Ultrafast initiation of a neural race by impending errors.

KEY POINTS: The brain makes decisions by means of races between neural units representing alternative choices. In the present study, we record the eyemovements made in the Wheeless task, when a visual stimulus is followed after a short delay by another stimulus demanding a different response. The behaviour can be very precisely described as a race between three independent decision processes: one Go process for each of the responses, and a Stop process that tries to cancel the first, now erroneous, response. To explain the high success rate for cancellation that we observe, the onset time for the Stop process must be some 10-20 ms shorter than for Go. As well as extending our understanding of the dynamics of complex decision-making, this task provides a rapid, non-invasive method for quantifying disorders of higher neural function. ABSTRACT: The brain makes decisions by means of races between neural units representing alternative choices, and such models can predict behaviour in decision tasks in a precisely quantitative way. But what is less clear is how soon after the stimulus the race actually starts. In the present study, we re-visit a complex decision experiment: the Wheeless task, in which a saccadic stimulus is followed after a short delay by a second stimulus, with the subject sometimes making a saccade to the first, now inappropriate, stimulus, and sometimes going straight to the correct one. We demonstrate that a simple model with three accumulator units, two 'Go' and one 'Stop', can then account in detail for the individual responses made, as well as their timing. This complex decision-making behaviour is predicted directly for each individual subject by their performance in a simple step saccadic task, which identifies the two free parameters that are specific for each subject. By contrast to previous assumptions, we find that it is necessary for the onset time of the Stop unit to be shorter than for Go by 10-20 ms. This suggests a specifically fast mechanism for altering responses in situations where urgent action is needed to prevent an impending error.

Saccadic foraging: reduced reaction time to informative targets.

The study of saccadic reaction times has revealed a great deal about the neural mechanisms underlying neural decision, in terms of Bayesian factors such as prior probability and information supply. In addition, recent work has shown that saccades are faster to visual targets associated with conventional monetary or food rewards. However, because the purpose of saccades is to acquire information, it could be argued that this is an unnatural situation: the most natural and fundamental reward is the amount of information supplied by a target. Here, we report the results of a study investigating the hypothesis that a saccade to a target whose colour provides information about the location of a subsequent target is faster than to one that does not. We show that the latencies of saccades to a location that provides reliable information about the location of a future target are indeed shorter, their distributions being shifted in a way that implies that the rate of rise of the underlying decision signal is increased. In a race between alternative targets, this means that expected information will be an important factor in deciding where to look, so that 'foraging' saccades are more likely to be made to useful targets.

Using saccades to diagnose covert hepatic encephalopathy.

Covert Hepatic Encephalopathy (CHE), previously known as Minimal Hepatic Encephalopathy, is a subtle cognitive defect found in 30-70 % of cirrhosis patients. It has been linked to poor quality of life, impaired fitness to drive, and increased mortality: treatment is possible. Despite its clinical significance, diagnosis relies on psychometric tests that have proved unsuitable for use in a clinical setting. We investigated whether measurement of saccadic latency distributions might be a viable alternative. We collected data on 35 cirrhosis patients at Addenbrooke's Hospital, Cambridge, with no evidence of clinically overt encephalopathy, and 36 age-matched healthy controls. Performance on standard psychometric tests was evaluated to determine those patients with CHE as defined by the World Congress of Gastroenterology. We then compared visually-evoked saccades between those with CHE and those without, as well as reviewing blood test results and correlating saccadic latencies with biochemical parameters and prognostic scores. Cirrhosis patients have significantly longer median saccadic latencies than healthy controls. Those with CHE had significantly prolonged saccadic latencies when compared with those without CHE. Analysis of a cirrhosis patient's saccades can diagnose CHE with a sensitivity of 75 % and a specificity of 75 %. We concluded that analysis of a cirrhosis patient's saccadic latency distributions is a fast and objective measure that can be used as a diagnostic tool for CHE. This improved early diagnosis could direct avoidance of high-risk activities such as driving, and better inform treatment strategies.

Re-starting a neural race: anti-saccade correction.

