Pursuit in the natural scene

The purpose of early smooth pursuit studies was to characterize the motor limb of the system without contamination by signals introduced by complex visual processing and cognition. To this end, most studies employed a single small spot to minimize the visual complexity of the stimulus and random-motion patterns to minimize target predictability. To understand fully how the pursuit system operates in a natural scene, recent research has been devoted to studying pursuit of visual stimuli that are more complex than the single spot and cognitive factors that contribute to pursuit control. This section will discuss recent research on target selection, gain control, and prediction, as well as work addressing how the pursuit system follows stimuli that are larger and visually more complex than the spot.

Gain Control and Target Selection for Pursuit

It has been suggested that when the pursuit system is confronted with multiple objects, selective attention biases the observer to select a single one in a winner-take-all fashion (Kowler, 1990). Other work supports this contention for certain spatial and temporal target separations (Ferrera and Lisberger, 1997; Recanzone and Wurtz, 1999). Interestingly, during pursuit initiation, a weighted vector average of the speed and direction of both targets appears to describe the data better than does winner-take-all selection (Lisberger and Ferrera, 1997). Vector averaging during pursuit initiation occurs even when an animal is cued to track one of the targets before they move (Ferrera, 2000). In this study, the pre-motion cueing biased the subsequent pursuit response toward the cued target, yet the response remained influenced by the nonrewarded target motion. The results of microstimulation in area MT are consistent with the idea that velocity signals that are read out from this structure are vector averaged for use by the pursuit system (Groh et al., 1997).

An increase in pursuit gain can occur after pursuit is engaged and the observer has selected an object to follow. Modulation of gain was first demonstrated in experiments where brief sinusoidal perturbations were superimposed upon step-ramp target motions (Goldreich and Lisberger, 1987). In this study, higher gain was observed when the perturbation occurred during pursuit than during fixation. Based on these data, the authors suggested that there is a neural “switch” that activates the pursuit system. Later work showed that the gain of the pursuit system with superimposed sinusoidal perturbations varied as a function of target speed (Keating and Pierre, 1996; Schwartz and Lisberger, 1994). Therefore, it appeared that a variable gain element, and not a switch, modified the pursuit system's level of activation.

Normally, smooth pursuit to step-ramp target motion is initiated with a smooth eye acceleration followed by a saccade that places the fovea close to or on the target (Keller and Johnsen, 1990). Lisberger (1998) used the step-ramp paradigm to investigate the interaction between the smooth and saccadic components of pursuit initiation and found that eye velocity just after the saccade was greater than before it. He argued that this “postsaccadic enhancement” resulted from a boost of the gain in the visuomotor pathways. Specifically, the oculomotor system anticipates that the selected target will be near the fovea following the catchup saccade, and the system boosts pursuit gain to ensure that eye speed closely matches target speed at this time. The finding that when two targets are present, heightened eye velocity after the catchup saccade to the foveated target supports the idea that gain control is used during target selection (Gardner and Lisberger, 2001). These data are also evidence that postsaccadic enhancement of pursuit is spatially selective.

Recently, the frontal eye fields (FEF) have been implicated in pursuit gain control (Tanaka and Lisberger, 2001). These authors found that the magnitude of eye velocity induced by electrical stimulation here was greater during pursuit than during fixation. This difference was seen both with and without the sinusoidal perturbations used in earlier studies (e.g., Schwartz and Lisberger, 1994). Stimulation of the supplementary eye field (SEF) also facilitates eye velocity during pursuit (Missal and Heinen, 2001). In this study, stimulated eye velocity increased as a multiplicative function of initial eye velocity, suggesting that, as in the FEF, the current was interacting with a gain element. While the stimulation results in FEF and SEF are evidence that both of these areas are involved in a gain control process, a key difference in the results suggests that these regions have different specific roles in pursuit control. Stimulation in the FEF produces smooth eye movements even during fixation (Gottlieb et al., 1993; Tanaka and Lisberger, 2001), but stimulation of the SEF does not (Missal and Heinen, 2001). This implies that the FEF is in the direct pathway for pursuit movement generation, while the SEF has a more general executive function, controlling the gain of intentional processes.

