Visual latency, visual reaction times

Transcription

1 VISUAL REACTION TIMES 1 Introduction Although we live in a visual world composed of richly varied surfaces, objects and events, complex perceptual decisions are arrived at quickly, often within a few hundred milliseconds. It would be reasonable to expect efficient information gathering systems like our brains to have incorporated into their design distinct subsystems, each specialised to process one or a few of these limited dimensions and attributes. Hence, the system as a whole would comprise a set of parallel information channels. Converging evidence from electrophysiological and psychophysical research indicates that information processing in the visual system occurs along at least two separate, parallel channels. Each channel processes spectral, temporal and spatial information in different ways and their relative importance varies with light level. One psychophysical experiment that can reveal the role of these parallel visual pathways is the study of the reaction time (i.e. the latency of visual process). Of particular interest is to determine how visual reaction time is affected by stimulus variables such as luminance, contrast, spatial frequency and retinal eccentricity. Before interpreting reaction time data, it would be essential to have a short review on the different stages of the retino-cortical pathways, and the basic physiological (Parvo vs Magno pathways) and electrophysiological (transient vs sustained channels) approaches of parallel processing. 2. Magno and Parvo pathways According to Livingstone and Hubel (1987) cells in the P and M pathways can be differentiated on the basis of colour selectivity, contrast sensitivity, spatial and temporal resolution. A physiological distinction between the two types of cells is their contrast sensitivity, with M cells having greater sensitivity than P cells. A corollary of this difference is the much greater contrast gain (rate of elevation of response per unit increase in contrast) of M cells than P cells; P cells require larger light differences before they will respond strongly (Kaplan et al., 199). Therefore, M cells may be especially important for the perception of low contrast objects, whereas P cells may be more important for seeing high contrast objects. Kaplan and Shapley (1986) and Sclar et al. (199) have emphasized the distinctive contrast signatures of the Parvo and Magno pathways in the LGN. Similar data are available at the retinal ganglion cell level (Lee et al., 199). With a steady-state luminance varying stimulus, the cell obtains a constant resting level. As Figure 1 illustrates, for the M cell, the amplitude response shows rapid growth with increase in contrast of the stimulus levelling off above.3 contrast. In comparison, the P cell amplitude response shows a steady growth with increasing contrast. Fig 1b shows that P and M cells are clearly distinguished by their responses to low-contrast stimuli (P cells exhibit no response at low contrast < 1). a b 6 P cell M cell Contrast Figure 1: Examples of the average contrast response for P (filled circles) and M (empty circles) cells in the LGN as a function of contrast. The stimulus was an optimal spatial frequency sinusoidal drifting grating (for each cell). The amplitude of the component at the frequency of stimulation was used as the measure of S PLAINIS, VEIC, UNIVERSITY OF CRETE 1

2 response. (a) (after Kaplan and Shapley, 1986). (b) (replotted from Sclar et al., 199). Figure 2 illustrates the difference in contrast gain between M and P cells at scotopic and low mesopic levels. Contrast gain is the product of the gain (impulse per quantum) of the receptive field centre mechanism and the average flux (quanta per second) effectively absorbed in the centre (it equals the slope of the response vs. contrast function in figure 4.3a). Purpura et al. (1988) found that, although contrast gain of both M and P cells dropped as mean illumination decreased, P cells were affected relatively more. They also suggested that P and M cells might differ in contrast gain at the same photopic levels as a consequence of their different receptive field sizes alone. It is also likely that the relative sensitivity of P and M cells at any luminance level depends on the amount of rod input to them, and that, in turn, must depend on retinal eccentricity. Figure 2: The average contrast gain, the slope of response versus contrast function, of 8 M cells and 15 P cells plotted against the mean retinal illumination. Below 1 photopic troland no responses could be elicited from P cells (after Kaplan, 1989; replotted from Purpura et al., 1988). Despite this difference in contrast response, there is a substantial overlap in spatio-temporal frequency response of the M and P pathways, although M cells appear to prefer somewhat higher temporal and lower spatial frequencies. In contrast, the pattern-sensitive sustained channels are characterised by preference for higher spatial frequencies and lack of preference for temporally modulated stimuli over stationary ones. This can be explained, as can be seen in figure 4.5, by the fact that P cells have smaller receptive fields, thus are more sensitive to spatial properties such as fine gratings and sharp contours; whereas M cells have larger receptive fields and are more sensitive to coarse gratings and flicker. Additional properties which differentiate between M and P cells are their respective response latencies and fibre conduction velocities. Along the entire visual pathway, from retina to visual cortex, conduction velocity in the M pathway is considerably higher than in the P pathway as revealed by response latencies. This advantage of M cells relative to P cells holds at all retinal eccentricities (Breitmeyer, 1984). Lennie (198) found that he was able to eliminate response latency differences between X and Y cells by using near-threshold stimuli. However, Maunsel (1987) showed electrophysiologically that with clearly suprathreshold stimuli (where the differences in response characteristics between the two pathways emerge), M pathway cells had a shorter visual response latency than P pathway cells. The most intriguing difference between M and P cells relates to their response to chromatic patterns (Lennie, 1986). The centres and surrounds of receptive fields of P cells show a colour-opponent organisation superimposed on the spatial organisation: centre and surround have different spectral sensitivities. There are clearly two distinct subgroups of P cells: those that have a "red-green" chromatic opponency (transmitted to LGN by midget cells, which convey M and L cone opponent signals) and those that have a "yellow-blue" chromatic opponency (transmitted to LGN by small bistratified cells, which convey S cone signals). In the latter, the centre-surround arrangement of the chromatically opponent mechanisms is less pronounced. M cells have in common with P cells the S PLAINIS, VEIC, UNIVERSITY OF CRETE 2