In the anti-saccade task, a subject must make a saccadic eye movement in the opposite direction from a suddenly-presented visual target. This sets up a conflict between the natural tendency to make a pro-saccade towards the target and the required anti-saccade. Consequently there is a tendency to make errors, usually corrected by a second movement in the correct anti-saccade direction. In a previous paper, we showed that a very simple model, with racing LATER (Linear Approach to Threshold at Ergodic Rate) units for the pro- and anti-directions, and a stop unit that inhibits the impending error response, could account precisely for the detailed distributions of reaction times both for correct and error responses. However, the occurrence and timing of these final corrections have not been studied. We propose a novel mechanism: the decision race re-starts after an error. Here we describe measurements of all the responses in an anti-saccade task, including corrections, in a group of human volunteers, and show that the timing of the corrections themselves can be predicted by the same model with one additional assumption, that initiation of an incorrect pro-saccade also resets and initiates a corrective anti-saccade. No extra parameters are needed to predict this complex aspect of behaviour, adding weight to our proposal that we correct our mistakes by re-starting a neural decision race. The concept of re-starting a decision race is potentially exciting because it implies that neural processing of one decision can influence the next, and may be a fruitful way of understanding the complex behaviour underlying sequential decisions.

Target direction rather than position determines oculomotor expectation in repeating sequences.

Saccadic latencies to targets appearing to the left and right of fixation in a repeating sequence are significantly increased when a target is presented out of sequence. Is this because the target is in the wrong position, the wrong direction, or both? To find out, we arranged for targets in a horizontal plane occasionally to appear with an unexpected eccentricity, though in the correct direction. This had no significant effect on latency, unlike what is observed when targets appeared in the unexpected direction. That subjects learnt sequences of directions rather than simply positions was further confirmed in an experiment where saccade direction was a repeating sequence, but eccentricity was randomised. Latency was elevated when a target was episodically presented in an unexpected direction. Latencies were also elevated when targets appeared in the correct hemifield but at an unexpected direction (35° polar angular displacement from the horizontal, a displacement roughly equivalent in collicular spacing to our unexpected eccentricity), although this elevation was of a smaller magnitude than when targets appeared in an unexpected direction along the horizontal. Finally, we confirmed that not all changes in the stimulus cause disruption: an unexpected change in the orientation or colour of the target did not alter latency. Our results show that in a repeating sequence, the oculomotor system is primarily concerned with predicting the direction of an upcoming eye movement rather than its position. This is consistent with models of oculomotor control developed for randomly appearing targets in which the direction and amplitude of saccades are programmed separately.

Antisaccades as decisions: LATER model predicts latency distributions and error responses.

Antisaccades are widely used in the study of voluntary behavioural control: a subject told to look in the opposite direction to a stimulus must suppress the automatic response of looking towards it, leading to delays and errors that are commonly believed to be generated by competing decision processes. However, currently we lack a precise model of the details of antisaccade behaviour, or indeed detailed quantitative data in the form of full reaction time distributions by which any such model could be evaluated. We measured subjects' antisaccade latency distributions and error rates, and found that we could account precisely for both distributions and errors with a model having three competing LATER processes racing to threshold. In an even more stringent test, we manipulated subjects' expectation of the stimulus, leading to large changes in behaviour that were nevertheless still accurately predicted. The antisaccade task is widely used in the laboratory and clinic because of the relative complexity and vulnerability of the underlying decision mechanisms: our model, grounded in detailed quantitative data, is a robust way of conceptualizing these processes.

Altered interictal saccadic reaction time in migraine: a cross-sectional study.

AIMS: The underlying mechanisms of migraine remain poorly understood, partly because we lack objective methods for quantitative analysis of neurological function. To address this issue, we measured interictal saccadic latency in migraineurs and controls. METHODS: In a cross-sectional study, we compared interictal saccadic latency distributions of 12,800 saccades in 32 migraineurs with 32 age- and sex-matched controls. RESULTS: The variability of migraineurs' reaction time distributions was significantly smaller (σ = 1.01 vs. 1.13; p < 0.05) compared with controls. In addition, a smaller proportion of migraineurs generated 'early' saccades (31% vs. 56%: p < 0.05). Sensitivity/specificity analysis demonstrated the potential benefit of this technique to diagnostic discrimination. CONCLUSIONS: The migraineur's brain behaves significantly differently from that of a control during the interictal period. By analysing whole distributions, rather than just means, data can be related directly to current neurophysiological models: specifically, the observed decrease in variability suggests a functional deficit in the noradrenergic systems influencing the cerebral cortex. From a clinical perspective, this novel method of characterising neurological function in migraine is more rapid, practicable, inexpensive, objective and quantitative than previous methods such as evoked potentials and transcranial magnetic stimulation, and has the potential both to improve current diagnostic discrimination and to help guide future research into the underlying neural mechanisms.

