You are here

The Visual and Motor System: An unthought parallel

After accounting for differences in the direction of axon information flow, the organization of the motor system parallels that of the visual system. In other words, information flow starting at the level of individual muscles and traveling backwards to the respective command centers in the motor cortex is conceptually comparable to information flow in the visual system traveling from retinal photoreceptors to higher cortical centers that recognize complex stimuli. Such parallels are likely to have arisen because of similarities in computational strategies between both systems as well as shared evolutionary pressures. In this paper, I will evaluate the evidence for such parallelism in each of the three levels of analysis of the motor system (Graziano, 2006): the properties of single neurons, cortical topography and the mapping from cortex to muscles. Furthermore, I will asses several areas where such parallelism does not seem to exist and possible underlying differences that may account for the divergent changes in system organization

  To conclude, I will propose that the organizational similarities may also be artificially produced given the shared scientific methodology used to research both systems.
At first glance, the organization of the motor and visual systems appears to be highly disjoint. On one hand, the motor system, defined as being the output of the central nervous system, comprises of three different types of pathways —somatic, autonomic, and neuroendocrine— each characterized by a unique type of output (control of the voluntary muscle cells, the viscera, or the hormonal output of the pituitary gland, respectively) (Swanson, 2003). On the other hand, the visual system is just a fraction of the sensory input to the central nervous system and consists mainly of one type of input from retinal photoreceptors. Nevertheless, both systems share one main characteristic: they are both involved in the process of creating or using a sparse internal representation of the external environment. This results in functional parallelism at each stage of processing, from muscle cells to command centers. Furthermore, both systems also share similar evolutionary pressures and possibly use similar underlying computational strategies to create optimal and effective representations.

The organization of the somatic motor system begins with individual muscles. These in turn are controlled by motoneurons via muscle fibers within each particular muscle. Furthermore, these neurons are organized into neuron pools in the spinal cord. Movement is produced by a complex and coordinated pattern of stimulation these motoneuron pools.  Finally, these pools are organized efficiently such that stimulation of one pool can laterally inhibit the corresponding antagonistic muscle. Parallel to this organization, the organization of the visual system begins with individual photoreceptors in the retina. At the first level of organization, input from one or more photoreceptors is sent to retinal ganglion cells. At this level, there is a similar process of lateral inhibition that results in an effective representation of the optical images using center-surround receptive fields. Finally, both systems show a high degree of optimization: the visual system optimizes for space (particularly at the retina), whereas the muscle system optimizes for energy expenditure, which includes smoothness of movement and response times.  For instance, it has been noted that the patterns of tension and relaxation among antagonistic muscles maximizes the efficiency of movement (Swanson, 2003) and that there is an optimization for endpoint smoothness (Bizzi, 1995).

The next aspect of parallel organization among both systems regards the mapping of information. The motoneuron pools in the spinal cord follow a well-known topographical distribution, with pools coding for muscles in the trunk being medial, and progressively becoming more lateral as one reaches the limbs (Swanson, 2003). This is similar to the arrangement of layers in the lateral geniculate nucleus (LGN) of the thalamus following a well defined topographical distribution. Furthermore, the topographical maps used in both systems elicit insight about the underlying function: they are concerned about the information, and not necessarily about the source.  In the motor system, the pattern of activation of nearby muscles is highly relevant, whereas in the visual system, the relevant information comes from a particular location in the visual field, and not necessarily from the particular eye that is the source of the information. Thus, the underlying parallelism concerns the use of a well defined topographical map at this level, but the details of the map may vary depending on the actual information being encoded.

    Similar to visual processing of complex stimuli, the generation of motor commands for highly complex behavioral sequences of well-coordinated movements occurs at the cortex. At this level of organization, representation takes the form of a more complex somatotopic map. Initially, it was believed that the primary motor cortex simply contained a map of muscles and joints very similar to the topographical organization in the ventral horn of the spinal column. However, this map can best be described as “overlapping, intermingled, and fractured”, which suggests that it represents the coordination of muscles from complex actions, rather than the individual movement of joints and muscles (Graziano, 2006). This parallel among both systems continues with the cortex at the level of the individual neuron. Namely, neurons in the motor cortex also exhibit the analog of a ‘receptive field’ of neurons in the primary visual cortex. Instead of showing patterns of activation that correspond to properties of the visual image (Hubel, 1981), they correspond to properties of complex behavioral events. For instance, these cells show a high correlation between the direction of limb movement and neuronal activation, and also show preference for end positions (Graziano, 2006).

