Learning to see again: Biological constraints on cortical plasticity and the implications for sight restoration technologies

Abstract

2. Introduction

The field of sight restoration has made dramatic progress in the last five years. The Argus II device (epiretinal, Second Sight Medical Products Inc., Rizzo et al. 2014, da Cruz et al. 2016) as well as the Alpha-IMS system (subretinal, Retina Implant AG, Stingl et al. 2015) recently completed clinical trials and are now available for commercial sale in the US and Europe. Several other electronic vision implants have either started or are planning to start clinical trials in the near future. These include the IRIS II (epiretinal, Pixium Vision, Hornig et al. 2017), PRIMA (subretinal, Stanford University, Lorach et al. 2015), as well as devices by the Bionic Vision Australia consortium (suprachoroidal, Ayton et al. 2014, Saunders et al. 2014), Nidek Co. Ltd. (suprachoroidal, Fujikado et al. 2016) and Nano Retina (epiretinal). At the same time, sight recovery technologies based on optogenetics and gene therapies are also making considerable strides; over a dozen human gene therapy trials are underway (Petrs-Silva and Linden 2014), and optogenetic clinical trials will likely begin within the next two years (Busskamp et al. 2012). Within a decade, many individuals suffering from blindness are likely to be offered a wide range of options for sight restoration that depend on widely different technologies (Fine et al. 2015, Ghezzi 2015).

Clinical trials suggest that, for some patients, visual prostheses can provide visual information that is useful in daily life; enabling simple object localization, motion discrimination, and letter identification (Zrenner et al. 2011, Humayun et al. 2012, Stingl et al. 2013, Dagnelie et al. 2017). However, these reports also highlight the limitations of current devices. Clinical and psychophysical measurements of patients’ experiences have made it apparent that the vision provided by current devices differs substantially from normal sight. Only a handful of patients show performance close to the theoretical limits that would be expected if performance were limited by the spacing and density of the electrodes. Insights from psychophysical experiments and theoretical considerations (reviewed in Fine and Boynton 2015) suggest that this may be due to interactions between implant electronics and the underlying neurophysiology of the retina, discussed in more detail in Section 3 below. In electronic and optogenetic devices these include unselective stimulation of retinal cells (Figure 1B), and in the case of epiretinal devices, spatial blurring due to axonal comets (Panel C). In the case of optogenetic technologies this includes temporal blurring (Panel D).

Open in a separate window

Figure 1

Example simulations demonstrating potential perceptual distortions and information loss for different sight recovery technologies. All images are based on the central 12° region of a movie and a 1000×1000 array (A.) Original image. (B.) Simulation of sub-retinal electrical stimulation, based on simultaneously stimulating ON and OFF pathways with no axonal stimulation. (C.) Simulation of epiretinal electrical stimulation, based on simultaneously stimulating ON and OFF pathways with axonal stimulation resulting in visual ‘comets’. (D.) One frame in a movie showing simulation of optogenetic stimulation, based on a model of simulating ON pathways with an optogenetic protein with temporal dynamics based on ReaChR (Sengupta et al. 2016). Motion streaks are the result of the sluggish temporal dynamics of ReaChR. Modified figure based on Fine and Boynton (2015).

Second Sight Argus II patients report the experience of prosthetic vision as being like “looking at the night sky where you have millions of twinkly lights that almost look like chaos” (Pioneer Press 2015). One particular challenge is dealing with the combination of a limited field of view and an external camera, which is disconnected from the patient’s direction of gaze (Barry and Dagnelie 2016). As a result, patients are required to take in the world in a piece-wise fashion, using head motion and mental imagery to connect individually perceived phosphenes to form a whole. One patient reported: “I do see boundaries. I can create an image. I know what a car should look like. I know what a tree should look like. I know what houses should look like. I know what objects should look like, and I have those images as memories in my brain. With this device, you can start to create an image by going back and forth, checking the boundaries and borders. We don’t have natural saccade movement helping us. The camera is right above my nose in a fixed position. It doesn’t move. You can create an image, but it takes a lot of time and a lot of work.” (Discovery Eye Foundation 2014).

