In our eyes, cells called photoreceptors convert the world around us into a pixel-like representation. Our brains must then reorganize this into a representation that reflects the identities of the objects we are looking at. The same object can be represented by very different pixel patterns, depending on its distance from us, the viewing angle and the lighting conditions.
Conversely, different objects can be represented by pixel patterns that are similar. This is what makes object recognition a tremendously challenging problem for our brains to solve, and we do not fully understand how our brains manage to recognize objects.
Nonhuman primates (such as rhesus monkeys) are routinely used to study object recognition because their brains are similar to ours in many ways. However, there are advantages to working with mice and rats, including access to an array of modern biotechnological tools that have been optimized for these species.
These tools include sophisticated ways to measure neural activity (Svoboda and Yasuda, 2006), to manipulate neural activity (Fenno et al., 2011), and to map how neurons are connected together within and between brain areas (Oh et al., 2014).
Skepticism that rodents could be used to gain insight into object recognition has largely been targeted at the ways in which rodent visual systems deviate from our own. For example, the retinae of mice and rats are specialized for seeing in the dark, and they lack a region called the fovea that allows humans to see objects in great detail at the center of the gaze.
The visual cortex is also organized differently in primates and rodents with regard to how neurons with similar preferences for visual stimuli are clustered together within each brain area, and a much smaller fraction of the rodent cortex is devoted to visual processing. In light of all of these differences, can we really learn much about how our brains recognize objects by studying how rodents see?
Rat Object Recognition
In an earlier study, Davide Zoccolan and colleagues presented behavioral evidence that rats are capable of identifying objects under variations in viewing conditions (Zoccolan et al., 2009).
Now, in eLife, Zoccolan and co-workers at SISSA in Trieste, the Istituto Italiano di Tecnologia and Harvard Medical School – including Sina Tafazoli and Houman Safaai as joint first authors – present evidence that this behavior is supported by four visual areas of the brain that are arranged in a functional hierarchy (Tafazoli et al., 2017). This is analogous to how object processing happens in the primate brain (DiCarlo et al., 2012).
Researchers had previously relied on anatomical evidence to argue that visual brain areas in rats are organized in a hierarchical fashion (Coogan and Burkhalter, 1993). Tafazoli et al. recorded the activity of four of these areas – termed V1, LM, LI and LL – in response to different objects as they systematically changed a number of variables (such as the position, size and luminance of each object). With this data, they quantified how much information each brain area reflected about the identity of the object, as well as how that information was formatted.
[caption id=“attachment_90446” align=“aligncenter” width=“2087”] Experimental design.
(A) Oblique insertion of a single-shank silicon probe in a typical recording session targeting rat lateral extrastriate areas LM, LI and LL, located between V1 and temporal association cortex (TeA). The probe contained 32 recording sites, spanning 1550 µm, from tip (site 1) to base (site 32). The probe location was reconstructed postmortem (left), by superimposing a bright-field image of the Nissl-stained coronal section at the targeted bregma (light gray) with an image (dark gray) showing the staining with the fluorescent dye (red), used to coat the probe before insertion.
(B) The stimulus set, consisting of ten visual objects (top) and their transformations (bottom).
(C) Firing intensity maps (top) displaying the RFs recorded along the probe shown in (A). The numbers identify the sites each unit was recorded from.
Tracking the retinotopy of the RF centers (bottom: colored dots) and its reversals (white arrows) allowed identifying the area each unit was recorded from. DOI: http://dx.doi.org/10.7554/eLife.22794.003[/caption]
A key insight came from analyzing the degree to which changes in the neural responses to different objects could be attributed to differences in object luminance as opposed to object shape. Compared to the other brain areas, the firing rate of the neurons in V1 (the first brain area in the hierarchy) depended more strongly on the amount of luminance within the region of the visual field that each neuron was sensitive to.
Moving through the hierarchy, an increasingly large proportion of the responses of the neurons reflected information about the shape of the object. At the same time, there was a systematic increase in the degree to which information about object identity was formatted in a manner that would make it easy for higher brain areas to access this information (DiCarlo and Cox, 2007).
In the face of considerable evidence that object processing in rats and primates is different, Tafazoli et al. have uncovered a compelling similarity. By design, their study has strong parallels with the studies that established a hierarchy for object processing in the primate brain, and their results suggest that rats and primates may perform object recognition in broadly similar ways.
Future work will be required to determine the degree to which the nuts-and-bolts of object processing are in fact the same between the species.
Tafazoli S, Safaai H, De Franceschi G, Rosselli FB, Vanzella W, Riggi M, Buffolo F, Panzeri S, Zoccolan D. 2017 Emergence of transformation-tolerant representations of visual objects in rat lateral extrastriate cortex eLife 6:e22794. doi: 10.7554/eLife.22794
Author: Nicloe C. Rust. © 2017 eLife Sciences Publications Ltd. Republished via Creative Commons Attribution license. Image: Sasha the Okay Photographer CC-BY