Directional prediction by the saccadic system.

One popular and fruitful approach to understanding what influences the decision of where to look next has been to present targets in a series of trials either to the right or left of a central fixation point and examine sequential effects on saccadic latency. However, there is a problem with this paradigm: Every saccade to a target is necessarily followed by an equal and opposite movement back to the center, yet the potentially confounding influence of this refixation saccade is rarely considered. Here, we introduce a novel random-walk paradigm that eliminates this difficulty. Each successive target appears to the left or right of the previous one, allowing us to study long sequences of saccades uncontaminated by refixations. This exposes a new stimulus-history effect, which is remarkably prolonged and relates primarily to movement direction: A saccade reduces the latency for subsequent movements made in the same direction and retards those in the opposite direction. Although in conventional refixation paradigms this effect cancels out, it is of particular significance in the real world--where our fixation point shifts constantly with the object of interest--and reflects a prediction of the way that real objects typically move.

Full reaction time distributions reveal the complexity of neural decision-making.

Measurement of the stochastic distribution of reaction time or latency has become a popular technique that can potentially provide precise, quantitative information about the underlying neural decision mechanisms. However, this approach typically requires data from large numbers of individual trials, in order to enable reliable distinctions to be made between different models of decision. When data are not plentiful, an approximation to full distributional information can be provided by using a small number of quantiles instead of full distributions - often, just five are used. Although this can often be adequate when the proposed underlying model is a relatively simple one, we show here that, with more complex tasks, and correspondingly extended models, this kind of approximation can often be extremely misleading, and may hide important features of the underlying mechanisms that only full distributional analysis can reveal.

Dietary treatment of phenylketonuria: the effect of phenylalanine on reaction time.

There is no evidence that high phenylalanine (Phe) levels have irreversible effects on the adult brain. Many adults with phenylketonuria (PKU) no longer follow a protein-restricted diet. Neuropsychological studies have shown that reaction time in adults with PKU is slower than controls. There are no data to show that this is directly related to Phe levels. Another way to assess reaction time is to measure saccadic latency. We have used a portable, head-mounted saccadometer to measure latency in the outpatient setting. Patients with PKU were split into three groups: off-diet (Phe >1,200 µmol/l), on-diet (Phe < 800 µmol/l) and maternal diet (Phe 100-400 µmol/l ). Reciprocal median latency (RML) was compared between groups. Latency was significantly slower in patients who were off-diet than in patients on-diet, on a maternal diet or in normal controls. Reaction times in both diet-treated groups were not significantly different from normal controls. In 16 women planning pregnancy we obtained values before and after they commenced the maternal diet. Stricter control of Phe levels resulted in a significant improvement in reaction times. We conclude that saccadometry is useful in monitoring PKU patients. Adult patients with PKU not on a protein-restricted diet have significantly slower reaction times than controls. In addition, off-diet patients have significantly slower reaction times than on-diet. Paired data show that effects of Phe levels on reaction time are reversible.

Predicting the timing of wrong decisions with LATER.

Response time, or latency, is increasingly being used to provide information about neural decision processes. LATER (Linear Approach to Threshold with Ergodic Rate) is a quasi-Bayesian model of decision-making, with the additional feature that it introduces a degree of gratuitous randomisation into the decision process. It has had some success in predicting latencies under various conditions, but has not specifically been applied to an equally important aspect of decision-making, namely errors: a complete model of decision-making should not only account for latency distributions of correct decisions but also of wrong ones. We therefore used a decision task that generates large numbers of errors: subjects are told to look at suddenly appearing targets of one colour, but not another. We found that subjects' faster responses are as likely to be correct as wrong, but eventually the latency distributions diverge, with errors becoming infrequent. It seems that colour information, arriving after a delay, results both in cancellation of the developing response to the mere existence of the target and in delayed initiation of the correct response. A simple model, using LATER units in a similar way to one that has previously successfully modelled countermanding, accurately predicts latency distributions and proportions of all responses, whether correct or incorrect, demonstrating that the LATER model can indeed account for errors as well as correct responses.