3 centre-surround arrangement of receptive fields. However, a few M cells show non-linearities. The differences between P and M cells in their responses to colour arise from differences in their synaptic connections in the retina. Recall that primates have three types of cones, with different absorption spectra. In an M (broad-band) ganglion cell, the receptive field centre and surround receive input (via bipolar, horizontal and amacrine cells) from the same type of cones, whereas for a P (colour-opponent) cell, the type of cone driving the two parts of the receptive field are different. 3. Contrast sensitivitiy of P and M cells As mentioned above, a striking difference between M and P cells is their achromatic contrast sensitivity. Recent investigations report a 1 times higher contrast sensitivity of M cells at low spatial frequencies than that of P cells (Kaplan and Shapley, 1982; Hicks et al., 1983), with P cells showing little or no response to gratings of less than about 8% contrast (see figure 1). They differ in other ways: M cells show spatial frequency tuning, whereas P cells are either low-pass or poorly tuned (see figure 3). By comparison, the human psychophysical contrast threshold over the spatial frequency range below 1 c/deg is near, or even below, 1% contrast (Kelly, 1961; Campbell and Robson, 1968). One would expect, therefore, that the human luminance contrast sensitivity function (CSF) particularly at low spatial and high temporal frequencies, would largely reflect the activity of M cells. A compensating factor that must be considered, however, is probability or neural summation (Legge, 1978). Probability summation suggests that the much greater number of parvocellular units increases the chance of detection, and hence the greater the measured sensitivity, particularly at the central retina. This hypothesis was favoured by Lennie and D Zmura (1988) who claimed that P cells exclusively determine the CSF. However, there has been no evidence that there is a 1-fold increase in the sensitivity of the P system, due to either probability or neural summation. The simplest interpretation of all the electrophysiological and psychophysical results is to consider that both P and M neurons (or systems) contribute to contrast sensitivity, but that the contribution of P cells increases with spatial frequency. As it was pointed out (Kulikowski, 1989) P cells should predominate near the psychophysical resolution limit as they have relatively high density compared with M cells. At the resolution limit the output of P cells has to be added in order to extract only luminance information (possibly the sensitivity is increased due to neural summation). Near the maximum contrast sensitivity (about 2 c/deg) the contribution of M cells to threshold detection is likely to be predominant. At intermediate spatial frequencies, it is likely that there is an overlap, i.e. both systems contribute to detection, depending on other stimulus parameters such as presentation time and mean luminance. 1 Magno-X cells Magno-Y cells Parvo-X cells Contrast Sensitivity Spatial frequency (c/deg) Figure 3: Contrast sensitivity functions of monkey Parvo and Magno channels. The open symbols (and dashed line) are from P cells; the filled circles (grey line) are from Magno-X cells cells; the filled triangles (black line) are from magno-y cells. Contrast sensitivity was the reciprocal of the contrast required to give a response with a fundamental Fourier amplitude of 5 impulses/sec (replotted from Kaplan and Shapley, 1982). S PLAINIS, VEIC, UNIVERSITY OF CRETE 3

4 As mentioned above, the findings of studies on differential fibre conduction velocity (on ganglion and LGN cells) lead to the assumption that the response latencies of transient neurons are shorter than those of sustained neurons. However, these physiological findings of differences in conduction velocity were obtained with animals only, and corresponding mechanisms in the human visual system can therefore only be extrapolated. Several psychophysical strategies have been applied in order to differentiate the response properties of the Parvo and Magno pathways of the human visual system. Contrast sensitivity is a candidate, but can only study bvisual performance at threshold levels If we wish to understand the roles that the M and P systems play in vision, two approaches are possible: we can study the properties of each cell type and relate them to visual capacities, or we can study visual performance under conditions which would selectively stimulate either the M or the P system. One psychophysical experiment that can reveal the role of the two distinct systems and distinguish their contribution is the study of the reaction time as a function of contrast. It appears that at a contrast level above about.1, reaction times become shorter as though a new system was activated (Harwerth and Levi, 1978). 4. Simple reaction time Simple reaction time is defined as "the interval between the onset of the stimulus and the response under the condition that the subject has been instructed to respond as rapidly as possible" (Teichner, 1954). Simple reaction time (RT) consists of three components: the latent period during the reception, the relay and transmission of the sensory impulses to the higher centres, and the time interval during the preparation and execution of the motor response. Any of the factors affecting these processes will obviously also affect the measured RT. Some of these factors are stimulus/receptor factors (e.g. sense stimulated, number of the sense organs -or receptors- stimulated, location of the stimulation in the visual field, intensity and duration of the stimulus), central and motor factors (e.g. age and sex of the subject, body position, responding organ) and special factors (e.g. temperature, drugs, alcohol, tiredness, lack of sleep). Figure 4 shows the components of reaction time for the sense of vision. Photoreceptors Striate cortex Light Optic nerve Motor neurons Muscle cells Figure 4: Representation of the components of information processing from photoreceptors to the responding organ (e.g. hand). Visual reaction time should consist of: a) the latent period during the reception of light from the photoreceptors (rods and cones), b) the relay and transmission of the neural signal (output of retina) by the optic nerve to the corresponding area in the brain (striate cortex), and c) the time interval during the preparation and execution of the motor response (contraction of muscles). One can represent a stage model of visual reaction time with a simple equality similar to that S PLAINIS, VEIC, UNIVERSITY OF CRETE 4

