Rebound Spiking as a Neural Mechanism for Surface Filling-in Hans Supèr1,2,3 and August Romeo1 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/23/2/491/1774688/jocn.2010.21512.pdf by guest on 18 May 2021 Abstract ■ Perceptual filling-in is the phenomenon where visual informa- hibition produces rebound or after-discharge spiking in neurons tion is perceived although information is not physically present. that otherwise do not receive sensory information. The behavior For instance, the blind spot, which corresponds to the retinal loca- of rebound spiking mimics the immediate surface filling-in illusion tion where there are no photoreceptor cells to capture the visual observed at the blind spot and also reproduces the filling-in of an signals, is filled-in by the surrounding visual signals. The neural empty object after a background flash, like in the color dove illu- mechanism for such immediate filling-in of surfaces is unclear. sion. In conclusion, we propose rebound spiking as a possible By means of computational modeling, we show that surround in- neural mechanism for surface filling-in. ■ INTRODUCTION filling-in processes but they are probably too slow to explain The blind spot is the region in the visual field that corre- the rather immediate surface filling-in at the blind spot (see sponds to the optic disk where the optic nerve leaves the Komatsu, 2006). The other hypothesis, the cognitive or sym- retina. At this location, there are no light-detecting pho- bolic filling-in theory, postulates that blind regions are ig- toreceptor cells to capture the visual events, and conse- nored and object representation is realized at high cortical quently this part of the visual field is not perceived. Yet we level on the basis of contrast information from lower areas do not see a hole in our visual scene when we look with (Pessoa et al., 1998). Feedback projections from these higher one eye because the location of the blind spot is filled-in areas have large axonal termination fields in the early visual by the surrounding visual information (see Figure 1A). areas and may so provide sensory information to neurons in This is shown by neurophysiological reports that describe the lower areas located at the blind spot region. However, neural responses related to filling-in at the blind spot in the it has been shown that feedback has a role in modulating early visual cortex (Matsumoto & Komatsu, 2005; Komatsu, stimulus-evoked responses and does not activate otherwise Kinoshita, & Murakami, 2000, 2002; Fiorani, Rosa, Gattas, & silent neurons (Ekstrom, Roelfsema, Arsenault, Bonmassar, Rocha-Miranda, 1992), which are consistent with neural de- & Vanduffel, 2008). This indicates that cortical neurons at scriptions of other forms of surface filling-in early visual cor- the blind spot region need to be activated, presumably by tex (Huang & Paradiso, 2008; MacEvoy, Kim, & Paradiso, feed-forward connections. 1998; De Weerd, Gattass, Desimone, & Ungerleider, 1995). How can retinal signals be effective in activating cells in The neural mechanisms for filling-in of are still a matter early cortical areas that do not receive feed-forward exci- of debate. Two different theories have been put forward to tatory projections? The excitatory retinal information is explain the filling-in completion phenomenon. One theory accompanied by inhibitory signals. Besides the global influ- postulates that spreading of neural activity in early visual ence, inhibition is robust, fast, and prominent in retina, areas is the basis for filling-in of visual information (Pessoa, LGN, and visual cortex (Alitto & Usrey, 2008; Solomon, Thompson, & Noe, 1998; Ramachandran & Gregory, 1991). Lee, & Sun, 2006; Blitz & Regehr, 2005). It is well known This theory is based on the assumption that cells at contrast that strong inhibition may cause rebound excitation at the borders spread their activity to surrounding cells. In such a end of the hyperpolarized period. Rebound or paradoxical case, filling-in is accomplished by the dense network of hori- excitation is a biophysical feature of neurons in which, fol- zontal connections that exist in the visual cortex. Horizontal lowing a period of strong hyperpolarization below the rest- connections have slow conduction velocities (0.1–0.2 m/sec; ing membrane potential, the membrane potential briefly Angelucci & Bressloff, 2006) and may explain slow surface rebounds to a more depolarized level resulting in firing spikes. Rebound spiking is thus triggered by inhibition and not by direct sensory activation. After-discharges may 1University of Barcelona, 2Institute for Brain, Cognition, and also be evoked by rebounds through inhibitory networks Behavior, 3Catalan Institution for Research & Advanced Stud- (Macknik & Martinez-Conde, 2004; Macknik & Livingstone, ies (ICREA) 1998). Here we prefer to use the term rebound spikes © 2010 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 23:2, pp. 491–501 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2010.21512 