Proceedings of The Physiological Society

Physiology 2012 (Edinburgh) (2012) Proc Physiol Soc 27, PC252

Poster Communications

Estimating human contrast-dependent visual delay: a new approach using saccadic competition

N. Nesaratnam1, I. Weinberg1, R. Carpenter1

1. University of Cambridge, Cambridge, United Kingdom.


  • Figure 1. A: Relation between proportion of responses made in the direction of the first, lower-contrast, target as a function of the delay of the second target, for one subject at different contrast levels; the lines are best linear fits (maximum likelihood), constrained to be parallel. B: Observed delays, for all subjects, as estimated from data as shown in A, plotted as a function of the delay predicted for the corresponding contrast, calculated from the existing formula (see text) using best-fit values of the parameters K and C0 for each subject. The dashed line shows the expectation for perfect prediction.

The study of reaction times has become a fundamental tool for studying neural decision-making, and quasi-Bayesian models such as LATER1 can give a good account of latencies in response to high-contrast targets. But when contrast is nearer threshold, the time taken for local detection of the target must also be taken into account; a complete model of visual reaction time requires two stages: a detection stage embodying a random walk, and a decision stage that follows LATER in having a linearly-rising decision signal2. To test this idea, we need more information about the time taken in the detection stage. Previous work, looking at simple reaction time as a function of target contrast, found that the average time taken for detection at contrast C was of the form K/log(1 + C/C0), where K and C0 are respectively latency-scaling and contrast-scaling factors3, 4. However, this method is imprecise and slow to generate data. Here, we show that an alternative approach can provide quantitative data of this kind more quickly and more precisely. Subjects completed a precedence task5, in which stimuli appear to both the left and right, and the subject chooses which to look at, whilst their eye movements were recorded and saccades registered, using standard techniques3. The two targets appear asynchronously, with the first-appearing being lower contrast than the second (which was 75% contrast), making it harder - and hence slower - to detect. Fig. 1a shows how increasing the interstimulus delay increases the probability of one subject looking at the first target to appear but reducing the contrast does the opposite. Our graph allows us to ‘titrate' this balance: there is a ‘50% point', at which the subject is equally likely to look at either stimulus, which can be estimated by linear interpolation. Comparing these at different contrasts allows us to quantify the extra decision time taken by the detection stage as a result of the reduced contrast. Using a precedence task exaggerates these differences because there are two LATER units which come close to threshold: they therefore inhibit one another. For each subject we can estimate K and C0 from such data, and then see how well the latency-predicting formula models the observed data. The average value of C0 across all subjects (4.94 ± 0.024%) was similar to what had previously been reported3, 4; the average K value (77.6 ± 8.1ms) is substantially higher, reflecting the enhancing influence of lateral inhibition in this competitive task. The technique appears to generate reasonably consistent results: Fig. 1b shows a comparison of observed and predicted reaction times across all subjects (Pearson R = 0.962, p < 0.0001); but they suggest that the original formula may need some reconsideration. We hope to use this method to examine the discrepancy in more detail.

Where applicable, experiments conform with Society ethical requirements