Pursuit to Larger and More Complex Visual Stimuli

A stimulus larger than a spot will activate a population of motion detectors that tiles a spatially extended region. Since the pursuit system can generate only a single eye velocity vector, the outputs of multiple motion detectors must be combined in this situation to produce one signal. The motion perception system integrates signals that stimulate multiple motion detectors to eliminate noise and to better detect and discriminate spatially extended stimuli (Watamaniuk and Sekuler, 1992). When presented with a random-dot cinematogram (RDC) composed of multiple coherent-motion dots, the pursuit system also appears to integrate motion signals, since pursuit gain is higher, and there is less variability in the response than when a single-spot stimulus is pursued (Heinen and Watamaniuk, 1998) (Fig. 94.1). This finding suggests that the same motion processing substrate limits smooth pursuit and motion perception. The same conclusion was reached when psychophysical judgments and smooth pursuit responses were compared in motion direction discrimination tasks (Beutter and Stone, 1997; Krauzlis and Adler, 2001; Watamaniuk and Heinen, 1999). In all of these experiments, perceptual and motor performances were similar when directional noise was added to moving stimuli. Other work has demonstrated that human observers can integrate the local motion of line-figure objects to derive a global motion signal (Beutter and Stone, 2000). Global information was used similarly to both perceive and pursue these objects, even when partial occlusion made the resulting image motion vastly different from the underlying object motion.

Figure 94.1..  

Eye speed during pursuit of single spots and random-dot cinematograms (RDCs) moving at different speeds. The left column (A and C) shows average eye speed during pursuit of 200 dot RDCs (dot size = 0.1 degree) that moved at 4, 5, 6, 7, and 8 deg/sec for one observer at the top (A) and for a second observer at the bottom (C); the right column (B and D) shows pursuit of a 0.1 degree spot moving at the same five speeds for each of the two observers. Each trace is an average of approximately 10 trials. Traces are aligned on pursuit onset (small vertical arrows). Note that during pursuit of RDCs, observers showed higher gain as well as better speed discrimination, as evidenced by greater separation between the traces. (Modified from Heinen and Watamaniuk, 1998.)

An unresolved issue that surfaces when larger stimuli are used to study pursuit is the degree to which the function and the underlying neural architecture of the pursuit system are distinct from those that subserve other smooth eye movements. One system in question is the one that generates the optokinetic reflex (OKN). The OKN is a response to full-field motion that supplements the vestibulo-ocular reflex (VOR), a nonvisual reflex originating in the semicircular canals. The VOR serves to rotate the eyes to counter head rotation. The distinction between smooth pursuit and OKN is not clear in the literature. Typically, pursuit is studied with a small-spot stimulus and OKN with a large, textured field. However, pursuing a spot can produce eye movement behavior similar to that seen when large fields are used, that is, a recurring pattern of slow and fast phase responses (van den Berg and Collewijn, 1986). Conversely, stimuli larger than a spot can produce eye movements that look like smooth pursuit (Beutter and Stone, 1997; Krauzlis and Adler, 2001; van den Berg and Collewijn, 1986; Watamaniuk and Heinen, 1999). Furthermore, the initial rapid OKN response to visual motion onset is thought to be controlled by the pursuit system (e.g., Cohen et al., 1977; Robinson, 1981), although this notion remains controversial (for a discussion, see Fuchs and Mustari, 1993).