Modelling Prosaccade Latencies across Multiple Decision-Making Tasks.

Oculomotor decision making can be investigated by a simple step task, where a person decides whether a target has jumped to the left or the right. More complex tasks include the countermanding task (look at the jumped target, except when a subsequent signal instructs you not to) and the Wheeless task (where the jumped target sometimes then quickly jumps to a new location). Different instantiations of the LATER (Linear Approach to Threshold with Ergodic Rate) model have been shown to explain the saccadic latency data arising from these tasks, despite it being almost inconceivable that completely separate decision-making mechanisms exist for each. However, these models have an identical construction with regards to predicting prosaccadic responses (all step task trials, and control trials in countermanding and Wheeless tasks, where no countermanding signal is given or when the target does not make a second jump). We measured saccadic latencies for 23 human observers each performing the three tasks, and modelled prosaccade latencies with LATER to see if model parameters were usefully preserved across tasks. We found no significant difference in reaction times and model parameters between the step and Wheeless tasks (mean 175 and 177 ms, respectively; standard deviation, SD 22 and 24 ms). In contrast, we identified prolonged latencies in the countermanding tasks (236 ms; SD 37 ms) explained by a slower rise and an elevated threshold of the decision making signal, suggesting elevated participant caution. Our findings support the idea that common machinery exists for oculomotor decision-making, which can be flexibly deployed depending upon task demands.

Saccadic latency in hepatic encephalopathy: a pilot study.

Hepatic encephalopathy is a common complication of cirrhosis. The degree of neuro-psychiatric impairment is highly variable and its clinical staging subjective. We investigated whether eye movement response times-saccadic latencies-could serve as an indicator of encephalopathy. We studied the association between saccadic latency, liver function and paper- and pencil tests in 70 patients with cirrhosis and 31 patients after liver transplantation. The tests included the porto-systemic encephalopathy (PSE-) test, critical flicker frequency, MELD score and ammonia concentration. A normal range for saccades was established in 31 control subjects. Clinical and biochemical parameters of liver, blood, and kidney function were also determined. Median saccadic latencies were significantly longer in patients with liver cirrhosis when compared to patients after liver transplantation (244 ms vs. 278 ms p < 0.001). Both patient groups had prolonged saccadic latency when compared to an age matched control group (175 ms). The reciprocal of median saccadic latency (µ) correlated with PSE tests, MELD score and critical flicker frequency. A significant correlation between the saccadic latency parameter early slope (σ(E)) that represents the prevalence of early saccades and partial pressure of ammonia was also noted. Psychometric test performance, but not saccadic latency, correlated with blood urea and sodium concentrations. Saccadic latency represents an objective and quantitative parameter of hepatic encephalopathy. Unlike psychometric test performance, these ocular responses were unaffected by renal function and can be obtained clinically within a matter of minutes by non-trained personnel.

A simple two-stage model predicts response time distributions.

The neural mechanisms underlying reaction times have previously been modelled in two distinct ways. When stimuli are hard to detect, response time tends to follow a random-walk model that integrates noisy sensory signals. But studies investigating the influence of higher-level factors such as prior probability and response urgency typically use highly detectable targets, and response times then usually correspond to a linear rise-to-threshold mechanism. Here we show that a model incorporating both types of element in series - a detector integrating noisy afferent signals, followed by a linear rise-to-threshold performing decision - successfully predicts not only mean response times but, much more stringently, the observed distribution of these times and the rate of decision errors over a wide range of stimulus detectability. By reconciling what previously may have seemed to be conflicting theories, we are now closer to having a complete description of reaction time and the decision processes that underlie it.

Dual LATER-unit model predicts saccadic reaction time distributions in gap, step and appearance tasks.