5 proposed by Lupp et al. (1978): RT = PT + MT where RT is reaction time, PT is perception time and MT is motor time. PT is the time for the evidence to reach the criterion for detection and MT is the time from detection (i.e. when the criterion level for detection is reached) to the response, e.g. keypress. Motor time is assumed to be independent of stimulus variables (e.g. luminance, wavelength, contrast etc.). Reaction time differences to these variables only reflect differences in the time required to reach the criterion level of evidence (perception time). The latency of the cortical response (perception time) has been proved to strongly depend on the stimulus intensity but is also influenced by parameters of peripheral stimulation other than intensity, such as wavelength, retinal locus, target size and stimulus duration. 5. Studies of reaction time Early investigators who examined the relation between reaction time and several stimulusdepending factors usually measured reaction time to a flashing light stimulus. Later studies investigated reaction times to more elaborate psychophysical stimuli, like sine-wave gratings, with which other stimulus variables could be manipulated, such as contrast and spatial frequency. 5.1 Effect of intensity and duration of stimulus Pieron (1952) enunciated a definite, quantitative relation between the latency of response in the human visual system and the luminance of the visual stimulus: the latency above the irreducible part varies as intensity, I, raised to a power. In mathematical terms Pieron s law can be expressed as: L - L = k I β, where β is the power of exponent, L refers to the asymptotic level, and k is a constant of proportionality. The analysis of data from Mansfield s study (1973) of reaction times demonstrated the validity of Pieron s formulation. According to Mansfield (1973) the reaction time can be separated into a fixed (L ) and an intensity-dependent component (L - L ), and can be described by Pieron s power function of intensity, where the value of the exponent β depends upon stimulus size. This indicates that the individual differences are mainly attributed to the efferent components of the response. The intensity-dependent component differs little between individuals and remains constant over sessions. Figure 5: Response latency as a function of peak luminance for a 3 msec flash of white light from a foveal target subtending.72. Standard errors for the means are indicated by the vertical bars (after Mansfield, 1973). S PLAINIS, VEIC, UNIVERSITY OF CRETE 5

6 Raab et al. (1961) also found that RT was a decreasing function of luminance and foreperiod duration and was independent of stimulus duration in the range of 1 to 5 msec for a given stimulus luminance. The form of the relationship between luminance and RT will depend on stimulus duration provided each stimulus duration is greater than the critical duration (see figure 6). Raab and Fehrer (1962) showed that the critical duration for dim targets (of about 1 cd/m 2 luminance) lies between 1 and 25 msec (figure 6). This closely agrees with the findings of Mansfield (1973), who stated that for constant luminance, latency decreases with increasing stimulus duration up to approximately 1 msec, then remains essentially constant. Figure 6: Reaction time as a function of stimulus duration for various flash luminances (expressed in Ft-L, 1 Ft-L=3.426 cd/m 2 ) (after Raab and Fehrer, 1962). The above RT studies were limited to flash-detection situations. Kaswan and Young (1965) used more complex pattern-discrimination tasks and found similar results to Raab and Fehrer (1962), which suggests that the range of exposure durations affecting RT in discrimination tasks and detection thresholds are similar. Their results also indicated that RT in discrimination tasks is affected by a much wider range of luminance than exposure-duration values. Lupp et al. (1976) used sinusoidal gratings of constant spatial frequency and found that exposure durations of 3 to 6 msec produce shortest RTs. They also suggested the effect of exposure duration to be of minor importance, and not directly related to the processing of spatial information. They concluded that the differences in perceptual latencies may be due to higher neural mechanisms, namely the strategy during perceptive experiments (Lupp et al., 1976). Pike (1973) also stated that "a more difficult discrimination task takes more time". 5.2 Effect of retinal locus Rains (1963) stated that the flash (stimulus) luminance is a critical factor in determining the relationship between the retinal locus of the stimulus and reaction time. He concluded that when stimulus intensity is above cone threshold, reaction time is shortest for foveal presentations and it becomes longer as it is moved farther out into the periphery. However, when the flash intensity is too weak to exite cones, reaction time is shortest at about 1 or 2 degrees (figure 7). Mansfield (1973) measured the luminance required to produce a constant latency (L- L ) above the asymptotic level as a function of retinal locus for both scotopic and photopic levels. He showed that, for photopic levels, this latency increases with distance from the fovea, whereas for scotopic levels, the luminance required to produce a constant latency above the asymptotic level decreases with distance from the fovea up to 2, but tends to increase for greater eccentricities. Moreover, at all extra-foveal loci, the difference between the asymptotic latencies (L ) for the photopic and scotopic levels amounted to approximately 4 msec. S PLAINIS, VEIC, UNIVERSITY OF CRETE 6