A possible difference between OKN and pursuit is that OKN gain increases with stimulus size up to full field (e.g., Cheng and Outerbridge, 1975; Dubois and Collewijn, 1979; van Die and Collewijn, 1982), which does not appear to be the case for pursuit. In a preliminary study investigating the benefit that larger stimuli provide the pursuit system, it was found that increasing the size of a stimulus to ∼12 degrees increased pursuit gain, but with further increases gain dropped (Watamaniuk and Heinen, 2000). Consistent with this, stimulating the retina beyond the ∼12 degree central region with oppositely directed motion boosted the gain again. These results implied that the receptive field for pursuit, that is, the pursuit field, has a center-surround architecture with an excitatory center and an inhibitory surround. Center-surround receptive fields are common in neurons throughout visual cortex, and neurons with similarly organized motion receptive fields have been found in area MT (Born and Tootell, 1992; Raiguel et al., 1995; Xiao et al., 1995) and could facilitate pursuit when an object moves in the world across a background.

Another type of smooth eye movement response that seems to overlap pursuit (and OKN) to a degree is ocular following. This response has been documented relatively recently and is present in both monkeys (Miles et al., 1986) and humans (Gellman et al., 1990). Ocular following is characterized by smooth eye movements that occur shortly after a large-field visual stimulus begins to move (latency of ∼50 msec in monkeys). That a large stimulus is required to evoke ocular following suggests that it may be more similar to OKN than is pursuit, as pursuit movements can be made to a small spot stimulus. Nevertheless, the optimal retinal stimulation for ocular following appears to have a center-surround organization similar to that of the pursuit field but on a larger spatial scale, with a central excitatory region of ∼20 degrees (Miles et al., 1986). Ocular following also has several properties that distinguish it from OKN as well as from smooth pursuit. First, ocular following is more pronounced when the stimulus is presented just after a saccadic eye movement (Kawano and Miles, 1986). Second, its gain is higher when the central field motion appears in the depth plane where the subject is attending and the motion in the periphery is presented with crossed disparity (Kawano et al., 1994a). This laboratory situation mimics the natural one that occurs when a moving observer attends to a central object and ignores the motion parallax induced by more distant ones.

Despite the differences in the optimal stimuli and responses for pursuit, OKN, and ocular following, the majority of neurons studied in cortical area MST (Kawano et al., 1994b), dorsolateral pontine nucleus (DLPN) (Kawano et al., 1992), and the cerebellum (Gomi et al., 1998) respond to all three types of eye movements. This observation suggests that these eye movement systems may share to a degree a common neural substrate.

Predicting Object Motion

It is well established that the pursuit system can override reflexive behavior and predict target motion when it is periodic and continuous (Deno et al., 1995; Dodge et al., 1930; Stark et al., 1962; Westheimer, 1954; Winterson and Steinman, 1978). After an observer learns the periodicity of such a target, it can be pursued with zero phase lag or even phase lead. However, when that target first moves, the eyes lag behind initially because the first motion is unpredictable, and the pursuit system must rely on a delayed signal arising from the retinal-image motion of the target. Pursuit in the step-ramp paradigm involves multiple discrete trials and can simulate the onset of multiple periodic target motions. Therefore, the step-ramp paradigm might be ideal for isolating signals related to retinal-image motion in the pursuit system. However, data obtained with the step-ramp paradigm must be interpreted carefully, because anticipatory pursuit can precede step-ramp motion even when motion parameters are randomized from trial to trial. Next we discuss some of the forces that guide anticipatory pursuit.

Anticipatory eye velocity is a function of target motion onset timing (Barnes and Asselman, 1991; Barnes and Grealy, 1992; Barnes et al., 1987) and target velocity (Kao and Morrow, 1994; Kowler and McKee, 1987; Kowler et al., 1984; Wells and Barnes, 1998), and can be enhanced by briefly blanking the fixation point (gap paradigm) before the target moves (Boman and Hotson, 1988; Heinen and Liu, 1997). Randomization does not necessarily eliminate anticipatory eye velocity, because the direction or speed of the target in trials that precede the one being studied are stored in short-term memory and can bias the pursuit response (Kowler and McKee, 1987; Kowler et al., 1984). There is also evidence that the timing of target motion onset is remembered from preceding trials and that this memory modifies anticipatory pursuit onset (Heinen et al., 2001). A cognitive expectation about the direction in which a target will move can override short-term memory and guide anticipatory pursuit (Kowler, 1989). As it does during continuous periodic motion, during step-ramp trials the predictive mechanism modulates eye velocity even while the target is moving (Kao and Morrow, 1994; Kowler and McKee, 1987; Kowler et al., 1984), including during the open-loop period which has been studied to understand visual-motion signals that drive pursuit.