Consequently, it might be more appropriate to think of current retinal prostheses as ‘providing visual input’ rather than ‘restoring sight’. Nonetheless, it is the hope of the sight restoration community that the artificial visual signal elicited by electrical prostheses can help patients in daily life. One critical open question is – what role will cortical plasticity play in helping patients make use of this artificial visual input? Previous experience with cochlear implants suggests that cortical plasticity is capable of compensating for significant loss of information in the sensory input. Although the auditory signal from the earliest cochlear implants was too diminished to allow for speech perception (Eisen 2003), current implants can mediate surprisingly good speech recognition in most adult-deafened subjects (Shannon 2012). Patients successfully adapt to the impoverished and distorted auditory input over the course of hours to approximately a year.

Plasticity for retinal implants may be very different than for cochlear implants. Early stages of the visual hierarchy seem to be much less plastic than the auditory or somatosensory systems: the preponderance of evidence, reviewed below, suggests that large-scale cortical reorganization at the level of the lateral geniculate nucleus (LGN) or the primary visual cortex (V1) may be very limited. This lack of plasticity early in the visual processing stream has important implications for what expectations are reasonable when hoping that patients can learn to interpret the input from visual prosthetics.

The goal of the present review is to briefly summarize the vast basic science literature on perceptual learning and cortical plasticity with an emphasis on how this literature might relate to the field of visual prostheses. The objective is not to provide concrete recommendations for prosthetic development or rehabilitative training, but to provide insight into the fundamental principles which govern perceptual learning and cortical plasticity throughout the lifespan, so as to inform the choices of engineers and clinicians who are designing devices or training patients.

We begin with a description of the distortions and information loss likely to be experienced by visual prosthesis users. We then discuss cortical plasticity and perceptual learning and describe what is known, and what is unknown, about visual plasticity; both across the hierarchy of brain regions involved in visual processing, and across different stages of life. Because we assume that most visual prostheses will be implanted in late blind individuals suffering from binocular loss (at least in the foreseeable future), certain important fields in the cortical plasticity literature are neglected in this review. These include the effects of stroke or trauma within subcortical or cortical structures, monocular deprivation (for reviews see Levi et al. 2015, Kiorpes 2016), and the development of cross-modal plasticity (for review see, Bavelier and Neville 2002, Lewis and Fine 2011) since this tends to be relatively limited in late blind individuals (for discussion see Collignon et al. 2013). We close by discussing what is known about brain plasticity in sight restoration patients and discuss biological mechanisms that might eventually be harnessed to improve perceptual learning.

3. Distortions and information loss in prosthetic devices

As mentioned in the Introduction, interactions between implant electronics and the underlying neurophysiology of the retina will differ depending on the device. Nonselective stimulation of retinal cells is a concern for all electronic and optogenetic technologies. Electrode sizes of current prosthetic devices (typically on the order of 100 μm) inevitably lead to the indiscriminate stimulation of thousands of morphologically distinct retinal cells (including simultaneous activation of both ON and OFF cells, Figure 1B). This is in contrast to natural stimulation, which precisely activates a number of specialized, functionally complementary, parallel processing pathways in the retina (for a review see Nassi and Callaway 2009).

While using large electrodes exacerbates the issue, nonselective stimulation remains a concern for electronic technologies with extremely small electrodes and optogenetic technologies. In optogenetics, while it is possible to target a single class of cell (e.g. bipolar vs. ganglion cells) it is not currently possible to selectively target ON vs. OFF cells (see Fine et al. 2015, for a review). Thus, for both these technologies, the lack of differentiation in neuronal responses across different cell types results in a reduction in the amount of information available to cortex (Moreno-Bote et al. 2014). Prostheses implanted in the LGN and cortex will similarly suffer from an information-limiting lack of differentiation across the stimulated neural population, with the exact effects depending on which layers/regions of cortex are implanted.