7 Figure 7: Reaction time as a function of retinal position for three intensities of white light (A=3555 ml, B=11.25 ml, C=.11 ml; 1 ml=3.183 cd/m 2 ) (after Rains, 1963). 5.3 Effect of stimulus contrast As noted above, there are many physiological and psychophysical indications which support the notion that contrast is one stimulus parameter that might distinguish transient and sustained systems, as transient channels (i.e. the M system) show a higher contrast gain. Therefore, an interaction between contrast level and reaction time is expected to be derived. Harwerth and Levi (1978) investigated reaction time to vertical sine-wave gratings as a function of, among other variables, contrast. They found that, for 5 msec flashes of gratings, reaction time to low (e.g..5 c/deg) and high spatial frequencies (e.g. 12 c/deg) generally decreased continuously and exponentially with increases of contrast ranging from threshold to suprathreshold values of 45%. However, for intermediate spatial frequencies (1-8 c/deg), the decrease of reaction time with similar increases of contrast was characterised by a discontinuity revealing that one exponentially decaying function dominated up to a contrast value of about 5-1%, followed by another function which dominated at higher contrast values. As figure 8 illlustrates, for a 4 c/deg spatial frequency stimulus, a biphasic RT versus contrast function was obtained, suggesting the presence of two different perceptual mechanisms. The faster RTs correspond to the relatively flat, high contrast portion of the function. The position of this break in the biphasic reaction time data, i.e. the contrast level at which the transition may occur was also proved to vary with stimulus parameters. Furthermore, when the stimulus duration was reduced to 5 msec, high spatial frequency data (8 and 12 c/deg) also showed a biphasic function. They interpreted this as revealing the operation of transient mechanisms at high contrast and sustained mechanisms at low contrast. They supposed that because shorter RTs were obtained at higher contrasts, these conditions favoured transient detectors. The same conclusion was reached by Felipe et al. (1993) who reported similar RT-contrast functions. These observations have been reinterpreted in the light of the background neurophysiology (Plainis & Murray, 2; Parry, 21). As discussed in detail below, it is now generally accepted that the detection of low contrast stimuli is mediated predominantly by the Magnocellular (M) pathway. Hence, it seems likely that, even though these are relatively slow, the RTs in the low contrast segment of the RT vs. contrast function are dominated by the activity of M neurons. Interestingly, Parry et al. (1988) also produced convincing evidence of the existence of two sections in the RT vs. achromatic contrast function, but not in the equivalent isoluminant chromatic function, where only one mechanism, the P pathway, is known to mediate the detection of the stimulus. S PLAINIS, VEIC, UNIVERSITY OF CRETE 7

8 Reaction time (msec) Log Contrast Figure 8 Mean reaction time and standard deviations as a function of contrast for three spatial frequencies (.5, 4. and 12. c/deg), with a viewing duration of 5 msec. The RT versus contrast function for the intermediate frequency is biphasic, whereas it shows an exponential unimodal form for the low and high spatial frequencies. The curve for the.5 c/deg is on a true scale, but the other curves have been succesively shifted to the right by 1 log unit for ease of viewing (after Harwerth and Levi, 1978). Burbeck and Kelly (1981) measured thresholds for detecting a vertical grating superimposed on a fixed-contrast horizontal grating and demonstrated that human transient channels (activated by test gratings of 2 c/deg or lower), as compared with sustained ones (activated by test gratings ranging from about 4 to 12 c/deg), are characterised by a gain that increases at a faster rate as a function of contrast. However, in agreement with Harwerth and Levi s results they found that at contrasts below 5% (low contrast), the gain characteristics did not differ significantly between transient and sustained channels. Felipe et al. (1993) studied a full range of spatial frequencies (1 to 4 c/deg) at different contrasts and found that the contrast sensitivity function (CSF) has an important influence on the curves produced by plots of reaction time versus spatial frequency. However, the influence becomes less important than the effect produced by the transition from transient to sustained activity in a narrow interval between 6 to 8 c/deg. They also concluded that at high contrast levels, the faster responses correspond to low spatial frequencies, whereas at low contrast levels (<.7) the faster responses correspond to intermediate frequencies of 6-1 c/deg. On the whole, these results are consistent with the electrophysiological studies of the gain characteristics of retinal ganglion cells reported by Purpura et al. (1988). The implication of these findings, might be the fact that at low contrasts, sustained channels may actually respond at a shorter latency than transient ones (Lennie, 198), but this relation reverses at higher contrasts. 5.4 Effect of spatial frequency The variation of reaction time with spatial frequency has been demonstrated by several authors. In all cases an increment of reaction time with increasing spatial frequency was reported, but there is no agreement as to its physiological significance. These studies used either high physical contrast or constant perceptual contrast and have been interpreted as showing that low spatial frequencies are processed by the transient channels (shorter latencies) while the high spatial frequencies are processed by the sustained channels (longer latencies). The spatial frequency range that reaction time increases can be generally interpreted as a transition zone in which both mechanisms operate. However, the limit of this transition zone has not been clearly established. There is evidence that for near-threshold stimuli the sustained channels operate for spatial frequencies at least as low as 3.5 c/deg (Tolhurst, 1975). Kulikowski and Tolhurst (1973), S PLAINIS, VEIC, UNIVERSITY OF CRETE 8