There are other examples of cognitive influences on open-loop pursuit. In humans, pursuit speed declines when the subject's attention is directed away from the moving target (Khurana and Kowler, 1987). Also, initial eye acceleration is systematically higher for lower-speed trials inserted in a block of high-speed trials or lower for high-speed trials inserted in a block of lower-speed trials (Carl and Gellman, 1987; Kowler and McKee, 1987; Lisberger and Westbrook, 1985). This phenomenon might be related to short-term memory of previous target motion but, alternatively, could be due to a cognitive expectation about what the speed of the target will be if the observer determines that all the targets in the block have been set to move at the same speed.

Although previously thought to be a special property of the human oculomotor system (Fuchs, 1967), anticipatory pursuit occurs in monkeys as well (Fig. 94.2). Furthermore, as in humans, the predictive mechanism can modulate open-loop pursuit in monkeys when target velocity is randomized (Fig. 94.2C). Since anticipatory and open-loop pursuit behave similarly during target randomization (Heinen et al., 2001), visual-motion processing and short-term memory might use the same neural substrate. This could occur if the activity of neurons in the motion pathway were altered by pursuit experience so that it biased the population response when new motion appeared. However, the visual stimulus can apparently generate a smooth response with different dynamics than anticipatory pursuit. It is often possible to identify two phases in the response, the first of which accompanies the onset of anticipation (Fig. 94.2B). Usually, an abrupt deflection of the eye velocity traces at ∼80 to 100 msec after the target starts to move marks the onset of the second phase, which is characterized by higher acceleration than anticipation onset. That the onset of this second phase is time-locked to the onset of target motion suggests that it reflects the contribution of the visual signal to the pursuit response. In addition, eye acceleration during this second phase depends on retinal-image motion (e.g., Keller and Kahn, 1986; Lisberger and Westbrook, 1985). Therefore, even if the visual-motion pathway is used to generate anticipatory pursuit, it might be possible to determine the relative contribution that the visual signal makes with the proper experimental design. A more detailed discussion of the interactions between processing streams for signals related to visual motion and cognition follows.

Figure 94.2..  

Predictive influences on monkey smooth pursuit. Eye and target position traces (A) and eye and target velocity traces (B) from a block of trials where the monkey tracked a target that always had the same constant velocity motion (50 deg/sec to the right). The target is shown as a solid gray line, individual eye traces as dashed gray lines, and mean eye traces as a solid dark line. Note that the anticipatory eye velocity began to build up before the target moved. Note also the sharp transition to rapid eye acceleration that occurs 100 msec after the target begins to move, consistent with activation by a retinal-image motion signal. C, The effect of target motion history on monkey open-loop pursuit. Data were taken during pursuit of a target that moved at a rate of 50 deg/sec, with the direction randomly chosen on each trial as either left or right. The rightmost node of each plot indicates mean eye velocity for all trials where the target moved right (R) or left (L). The nodes on the left indicate the mean eye velocity of all trials with a given direction that are preceded by a trial with the same or opposite direction. For example, the point designated “R(L)” indicates the eye velocity of all trials where the target moved to the right, given that in the preceding trial target motion was to the left. Note that overall eye velocity was more rightward when the preceding target motion was to the right and more leftward when the preceding motion was to the left. The velocity of the eye was characterized at the end of the open-loop period.