by guest on 01 October 2021 Figure 1. Perceptual filling-in and model architecture. (A) A surface (input) is perceived although at some retinal parts there are no sensory receptors. The blind spot region is filled-in by the surrounding visual information so that cortical neurons whose receptive field locations correspond to the blind spot region respond to the surface stimulus. (B) The model consists Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/23/2/491/1774688/jocn.2010.21512.pdf by guest on 18 May 2021 of two neural layers, which are unidirectional connected. All layers receive point-to-point (retinotopic) excitatory input (black arrow). The first neural layer receives sensory input. Neurons in the second layer also receive inhibitory input from all preceding neurons (gray shading). instead of after-discharges (Adrian & Matthews, 1927) be- METHODS cause in our study, neurons become active (rebound) after Model Architecture and Inputs the end of suppression rather than continue firing spikes after removal of the receptive field stimulus. The model is composed of two layers, each containing two In the visual system, rebound spiking is observed in arrays of 64 × 64 units of neurons of the Izhikevich type the retina (Margolis & Detwiler, 2007; Mitra & Miller, (Izhikevich, 2003; Figure 1B). Each layer corresponds to a 2007a, 2007b), LGN (Bright, Aller, & Brickley, 2007; Zhu & visual region. We consider neurons in the first layer as ret- Lo, 1996; Mastronarde, 1987), and visual cortex (Moliadze, inal ganglion cells, which transform continuous or graded Zhao, Eysel, & Funke, 2003). In the retina amacrine cells input into spike activity. In this layer, the region of the optic may inhibit ganglion cells over a large region causing re- nerve corresponding to the blind spot was modeled by a bound spiking in these cells (see Mitra & Miller, 2007a), center region (16 × 16) void of neuronal cells. For the color and in the LGN, reticular cells may evoke rebound burst dove illusion, the center part represented the location of in relay cells (Destexhe & Sejnowski, 2002). Hence, strong, the empty object and contained neurons like in the normal global inhibition, and rebound spiking are prominent in visual field. The second layer may correspond to the LGN or early visual structures. Therefore, although it has been ar- V1. Neurons in the first layer receive surface input, which is gued that rebound activity may not represent visual infor- an array of 64 × 64 pixels. The pixel values of the input array mation (Buzsaki, 2006), we consider the possibility that are 1 and correspond to the preference of a single visual fea- rebound activity induced by widespread suppression can ture, like direction of motion or color. For the color dove be an alternative explanation for surface filling-in. illusion, the pixel values were set to 0 for the background To test this idea, we used computer simulations of a and object region. neural network model composed of biologically plausible spiking neurons (Izhikevich, 2003) that permit to investi- Feed-forward Connections gate such dynamic network behavior. Our results show that inhibition produced rebound spiking in neurons cor- The excitatory feed-forward projections from the input responding to the blind spot after surface stimulation. Sur- layer to the first neural layer and from the first to the sec- rounding cells also responded to the surface stimulus, ond neural layer are retinotopic (point-to-point connec- although they received the same inhibitory input as the tions), where pixel/neuron Nij in the one layer solely cells at the blind spot. The strength and onset latency of connects to neuron Nij in the next layer (Figure 1B). Thus, the rebound responses were similar to the ones of the stim- the excitatory part of a neuronʼs receptive field has size ulus evoked response, which agrees with complete and im- one. Neurons in the first neural layer do not receive inhib- mediate perceptual filling-in of the blind spot (Komatsu, itory signals from the surface stimulus input. Neurons in 2006; Ramachandran & Gregory, 1991). Finally, our model the second layer receive inhibition from all neurons lo- can explain the immediate filling-in of an empty object at cated in the preceding layer. Thus, inhibition is global. In- the normal visual field location at the end of a background hibition is achieved by assigning negative weights to the color flash, as happens in the color dove illusion. So we pro- connections. Neither intralaminar connections, that is, hori- pose rebound spiking as an alternative neural mechanism zontal connections between neurons within a layer, nor for some types of surface filling-in. feedback connections, that is, connections from the second 492 Journal of Cognitive Neuroscience Volume 23, Number 2 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.2010.21512 by guest on 01 October 2021 neural layer to the first neural layer, are included in the net- fact applied to vij, uij,and∀ij.