There are of course many other causes of perceptual distortions and information loss. For epiretinal prostheses a significant concern are the visual “comets” (Figure 1C) that result from axonal stimulation. For optogenetic technologies, there is likely to be loss of temporal resolution, due to the sluggish temporal dynamics of optogenetic proteins (e.g. ReaChR). As is observable in Figure 1D, this loss of temporal resolution can result in significant spatial distortions for rapidly moving objects. Finally, the distortions and information losses for subcortical and cortical prostheses are likely to be complex, and depend significantly on which layers (Pezaris and Reid 2007) and/or cortical areas are stimulated. Direct electrical stimulation of the LGN (Pezaris and Reid 2007) or early visual cortex (V1/V2) elicits phosphenes (Tehovnik and Slocum 2007, Murphey et al. 2009). In contrast, electrical or magnetic stimulation of areas like the middle temporal visual area (V5/MT, Beckers and Homberg 1992), inferotemporal cortex, or specific regions within the fusiform gyrus (Afraz et al. 2006, Rangarajan et al. 2014) can disrupt motion and face processing, respectively, but does not elicit percepts (Murphey et al. 2009, Rangarajan et al. 2014). Representations in these higher-level areas may occur at a distributed network level (Levy et al. 2004). As a consequence focal stimulation does not create an interpretable percept, but can disrupt the network activity elicited by a natural visual stimulus.

Because each type of device is likely to suffer from unique types of information loss, quantitative measures of the quality of visual information provided by a device may become critically important as patients are given choices across multiple very different technologies. Similarly, being able to assess the ‘visual cost’ of a given distortion could play an important role in prosthetic design. There are already numerous metrics that can assess the perceptual quality of an image or movie based on models of the visual system (Haines and Chuang 1992, Wang et al. 2004, Laparra et al. 2016), but these assume no capacity for plasticity. Metrics describing the amount of ‘neurophysiologically available’ information (i.e. assuming perfect plasticity) in a device are therefore also important. Figure 2 describes one way in which loss of information (in this example resulting from axonal stimulation in an epiretinal prosthesis) might be quantified. A variety of devices with the number of electrodes ranging from two to 1500 were simulated. The field of view was held roughly constant at 10×20 degrees of visual angle and the electrode radius was set to 20% of the electrode-electrode spacing. Spatial distortions due to axonal stimulation was predicted using a model based on Nanduri et al. (2011). The resulting ‘predicted percepts’ (Panels A and B) for each device were used as input into a principal components analysis (PCA). If every electrode acted independently, the resulting visual output would be able to span a space of possible percepts with a dimensionality equal to the number of electrodes in the array. We can quantify the actual dimensionality of the space by the number of principal components needed to explain 95% of the variance – this can be thought of as the number of independently acting, ‘effective’ electrodes. Panels C and D (note the difference in y-axis scaling) show the number of effective electrodes as a function of the number of physical electrodes in the device both excluding (Panel C) and including (Panel D) axonal stimulation. Without axonal blurring, the number of principal components increases linearly with the number of electrodes with a slope of roughly 0.5: in an array of 1400 electrodes about 700 principal components are required to explain 95% of the variance. When axonal stimulation is included the number of effective electrodes drops by about an order of magnitude; with 1400 electrodes, only around 50 principal components are required to explain 95% of the variance.

Open in a separate window

Figure 2

Quantifying loss of information in epiretinal devices. (A.) The retinal location of a simulated Argus II array (6×10 electrodes) and the predicted percepts generated by stimulating single electrodes both without (left) and with (right) axonal stimulation. (B) The retinal location of a simulated high-resolution array (30×50 electrodes) with analogous percepts. (C, D.) Spatial distortions result in a reduced number of effective electrodes (i.e., the number of principal components needed to explain 95% of the variance) as a function of the number of physical electrodes in the array, excluding (C.) and including (D.) axonal stimulation.

4. Cortical plasticity and perceptual learning: two sides of the same coin

Cortical plasticity (also known as neuroplasticity) is an umbrella term that describes the ability of cortex to change its structure or function in response to experience. Cortical plasticity can be observed at multiple temporal scales (Horton et al. 2017), ranging from short-term (seconds to minutes) to long-term (days to many months). Cortical plasticity also occurs across a wide range of spatial scales, ranging from alterations in the tuning characteristics of individual neurons up to reorganization of entire neuronal circuits (“cortical remapping”, see Wandell and Smirnakis 2009, for a review).