9 Vassilev and Mitov (1976) indicated that a spatial frequency of about 5-8 c/deg could be the above mentioned limit. Harwerth and Levi (1978) found an interval between 1 and 8 c/deg within which they claimed that both transient and sustained channels are effective. Tolhurst (1975), by using near-threshold contrasts of a sinusoidal grating, claimed that the reaction time histograms can distinguish transient from sustained responses. He showed that the reaction times for transient channels (low spatial frequency -.2 c/deg) appear to be distributed bimodally, with each mode corresponding to the abrupt onset or offset of the grating (RT histograms show two peaks, see figure 4.14). According to Tolhurst (1975), these modes reflected the probabilistically distributed activities of transient channels to grating onset and offset. On the other hand, for a higher spatial frequency grating (3.5 c/deg) activating predominantly sustained channels, the reaction time distribution was consistent and, moreover, of longer latency than the onset reaction time to the low frequency grating (figure 4.14). This supports the physiological findings of neural responses (see figure 4.7), which show prolonged excitation (or inhibition) for sustained cells, and an abrupt excitation at the onset of the stimulus followed by an abrupt inhibition at the offset, for transient cells. Furthermore, Tolhurst (1975) proved that a more extended (longer) stimulus is more detectable than a less extended (shorter) stimulus, also because of the probabilistic nature of the threshold. However, stimulus duration only affects the peak (of transient responses) associated with the stimulus offset, which would shift to the right (longer RTs) for longer duration stimuli, and not the peak associated with the stimulus onset. Reaction time histograms could not distinguish transient from sustained responses when the stimulus is of high contrast, because in that case the stimulus would be detected near the beginning of the trial and the reaction times will be only a little scattered. Similar findings have been reported by Breitmeyer et al. (1981), who used a flicker masking technique. Number of RTs Stimulus Time from beginning of stimulus Figure 9: Reaction times to near-threshold gratings of 3.5 c/deg (left) and.2 c/deg (right) spatial frequency. The stimuli had sudden onset and offsets. The RT distribution is bimodal for the low spatial frequency grating (with two peaks corresponding to the onset and offset of stimulus) and unimodal for the higher spatial frequency grating (replotted from Tolhurst 1975). Early studies used the same physical contrast for all spatial frequencies used. Although the differences in reaction time with increasing spatial frequency (about 1 to 15 msec) may partially reflect the differences between response latencies of low spatial frequency transient channels and high spatial frequency sustained channels, they also may be due to the fact that reaction time increases as the subjective contrast of gratings of equal physical contrast decreases with increasing spatial frequency. Therefore, this procedure does not control for the attenuation of contrast with spatial frequency due to dioptric or neural factors. Lupp et al. (1976) chose suprathreshold contrasts that were 5 times threshold to overcome this problem. However, Georgeson and Sullivan (1975) found that the increase in perceived contrast with physical contrast is greater at low contrast for both low and high spatial frequencies. So, it seems unlikely that the stimuli used by Lupp et al. (1976) were matched in terms of perceived contrast or detectability at suprathreshold contrasts. This may happen at higher levels of physical contrast (i.e., 5 to 7%), as was proved from contrast matching experiments (Watanabe et al., S PLAINIS, VEIC, UNIVERSITY OF CRETE 9

10 1968; Blakemore et al., 1973; Georgeson and Sullivan, 1975). Contrast matching data showed that, at these high contrast levels, suprathreshold functions become increasingly flat across the whole spatial frequency range (figure 1) Spatial Frequency (c/deg).7 1. Figure 1: Functions of apparent iso-contrast at two different suprathreshold contrasts of.7 (squares) and.22 (triangles), compared to the CSF of the subject (circles) (replotted data from Blakemore et al., 1973). The subject fixated between two gratings, a reference of 5 c/deg and the test which is plotted on the x-axis. The spatial frequency of the test grating was varied and its contrast was adjusted until it seemed similar to the contrast of the reference grating. Breitmeyer (1975) measured the visual reaction time to the onset of a 5 msec presentation of sinusoidal gratings at a contrast of 5% and at spatial frequencies ranging from.5 to 11. c/deg. He found that reaction time over this range increased monotonically with increasing spatial frequency. RT was roughly 2 msec at a spatial frequency of.5 c/deg and increased to a value of 35 msec at a spatial frequency of 11. c/deg (figure 11). Since contrast sensitivity declines at higher spatial frequencies, Breitmeyer (1975) also measured reaction time to the same gratings but with their individual physical contrasts adjusted so that their apparent contrasts were equal. Despite this cancelling out of differences in contrast sensitivity, he found that the reaction time increased by about 4 msec over the same range of spatial frequencies employed. Figure 11: Mean RT to vertical sinusoidal gratings as a function of spatial frequency at an objective contrast of.5 (left) and adjusted to have equal subjective contrasts (right) (after Breitmeyer, 1975). S PLAINIS, VEIC, UNIVERSITY OF CRETE 1

