Information processing in the primate visual system


  • Paul R. Martin,

    1. ARC Centre of Excellence in Vision Science, School of Medical Sciences, and Save Sight Institute, University of Sydney, 8 Macquarie St, Sydney, NSW 2001, Australia
    Search for more papers by this author
  • Samuel G. Solomon

    1. ARC Centre of Excellence in Vision Science, School of Medical Sciences, and Save Sight Institute, University of Sydney, 8 Macquarie St, Sydney, NSW 2001, Australia
    Search for more papers by this author


Imagine that you are a medical student in the year 1985. Visual physiology is well-understood, and the subject is taught as a two-lecture narrative. In the first lecture you learn that light is transduced by photoreceptors in the retina, how lateral inhibition in the retina creates centre–surround receptive fields, and that visual information is sent from the eye along two pathways called parvocellular and magnocellular (Fig. 1A). The parvocellular cells (PC) signal colour and fine detail in the visual world, whereas magnocellular cells (MC) signal movement and visual distance. Their signals pass to the primary visual cortex (V1) though a dorsal thalamic relay called the lateral geniculate nucleus (LGN).

Figure 1.

Traditional view and modern additions to understanding visual processing in primates
A, traditional schematic view of the main feed-forward pathways. Two classes of ganglion cell in the retina send signals to the dorsal thalamic relay nucleus (lateral geniculate nucleus, LGN). Relay neurones in the parvocellular (PC) and magnocellular (MC) pathways project to primary visual cortex (V1) where some mixing of their signals probably occurs. Signals from V1 pass to two series of association cortical areas: areas in the dorsal series analyse the movement and spatial arrangement of visible objects (‘where’); areas in the ventral inferotemporal series allow object recognition and categorization (‘what’). B, traditional view of signal processing in V1. Circular concentric receptive fields from a line of LGN afferents with neighbouring receptive fields are combined as inputs to a cortical neurone (red silhouette). The envelope of the input receptive fields can signal an ‘edge-detection event’ for a narrow range of edge orientations. C, modern view of non-standard afferents to visual cortices. In addition to PC and MC pathways, small numbers of ganglion cells with heterogeneous properties project to the koniocellular (KC) layers of the LGN as well as to the superior colliculus (SC) and lateral posterior–pulvinar complex (LP). The cells in these areas project directly or indirectly to association cortices and/or to V1. D, schematic view of feedback pathways. Association cortices send reciprocal cortico-cortical signals, to lower-order areas, and cortical visual centres send signals to subcortical visual areas. E, schematic view of signal processing in primary visual cortex. Populations of cortical neurones serving a region of visual space are reciprocally connected by local axon collaterals to generate edge detection by mutual excitation. Each cell contributes to feed-forward projections to other cortical areas and to feedback projections to the thalamus, and the cells receive top-down projections from other areas. Response of each cell thus depends not only on afferent convergence but also on the state of the local network and the activity of other cortical areas.

Lecture two of our imaginary series teaches you that thalamic afferents are recombined in V1 to supply ‘edge detector’ neurones (Fig. 1B). The outputs of these neurones, together with outputs of colour detectors and motion detectors, feed cortical pathways that generate increasing levels of selectivity. At the end of this process, neurones come to represent the identity (‘what’) and position (‘where’) of visible objects. The comfort of the narrative is disturbed only when the lecturer raises a warning finger: some questions do remain open. Does this serial process culminate in ‘grandmother detector’ cells? And how can one discover where grandmother is standing, if her position is signalled by a different neurone in quite a different part of the brain?

Today, a series of lectures on vision would keep much of the data outlined above, and would incorporate many new facts. Most importantly, discoveries have been made that change our two-channel, serial processing view of vision. We now know that in addition to parvocellular and magnocellular pathways, more than a dozen smaller pathways leave the eye. Neurones in some of these pathways show properties as complex as those of cortical neurones, and their signals can reach visual cortical areas along a variety of routes, including the koniocellular (KC) layers of the LGN, the superior colliculus and the lateral pulvinar complex (Fig. 1C). We have learned, despite formidable experimental obstacles, much more about the function of feedback pathways and their ‘top-down’ influence from higher cortical areas to V1, and from V1 to the thalamus (Fig. 1D). The feed-forward depiction of signal processing in V1 has been updated to view each cortical neurone as part of an interconnected local network. The local network gets top-down signals from other cortical areas, and itself sends top-down signals to the thalamus (Fig. 1E). The grandmother cell does not speak alone: her voice is part of a chorus of activity across highly connected cortical circuits.

The review and research papers in this issue address only a few of many important questions on primate vision. In a review of subcortical pathways and their relation to psychophysically defined ‘channels’ of human vision, Lee (2011) shows how attribution of high-acuity and colour vision to PC neurones, and motion vision to MC neurones, is overly simplistic: the spatial acuity of PC cells is only marginally better than the acuity of MC cells, some colour signals travel in the KC pathways, and psychophysically characterised visual channels have much worse time resolution than the subcortical pathways that feed them. A related research paper by Lee et al. (2011) shows how combined colour and brightness modulation is ‘handled’ by PC and MC pathways; here activity in the PC pathway is found wanting as a source of high-acuity brightness signals.

The independence and diversity of signals leaving the retina is addressed by Greschner et al. (2011) who use multi-electrode recording to identify in vitro simultaneous activity of hundreds of output neurones (ganglion cells) in macaque retina. They tame a behemoth of data to find that neighbourhoods of like-type cells show correlated activity that probably arises from fluctuations of transmitter release in photoreceptors. As the intrinsic correlations are preserved across uniform or patterned light stimulation they also bring a general result: the functional circuitry generating shared activity does not change when the retina moves from a uniform state to a dynamic one.

The thalamus is where retinal signals are first brought into contact with signals from other parts of the brain – from mid-brain systems controlling arousal and sleep, and from visual structures such as the visual cortex and superior colliculus. The functional properties of cortical feedback to the thalamus are growing clearer, as Briggs & Usrey (2011) outline in this issue. Surprisingly it seems that despite mixing of afferent channels at the cortex, the feedback pathways show signatures of the afferent PC, MC and KC streams. A research paper by the same authors (Alitto et al. 2011) addresses the related question of how anaesthesia changes temporal tuning properties in the LGN. Few systematic studies of the functional properties of cells in awake and anaesthetized animals have been made at any level of the visual system. Briggs & Usrey make one of the closest comparisons to date, taking specific care to keep stimulus conditions as comparable as possible in waking and anaesthetised conditions. They show that the main temporal tuning properties of LGN cells are preserved under anaesthesia, with some reduction in sensitivity and high-frequency response. The enigmatic problem that remains is the drastic difference in temporal tuning of single neurones and the ‘low-pass’ temporal properties of human vision pointed out by Lee (2011), and others (Williams et al. 2004).

Studies of the anatomical organisation of cortical feedback pathways are slowly being given functional relevance, both from functional imaging in humans and from single-neurone studies of waking primates. A long-standing hypothesis is that feedback pathways can allow attention to modulate visual sensation. Our knowledge of neural bases of attention has advanced substantially in recent years, and here Bisley (2011) reviews how the brain allocates visual attention, and how this attention-related signal interacts with the incoming sensory signal.

It should not be surprising that our view of the visual system has evolved across 25 years of research and, likewise, unsurprising that much remains to be done. The simple models of the last century cannot account for visual sensations, yet still provide crucial keys to understanding. New knowledge has changed our view: from vision as series of steps, to vision as a product of activity in cortical and subcortical networks. But it remains obvious that today's models provide (at best) only a slightly less rudimentary account of visual sensations than their forebears.