Saccadic latencies have long been known to depend on the relative timing of the appearance of the new target, and offset of the original fixation target. Previous studies have tended to conclude that two separate effects are at work, one equivalent to competitive inhibition from the fixation target, and the other due to its offset providing a warning that shortens latency. In this study, we propose a simpler explanation, based on a well-established model of reaction time, LATER (linear approach to threshold with ergodic rate), that in addition to predicting mean latencies also--more challengingly--predicts latency distributions. We show that observed distributions, using gap, step and appearance tasks under three conditions of prior probability, can be accurately predicted by using a pair of LATER units, one corresponding to fixation target offset and the other to peripheral target onset. Because fixation offset is probabilistically associated with target appearance, when the fixation unit is activated it increases the target's decision signal (that represents probability) in a fixed proportion, speeding responses. In contrast, when the fixation target remains present, the fixation unit is not activated, and responses are slower. Both these effects generate characteristic changes in the shapes of the latency distributions that can be accurately predicted by the model.

Visual pursuit: an instructive area of cortex.

Recent experiments have revealed an area of visual cortex that provides a velocity error signal which enables the eye to learn to pursue targets when they move in a predictable way.

The effect of stimuli that isolate S-cones on early saccades and the gap effect.

Disappearance of the fixation spot before the appearance of a peripheral target typically reduces average saccadic reaction times (the gap effect) and may also produce a separate population of early or express saccades. The superior colliculus (SC) is generally believed to be critically involved in generating both effects. As the direct sensory input to the SC does not encode colour information, to determine whether this input was critical in generating the gap effect or express saccades we used coloured targets which this pathway cannot distinguish. Our observers still made early saccades to colour-defined targets, but these were anticipations in response to the offset of the non-coloured fixation target. We also show that a gap effect still occurs when either the fixation target or the peripheral target is colour defined, suggesting that direct sensory input to the SC is not required and that information about the location of colour-defined targets is abstracted prior to processing within the SC.

A single mechanism for the timing of spontaneous and evoked saccades.

The study of saccadic latency-the reaction time between presenting a visual stimulus and initiating an eye movement to look at it--has led to a better understanding of decision mechanisms in general, through the development of quantitative models such as LATER. But outside the laboratory, evoked saccades of this kind are rare. Most saccades are made spontaneously while viewing static scenes. Can their initiation be explained by the same decision mechanism? We suggest that in a series of spontaneous saccades, each can be considered to be evoked by the new retinal image generated by its predecessor, so that the intersaccadic interval (ISI) can be regarded as equivalent to latency. We measured ISIs in subjects spontaneously viewing static scenes, and found their distributions to be qualitatively similar to those of evoked saccades, differing quantitatively in just two respects: (1) the main part of the distribution is slower; and (2) there is an increased number of very early responses. By using novel saccadic tasks we show that (1) can be accounted for by lateral inhibition between multiple, suddenly presented image elements, and (2) by the fact that the stimulus is necessarily extremely predictable. Adding these two factors to an evoked saccadic task produced latency distributions indistinguishable from those of spontaneous ones. This suggests that the mechanisms generating evoked and spontaneous movements may be less functionally distinct than is commonly assumed. Both clinically and scientifically, a common, unified framework for explaining both spontaneous and evoked movements is an exciting prospect.

Supplementary eye field: keeping an eye on eye movement.

The supplementary eye field has the biggest say in choosing what we look at, but has long been an enigma. Recent studies are beginning to make more sense of what it actually does.

Contrast, probability, and saccadic latency; evidence for independence of detection and decision.

Many factors influence how long it takes to respond to a visual stimulus. The lowest-level factors, such as luminance and contrast, determine how easily different elements of a target can be detected. Higher-level factors are to do with whether these elements constitute a stimulus requiring a response; they include prior probability and urgency. It is natural to think of these two processes, detection and decision, as occurring in series, so that overall reaction time is essentially the sum of the contributions of each stage. Here, measurements of saccadic latency to visual targets whose contrast and prior probability are systematically manipulated demonstrate that there are indeed separable stages of detection and decision. Both can be quantitatively described by rise-to-threshold mechanisms; the average rate of rise of the first is a simple logarithmic function of target contrast, whereas the second shows the linear rise characteristic of the LATER model of neural decision making. The implication is that under normal, high-contrast conditions, in which detection is very fast, the random variability that is characteristic of all reaction times is not caused by sensory noise but is gratuitously introduced by the brain itself; paradoxically, by conferring unpredictability it may aid an organism's survival.