11 6. Recent studies on Reaction Times In following experiments a series of experiments designed to characterise the processing of suprathreshold contrast using reaction times (RTs) is described. Some novel findings are reported; first, it is possible to explain all the RT data with a simple equation linking contrast, spatial frequency and luminance. Second, it is shown that the RT equivalent of contrast gain corresponds closely to contrast gain values obtained neurophysiologically, in terms of both spatial frequency and luminance. Moreover, RT data are transformed to produce sensitivity functions with varying luminance levels, eccentricity and stimulus duration, in order to establish whether RT-based contrast coding and contrast sensitivity are constrained by the same neural processes. Both of these strategies have the same aim to determine the link between RTs and the neural mechanisms in the early stages of visual processing. 6.1 Methods - Procedure The stimuli were vertical sinusoidal gratings, modulated in luminance, and displayed on a Barco CCID7651 Calibrator colour monitor. The red, green and sync inputs to the monitor were supplied by a 12-bit, two-channel grating generator card (Millipede Prisma VR1 series 2) in a PC. The mean luminance of the screen [L= (Lmax+Lmin) / 2] was 2 cd/m 2, and this was attenuated with neutral density filters to give lower luminances. The circular target subtended an angle of 7.13 degrees at a viewing distance of 114 cm. Normal pupils were used apart from one control experiment. Subjects fixated on a cross located in the centre of the illuminated area of the screen for central viewing and on a series of red LEDs when eccentric viewing was tested. RT data were collected for a range of contrasts from suprathreshold (.5) to threshold (C o ) detection. Contrast was defined after Michelson: C = (Lmax-Lmin) / (Lmax+Lmin), where Lmax = maximum luminance and Lmin = minimum luminance. A series of spatial frequencies and mean luminances were used. Eccentricities of, 5, 1 and 15 degrees for both hemi-fields were tested. Generally stimuli were presented for 34 msec. In some experiments stimulus duration varied between 2, 5 and 5 msec. RTs were determined using a CED 141 smart interface (1 ms temporal resolution), linked to a PC, and a purpose-designed computer programme. They were measured by displaying vertical gratings with an abrupt onset and offset. Subjects responded by pressing a button which triggered the interface (CED 141). The subject was instructed to press the response button immediately he/she detected the stimulus. Only responses between 15 and 1 msec were accepted.\ Figure 12: Schematic diagram showing the layout of the set-up. S PLAINIS, VEIC, UNIVERSITY OF CRETE 11

12 6.2 RT vs contrast functions In figure 13a RT data for one condition (11.22 c/deg, 5 msec duration) are plotted as a function of contrast on a logarithmic axis. It is evident that RTs decrease exponentially as contrast is increased. As was previously shown (Plainis & Murray, 2), the RT vs. contrast curves can be described satisfactorily by the following monotonic function: τ = τ +β. C -1 (1) This is a Piéron function (described above) with the exponent α being equal to -1, where τ is the measured reaction time, τ is the asymptotic reaction time reached at the highest contrasts (comprising motor time and other non-visual factors), β is a constant (characterizing the steepness of the curve) and C is Michelson contrast. From equation (1) it follows that, if the data are re-plotted in terms of 1/C, the resulting slope, k, would be linear, as is confirmed in figure 13b. This relationship is extremely robust for many observers and a wide range of stimulus conditions (eg luminance, spatial frequency), as shown in Plainis and Murray (2). Fig. 13. Plots of RT vs. contrast (a) and vs. the reciprocal of contrast (b) for subject LG and for a specific stimulus (spatial frequency: 11.22c/deg). Each data point represents the mean of at least 24 measurements (maximum¼32) and the error bars ±1 s.e. The solid line drawn through the data is the best fit of Eq. (1) (a), or the least square regression fit (b). The vertical dotted line indicates C =.1. However, some stimulus conditions produce a bi-linear function. For example, as seen in Figure 14a, RTs decrease as contrast is increased, but tend to level off producing an asymptote at around C =.1. As C increases, RTs again reduce, as if a different detection mechanism operates. The break is more obvious in Figure 14b, where RTs are plotted as a function of 1/C. This observation confirms previous findings. There are two points to note from these data; first, the slope of the high contrast region is much steeper than for the low contrast region, and second, the overall slope, k, is determined by the low contrast region, as illustrated by the proximity of the solid line to the dashed line in the low contrast region. In subsequent figures, solid lines are best fit regression lines for the full contrast range and dashed lines are fits to the low and high contrast segments. Fig. 14. Plots of RT vs. contrast (a) and vs. the reciprocal of contrast (b) for a specific stimulus (spatial frequency: 3.74 c/deg, duration: 5 ms, luminance: 2 cd/m 2 ) which produces a biphasic function. The dashed lines drawn through the data are the best fits of Eq. (1) (a) or the least square regression fits (b) for high contrast levels (.5.1) and low contrast levels (.1 to threshold). The solid lines are the fits for all the data points. S PLAINIS, VEIC, UNIVERSITY OF CRETE 12

13 The Piéron function (1), used to model RTs is identical to the well-known Naka-Rushton equation, if the reciprocal of RT is given as a function of contrast (C): τ C τ = = (2) 1-1 ( + k C ) (C + k ) τ τ RTs are reciprocally related to sensitivity; RTs to high suprathreshold targets are short and those to close-to-threshold targets are longer. The Naka-Rushton equation is frequently used to describe the contrast-response functions of neurons in the visual pathway (eg Kaplan & Shapley, 1986; Sclar et al., 199). In this way RTs can be linked to response amplitudes and gain characteristics of P and M cells. Moreover, contrast gain, the slope of the Naka-Rushton function at % contrast is used to describe sensitivity of cells. The slope of the Naka-Rushton function (2) at contrast C is: τ τ 1 C ) (C + k 1 1 ( τ (C)) = -1-1 (C + k τ τ The slope at % contrast is: τ 1 ( ()) τ τ = k 1-1 = k 1 ) Thus, k -1 is an index of sensitivity (the gain) of the underlying detecting mechanism: steep slopes indicate low gain and consequently low sensitivity, shallow slopes indicate high gain (ie high sensitivity). It characterises the link between contrast and RT for each of the different stimulus conditions. Figure 15 shows the effect of spatial frequency on RT as a function of 1/C. k becomes steeper as spatial frequency increases (and as sensitivity decreases). At low spatial frequencies, where small increments/decrements in contrast influence RT very little, the values of k are low (high sensitivity), whereas at high spatial frequencies, where small increments/decrements have a large effect on RT, the values of k are high (low sensitivity). It is evident that in this case k gradually and systematically increases (ie sensitivity decreases), reaching a maximum at 11.22c/deg. Fig. 15. Plots of RT vs. the reciprocal of contrast (1=C) for a stimulus of 2ms duration and for a range of spatial frequencies. Mean screen luminance was 2cd/m 2 and eccentricity deg. The solid lines represent the least squares regression fit for all the data points, whereas the dashed lines represent the least square regression fits for the two segments (i.e. high levels of contrast (.5.1) and low levels of contrast (.1 to S PLAINIS, VEIC, UNIVERSITY OF CRETE 13