As Swanson (2003) suggests, there is still much to be understood regarding the motor networks, particularly the underlying hierarchy and control. This problem becomes more evident as one climbs each step higher in the ladder of motor command control. It is here that the parallelism to the visual system becomes looser. Among the issues to be fully understood include how complex sequences of movements emerge and how they are encoded, as well as the complex interactions among the underlying network. To address the first, it has been suggested that sequences of movements that need to be carried out in a particular order are related to the supplementary motor cortex (SMA) (Graybiel, 1995). Unfortunately, no study has addressed this actually takes places or the actual information needed to command such sequences. In terms of the encoding, Graziano (2006) states that both systems continue to be parallel at this level by suggesting that coding for movement may be analogous to how the inferior temporal cortex (IT) codes visual information: each cell corresponds to a complex motor event and can dynamically change its response patterns depending on what is relevant in terms of the animal’s behavior.

At this level of representation, a commonality among both systems becomes even more relevant and may be one reason why such similarities in organization are observed. Both systems undergo a constant process of adaptation to optimally maintain a sparse yet accurate internal representation of the body and the environment. On one hand, the visual system is in a “continuing process of normalization and calibration” (Gilbert, 1996), which allows it to adapt to environmental conditions and extract effectively fine detail under various lighting and contrast conditions. Such adaptations are observed all throughout the visual pathway, starting with receptive fields that dynamically change their size, to neurons at visual cortex that vary the strength and density of their connections (Gilbert, 2006). On the other hand, the motor system is also highly adapted to reflect behavioral needs of the animal (Graziano, 2006). This ranges from neurons in the striatum which exhibiting context dependent firing patterns (Graybiel, 1995) to optimal control strategies that take into account intricate task-specific parameters (Graziano, 2006).

    Given the nature of the problem solved by each system, it is not unexpected that the parallelism does not extend to all arenas. One area where this is evident is in the feedback control mechanisms of the motor system. Motor feedback can have an instantaneous effect, may involves several learning mechanisms (e.g. unsupervised learning in the cortex, reinforcement learning in the basal ganglia, and supervised learning in the cerebellum). Even though visual cognition is not a passive process, feedback mechanisms in the visual are less evident. For instance, a known interaction between cortex and cerebellum, where the cortex outputs a basic motor command and the cerebellum is in charge of the details and fine control of the movement (Swanson, 2003) is not found in the visual system. Namely, the visual system has no opportunity to influence the input: once a photoreceptor misses visual information, the loss cannot be recovered. Another area where the organization of both systems has to be different in order to ensure survival concerns instinctual patterns of behavior, such as defensive, hand-to-mouth movements and mating behaviors (described in detail in Graziano, 2006). It has been observed that such behaviors can be evoked by stimulating the motor cortex in the macaque brain. There is no such parallel (at least known to humans) regarding a repertoire of instinctual visual programs that are inherent in the cortex and are essential for survival.

    Interestingly, several conclusions reached about the organization of the motor and visual systems result from a particular research methodology. For instance, the researcher stimulates the system (either via electrical stimulation in the motor cortex, or via visual stimuli), and observes the underlying changes as time progresses. Nevertheless, there is one main difference regarding the direction of information flow: whereas motor stimulation results in observable sequences of movements, visual stimulation results in an abstract representation that is not readily accessible. In other words, evoked movement provides an immediate hypothesis about the function of the activated tissue, whereas this is not the case with visual cognition. Although certain areas of the visual system are thought to be involved in extracting patterns of stimuli across time, I believe that such understanding has yet to reach the level of understanding in the motor system. For instance, fully understanding the role of timing would lead to much greater understanding of how learning regarding visual stimuli takes place, such as learning to recognize a particular object, such as faces, based on how it moves across time. Based on my understanding, the mechanisms of learning in the motor system are well understood, unlike in the visual system.

Both the visual and motor systems are constantly adapting to our environment, and share several features at all levels of analysis, starting from individual neurons, to more complex computational strategies. This approach of reversing the direction of information flow, similar to reverse engineering of the motor system, furthers our understanding of both systems by gaining insight into how the brain stores internal representations. By understanding the parallels and differences in organization, we get closer to understanding the basic principles of brain architecture in a world with demanding evolutionary pressures that demand complex yet optimal organization.