RTD for the three observers in Figs. 2 and 3. (a) Achromatic contrast. (b) Chromatic contrast.

Download Full Size | PDF

 

Fig. 5. Upper row: mean RT (${{\pm 95}}\%$ CIs) versus red–green ratio (RGR) for slow and fast onset RTs. Lower row: RTD versus RGR. The data exhibit a clear minimum. Isoluminant values based on HFP are NRAP 0.5, IJM 0.42, and DMCK 0.5, as indicated by inverted triangles. Note the increased sampling of RGR around the isoluminant point.

Download Full Size | PDF

C. RTD for Different Red–Green Ratios

As highlighted above, varying red–green ratio enables small, controllable levels of achromatic contrast to be introduced into the stimulus. By definition, when the stimulus is isoluminant (as determined by HFP) for a particular observer, the red and green parts of the stimulus are equal in luminance. This absence of luminance contrast means the target is detected more or less exclusively by the chromatic system. If, as proposed above, the technique of taking the difference between slow and fast onset RTs is sensitive to achromatic intrusions then obtaining this for a range of red–green ratios should reveal a conspicuous minimum at the individual isoluminant point for a particular observer. This idea is tested in Fig. 5, which depicts both the mean RT versus RGR (upper graphs) and the RTD versus RGR (lower graphs) for each observer. The occurrence of a minimum in this function corresponds to isoluminance and is indicated by the inverted triangles in the figure. Interestingly, these functions are not particularly symmetrical about the isoluminant point. This issue has been reported using different stimuli in a previous paper in which RTs were measured either side of isoluminance [15].

4. DISCUSSION

The main goal of the experiments described was to identify the presence of transient activity in a temporally modulated stimulus. To achieve this, we compared simple RTs generated by either fast or slow onset stimuli at zero spatial frequency. The temporal waveform of the stimulus onset, either abrupt or ramped, markedly affected the RT versus the contrast functions. Other factors such as eye movements may be important. Future RT experiments, particularly with patterned stimuli but controlling for the role of shifts in fixation, may be of interest. However, we will argue that the differences between chromatic and achromatic RTs versus the contrast functions for our stimuli that contain no spatial information highlight the operation of different underlying visual mechanisms. The rationale to compare slow and fast onset RTs is quite simple: It has been known for many years that rapid onset RTs bias the responses in favor of transient mechanisms. It is equally well established that achromatic and chromatic information is processed by predominantly transient and sustained visual mechanisms via non-opponent and cone-opponent channels [16]. It therefore follows that comparing slow and fast onset RTs can be expected to identify non-opponent activity in a stimulus.

To test the principle of comparing slow and fast onset RTs as a means to identify transient detectors, a range of red–green ratios was used. As discussed below, manipulating RGR is a useful means to control the level of achromatic (transient) or chromatic (sustained) activity in a stimulus. When transient activity is totally absent, the difference between slow and fast onset RTs should reach a minimum. This prediction was verified for the three subjects with normal color vision.

RT data are influenced by the temporal onset profile of the stimulus. This effect is seen most conspicuously for low contrast (below about 10%), low spatial frequency gratings and was first described by Parry [11]. He showed that, regardless of the contrast, the mean RT increased as a linear function of the raised cosinusoidal ramp duration between 50 and 1000 ms. The implication is that there is an RT trigger point in the contrast domain at which an RT event/response occurs. Note that this point occurs later in each successively slow stimulus. As the slope of the RT versus the duration function is linear, the trigger contrast can be calculated from the ramp duration. This means the contrast corresponding to the additional delay due to the ramp can be calculated for a range of contrasts. This trigger contrast delay is constant across the contrast, confirming that the RT event occurs at the same contrast, consistent with the activity of a transient mechanism. At higher contrasts, the RT delay was time dependent rather than contrast dependent, and the RT event was triggered at different points on the ramped contrast function. This latter behavior is consistent with a sustained mechanism. Of course, there are other characteristics of the chromatic system that are relevant, such as long integration times, sluggish responsivity, and a reduced ability to signal fast events. We would argue that these are all tightly interconnected phenomena related to sustainedness.

In these experiments, we recorded RTs generated by 64 presentations split equally between fast and ramped onset temporal profile. It is evident from Fig. 2 that the choice of achromatic contrast is important when comparing fast and slow onset RTs. The analysis is valid only if around 80% of presentations are detected. If the contrast is too low an insufficient number of stimuli will be detected and the entire session will be dominated by rapid onsets, simply because the visual system is more sensitive to this form of stimulation. Also, as Tolhurst [10] showed, when the onset of a transient stimulus is detected at less than 100% probability, there is a second chance to detect the stimulus at its offset, giving rise to a bimodal RT histogram. On the other hand, if the contrast is too high, the data will be dominated by sustained activity and the transient system will be under-represented. Hence, it is important to note that a series of trials for different contrasts is necessary for each observer to identify the appropriate contrast that will allow the transient-sustained dichotomy to be manifested.

We might ask to what extent we can be sure a pure chromatic stimulus is exclusively mediated by sustained mechanisms. As reported for many years (see [17] for a review) and discussed briefly in the introduction, there is clear evidence of the close linkage between the cell systems in the retina/LGN and psychophysical performance of some, but by no means all, tasks. Most unambiguously, the ${\rm{M}}$ pathway provides the substrate for a luminance system. Psychophysical detection of flicker corresponds to the spectral and temporal properties of this pathway. Historically, these experiments gave rise to the definition of luminance, as discussed by Kaiser [18] and Lennie et al. [19]. Similarly, in a cone space, detection of chromatic changes map onto the M-L cells of the ${\rm{P}}$ pathway and the ${\rm{S}}$ cone driven ganglion cells of the koniocellular pathway [20] so that the spatial and temporal properties of the two pathways complement each other. But the overwhelming observation is that ${\rm{P}}$ cells show highly sustained responses in both the retina and LGN. This is exemplified in Fig. 2 in Lee [1], which highlights the differences between the two pathways and illustrates their link with perception. In almost all those early experiments, data were collected close to the threshold, where neurons can be expected to behave linearly. Of course, M-cells, by definition, are far from linear but it is remarkable that many psychophysical observations are reflected in neuronal physiology given the very large numbers of synapses between signals in the retina and a behavioral response. Note, however, that ${\rm{M}}$ cells do show some linearity; their flash responses can be predicted from their sinewave responses, provided low contrasts are used. It can be concluded that, perceptually speaking, pure chromaticity arises exclusively from sustained responses in the subcortical pathway.

However, as outlined in Lee [1], there are many unanswered questions regarding the extent to which function matches structure in the subcortical visual pathway. As far as this paper is concerned, the fact that RT measurements involve suprathreshold stimulation is highly pertinent to this discussion. Hence, the ideas arising from the observations made here are quite remarkable. If the appropriate stimuli are used, the data show that the specificity between the ${\rm{P}}$ and ${\rm{M}}$ pathways is maintained for suprathreshold stimuli.