14 threshold)). The legend indicates the spatial frequency of the grating used. Also, k is the slope for all the data points (solid line) and k 1 and k 2 are the slopes for high and low contrast levels (dashed lines), respectively. The asterisk indicates a statistical significance difference (p < :5) between k 1 and k 2 ; NS indicates no significant difference between the slopes. Turning now to the dashed lines which are least square regressions for the high and low contrast regions of the curves, the discontinuity in the RT vs. 1/C function is highly conspicuous for certain stimulus conditions. Note that for all conditions the break occurs around contrast.1 as shown by previous authors for RTs and VEPs. Values k 1 and k 2 which depict the slopes for the high and low contrast regions, respectively, are given in the top left hand corner of each panel. When the function is bi-linear, the high contrast region of the curve always has a greater slope (ie low gain/sensitivity) than the low contrast region. It is also evident that the slope k 1 of the high contrast region does not change much with spatial frequency, but the slope k 2 of the low contrast region increases dramatically with spatial frequency. At the high spatial frequencies (7.48 and c/deg) the RT vs. 1/C function is no longer bi-linear. We now consider the effect of stimulus eccentricity. Figure 16 illustrates how RT varies with horizontal eccentricity for a spatial frequency of 5.57 c/deg. The uppermost panel in figure 16 shows the data for central fixation. The panels below are displayed in pairs showing data for 5, 1 and 15 degrees eccentricity with the left hand panel showing the data presented to the left hemi-field and the right hand panel the data presented to the right hemi-field. There is an obvious break-point at around C =.1 and the low contrast region has a much shallower slope than the high contrast region, again revealing the presence of two mechanisms. The slope of the low contrast region increases at 5 degrees until it coincides with the high contrast slope at 1 degrees, when the break disappears. This probably indicates that only a single mechanism operates at higher eccentricities.. Fig. 16. Plots of RT vs. the reciprocal of contrast (1=C) for a range of stimulus eccentricities. The spatial frequency of the grating was 5.57 c/deg, the stimulus duration 34ms and the mean screen luminance 2cd/m 2. The legend indicates the eccentricity used (LH: Left Hemifield, RH: Right Hemifield). Other details as in figure 15. The figures so far described show that the slope, k, varies systematically with stimulus duration, eccentricity, luminance and spatial frequency. In the following figure the RT-based contrast sensitivity (k -1 ) is compared with conventional threshold-based contrast sensitivities. In figure 17a the spatial tuning of k -1 for a wide range of luminances is shown. For both subjects the high S PLAINIS, VEIC, UNIVERSITY OF CRETE 14

15 luminance (2 cd/m 2 ) data exhibit band-pass characteristics. As luminance is reduced the function becomes low-pass and gain decreases. It is interesting to compare these data with the classical contrast sensitivity vs. luminance data of Van Nes and Bouman (1967) and Daitch and Green (1969). The overall effects of reducing luminance are qualitatively similar. In figure 17b the effect on RT of changing eccentricity is summarised. Data from the left and right hemi-fields are shown either side of zero eccentricity. It is evident that, K -1 adopts the shape of the contrast sensitivity function as it varies with eccentricity (Johnson, Keltner & Balestrery, 1978; Robson & Graham, 1981; Pointer & Hess, 1989). At 2 cd/m2 (solid lines) it is maximal at the fovea and decreases approximately linearly with eccentricity. Under mesopic conditions (.2 cd/m 2 ), k -1 remains largely independent of eccentricity, when low spatial frequencies (.49 and 1.71 c/deg) are used. Note that the functions are symmetrical for the two hemi-fields. Fig. 17. Plots of the RT-based contrast gain (k -1 ) as a function of: (a) spatial frequency for a range of luminances, (b) eccentricity (both hemifields) for two subjects. 6.3 Discussion The transition point between the two slopes occurs at around C =.1. The segments above and below.1 are interpreted as revealing the activity of Parvo and Magno pathways, respectively. This suggests that RTs are regulated by the characteristics of neurons at the early stages of visual processing. In addition, it emerges that the bi-linear function may be present or absent, depending on sensitivity, for variations in eccentricity, luminance and duration of the grating stimulus. By taking the reciprocal of the slope of the RT vs. 1/C functions, k-1, as a measure of sensitivity (gain), we show that close-to-threshold RTs reflect how contrast sensitivity varies with these parameters. This is consistent with the suggestion (Crook, et al., 1988; Kulikowski, 1989; Shapley & Hawken, 1999) that the M system forms the physiological substrate of the contrast sensitivity function (CSF). RT-contrast functions reveal P and M processing The bi-linear RT-contrast relationship presented here, offers new evidence regarding the physiological mechanisms underlying RTs and supra-threshold contrast coding. The discontinuity in the RT-contrast function between low and high contrast levels indicates a transition from M- dominated to P-dominated activity. At low contrasts, only a relatively small number of neurons, having high gain and fast responses (M cells) are activated. Increasing the contrast of the stimulus, recruits additional neurons (the more numerous P cells) and thus reduces synaptic delay, probably via a probability summation mechanism. Moreover, M cells tend to saturate at high contrasts (Kaplan & Shapley, 1986). Therefore, it seems that the faster, high contrast branch of the bi-linear RT-contrast plot represents the contribution from a second population of neurons, the P cells. There are three lines of evidence to support the above explanation. First, where there is a discontinuity, it falls close to contrast.1. It is well known that M cells are selectively activated at S PLAINIS, VEIC, UNIVERSITY OF CRETE 15