The behavioral manifestation of cortical plasticity is perceptual learning. In a traditional perceptual learning paradigm, training on a specific task leads to a long-lasting improvement in behavioral performance (Karmarkar and Dan 2006, Deveau et al. 2014), such as a decrease in the minimal orientation difference that can be detected, or an increase in the speed for detecting target shapes embedded in distracters (for a review, see Sagi 2011). Similar to cortical plasticity, perceptual learning is a well-described feature of mammalian visual systems that occurs over multiple time scales – ranging from seconds (Mooney 1957) and minutes to years (Krupinski et al. 2013).

Although perceptual learning and cortical plasticity represent corresponding behavioral and physiological measurements of the same phenomenon, it has often proved difficult to understand the relationship between the two. One reason for this is that performance improvements for apparently quite similar tasks can be mediated by very different cortical substrates. For example, training in auditory (Recanzone et al. 1993) or tactile (Recanzone et al. 1992a, Recanzone et al. 1992b) frequency discrimination seems to result in “cortical recruitment” within primary sensory areas; that is, expansion of the amount of cortical territory/number of neurons representing the trained frequencies. In contrast, training on visual orientation discrimination tasks does not seem to substantially alter either the number of neurons tuned to the trained orientation or the orientation tuning of neurons whose orientation preference is close to the trained frequency (Crist et al. 2001, Ghose et al. 2002; though see Schoups et al. 2001 for evidence of subtle changes in tuning). Rather, visual perceptual learning is linked to alterations in extra-classical contextual responses (Crist et al. 2001, Li et al. 2004), possibly mediated by altered response properties within V4 neurons with orientation tuning relevant to the task (Yang and Maunsell 2004).

5. Perceptual learning

Perceptual learning is likely to be very different in prosthesis users than in sighted subjects. In subjects with normal vision it can be assumed that the organization of the visual pathways is already close to optimal, especially in the fovea (Westheimer 2001). As a result, task performance for naturalistic tasks such as object recognition is generally at ceiling, making it necessary to titrate difficulty by reducing stimulus difference (e.g. orientation difference), contrast or stimulus duration, or by adding external noise. However, even if the task is made more difficult, learning is likely still limited by the fact that the visual system is already optimized for these stimuli - indeed this is probably why familiar objects show so little learning (see task 6, Figure 3, below).

Open in a separate window

Figure 3

Performance improvement as a function of practice over time for 16 different perceptual tasks (ordered according to the estimated learning slope). For all studies, performance on each session was converted into d′ (Green and Swets 1966). The learning index, L, measures improvements in performance with practice, Ls=ds′/d1′, where s is the session number. A learning index remaining near 1 implies that observers showed no improvement in performance with practice. Reprinted with permission (Fine and Jacobs 2002).

6. Cortical plasticity within the sensitive period

Although implantation of prosthetic implants will be restricted to adults for the foreseeable future, the literature on plasticity in the sensitive period is nonetheless highly relevant. Important aspects of visual cortical architecture, such as retinotopic organization, are driven by molecular signaling and are robust to visual perturbations even during early development. It is important to recognize that aspects of the visual architecture that are minimally dependent on experience during early development are unlikely to be modifiable via training in adulthood.

7. Adult plasticity

One important question is whether adult plasticity in visual cortex differs from juvenile plasticity quantitatively, or whether there are also qualitative differences (Karmarkar and Dan 2006). Specifically, as described below it is not yet clear whether or not adult visual cortex can generate novel, functionally appropriate synaptic connections.

8. Effects of training in visual prosthesis patients

Visual prosthesis and optogenetic technologies offer a unique platform to study cortical plasticity in adulthood. We have seen that local connections in adult cortex might be at least partially modifiable, but is this plasticity sufficient to make visual prostheses useful?