16 contrasts below.1. The evidence for this comes from many different types of experiments; Tootell, Hamilton and Switkes, (1988) showed that low contrast (<.8) gratings produced de-oxyglucose staining only in the M projection to V1 macaque. Hicks et al. (1983) and many other authors (eg. Derrington & Lennie, 1984; Kaplan & Shapley, 1986; Lee et al., 1989; Sclar et al., 199) have used electrophysiological methods to show that P cells have much poorer sensitivity to luminance contrast than M cells for an extended range of spatial and temporal frequencies. It therefore seems unlikely that they subserve the low contrast RTs. Second, the RT-contrast functions show higher gain (shallow slopes) at low contrasts and Kaplan and Shapley (1982) demonstrated that M cells have 1x higher gain than P cells. Third, the discontinuity is not present under all conditions; crucially, it is only obtained at low to moderate spatial frequencies and when sensitivity is high. When sensitivity is compromised (eg. high spatial frequencies, low luminances, eccentric viewing) the gain of the underlying mechanism is low (steeper functions), and the data can be fitted by a single function, which is likely to reflect the activity of a single mechanism or, as discussed below, the combined activity of P and M cells. This interpretation is also supported by the observation that, in the bi-linear RT-contrast plots, the slope of the low-contrast branch becomes gradually steeper with increasing spatial frequency, until it coincides with that of the high contrast region. It follows that under conditions where only a single mechanism is known to operate, a simple linear function should be obtained. This has been shown to be the case for RTs obtained from isoluminant chromatic stimuli which are processed exclusively by the P system (Parry et al., 1988; Parry, 21; Burr & Corsale, 21). Similarly, monophasic RT-contrast functions emerge when slow onset/offset stimuli are used (Parry, 21), or when high spatial frequencies are tested (see figure 15), where again detection is presumably mediated by a single mechanism, the P system, or perhaps combined activity of both systems (see below). On the other hand, low luminances (see Plainis & Murray, 2) and parafoveal presentation (> 5 degrees) seems to favour a single system, presumably the M pathway (Thomas et al., 1999). The dichotomy in the RT-contrast function is also present in the eccentricity data (at 5 degrees; see figure 16), revealing again the presence of two mechanisms. The discontinuity disappears at greater eccentricities for 5.57 c/deg but remains at the lower spatial frequencies (.49 and 1.71 c/deg) to 15 degrees eccentricity, because the sensitivity to low frequencies at these eccentricities is relatively high. When sensitivity decreases dramatically, as occurs for a 5.57 c/deg grating at 1 to 15 degrees, a single RT-contrast function is obtained. This may reflect the activity of M-cells only. The important point is that sensitivity is reduced with eccentricity. This is due to pre-neural factors such as optics and the anatomical and directionall characteristics of the photoreceptors (Curcio, 199; Banks, Sekuler & Anderson, 1991, Lee, 1996; Malpeli, Lee & Baker, 1996). Note that it has been claimed (Croner & Kaplan, 1995) that contrast gain of P cells increases in the periphery to counteract the blur introduced by optical aberrations. The link between neurophysiology and RTs is tested this idea in fig.18 which compares contrast gain values at different luminances, derived from electrophysiological responses of M and P macaque retinal ganglion cells, with the RT-contrast gain for a range of spatial frequencies. Again we have taken the simplistic view that low spatial frequency RTs (open circles) are mediated by M cells and high spatial frequency RTs (open squares) are mediated by P cells. Medium spatial frequencies show the transition from M-dominated to P-dominated stimuli and the change in the contrast gain, effectively the RT equivalent of contrast gain, is consistent with this. M- and P-cells contrast gains (filled circles and squares, respectively) are re-plotted from Purpura et al. (1988). RTs and contrast sensitivity There is little doubt that RTs are greatly influenced by the contrast sensitivity (gain) of the underlying detection mechanisms. It would be surprising if this were notthe case, but precisely how RTs reflect the processing of supra-threshold contrast is less intuitively obvious. As known the RT paradigm is biased toward transient activity and the traditional methods for assessing supra-threshold perception, contrast matching and magnitude estimation (eg Watanabe et al., 1968; Blakemore, Muncey & Ridley, 1973; Georgeson & Sullivan, 1975) probably reflect slow, sustained-type processes. Figures 15 and 16 illustrate that the overall slope, k, of the RT-contrast functions coincides with the slope, k 2, of the low-contrast branch, which reflects M activity. If the RT-based contrast sensitivity data are qualitatively similar to conventional sensitivity measures, then this is S PLAINIS, VEIC, UNIVERSITY OF CRETE 16