Previous experience with cochlear implants would suggest optimism. Adult-implanted cochlear implant users initially report extremely unnatural and incomprehensible perceptual experiences. Over remarkably short time periods, not only do subjects become better at important tasks such as speech recognition, but their perceptual experiences become more qualitatively ‘natural’. As described by one patient: “It sounded like popcorn… every time someone said something it popped… as soon as I opened my mouth it popped… Now it sounds a little like Alice in Wonderland when she stands there in the tunnel talking. It is like more of a clanging sound now and it is very clear but still a bit lighter… more treble and like talking in a can.” (Hallberg and Ringdahl 2004).

The abilities of visual prosthetic patients also improve with training. However there is little evidence that this is due to distortions becoming less perceptually apparent. Instead most improvements seem to be relatively task specific, and are the result of patients becoming better at interpreting distorted input.

One study did find reductions in visual field loss as measured with Goldmann perimetry after prosthetic implantation (Rizzo et al. 2015) for both implanted and non-implanted eyes. This may have been due to ‘reactivation’ of the visual system (though other phenomena such as Charles Bonnet syndrome (Cogan 1973) and the photophobia often observed after cataract surgery suggest that long-term deprivation leads to upregulation of deprived visual cortex). Alternatively, these results may reflect changes in behavioral response criteria (Shapiro and Johnson 1990). Another study found decreases in detection thresholds as a function of time since surgery (Castaldi et al. 2016). However subject performance on other relatively simple perceptual tasks, such as contrast sensitivity (Castaldi et al. 2016) and motion discrimination (Dorn et al. 2013, Castaldi et al. 2016) barely improved, even after extensive training over several months.

The largest perceptual improvements found in prosthetic users have generally been reported in tasks such as moving independently in space, locating a large bright square on a screen, and identifying large-print letters (~40°) at above-chance levels (Chader et al. 2009, Zrenner et al. 2011, Humayun et al. 2012, da Cruz et al. 2013, Dorn et al. 2013, Stingl et al. 2013, Rizzo et al. 2015). After practice, single-letter recognition seems to take somewhere between a few seconds to three and a half minutes, enabling the best performing patients to read short words (da Cruz et al. 2013).

One possibility is that the perceptual ‘experience’ of prosthetic users will gradually improve over time, as seems to occur for cochlear implants, due to either a reduction in perceptual distortions or an increase in the perceptual discriminability of stimuli. Alternatively, patients may learn how to use distorted information to perform specific tasks. Dissociating these two kinds of learning, and understanding how they contribute to prosthetic vision, will require careful thought in future longitudinal studies of sight restoration patients.

Importantly, many assessments of the effects of training on visual prosthetic performance have focused on “closed” tasks under somewhat unrealistic contexts. This may lead to overly optimistic assessments of performance and the effects of training. An example of a strongly closed task is “is that an ‘A’ or a ‘B‘“. An example of a less closed task is “what letter is this?” An open task is “What is on the screen?” One difficulty with closed tasks, particularly those that mimic open tasks, is that they can give a highly misleading impression of performance. As in Figure 4, when subjects are told that the four items on the table include a plate, a cup, a napkin and a fork it is easy to see how subjects can successfully ‘find the fork’. Real dinner tables tend to be more cluttered and disorganized. An individual who can successfully find a fork among a preselected set of four items may not be able to find a candlestick on a cluttered dinner table. A second issue with closed tasks is that they are likely to result in perceptual learning that is highly specific to the particular context or task used. The literature on perceptual learning (see Section 5.2 above) indicates very strongly that learning will almost always occur at the most specific level possible given the training task. Thus, care should always be taken when designing training protocols to ensure that improvements are likely to generalize to performance outside the laboratory or clinic.

Open in a separate window

Figure 4

Two examples of a dining table under conditions of simulated prosthetic vision. Under ‘laboratory’ conditions the plate, cup, napkin and fork are easily differentiable. Real-world conditions for Dr. Fine’s dining table include multiple additional and unexpected objects. Finding the fork is challenging, even for an individual with normal vision.

9. Reawakening the critical period

One exciting avenue for improving prosthetic implant outcomes may come from harnessing new methods for ‘reawakening the critical period’ (Bavelier et al. 2010, Werker and Hensch 2015) via pharmacological intervention.

Recent advances in neurobiology have provided an increasingly detailed picture of the cascade of events that control critical period plasticity. During early development, the synaptic neurochemical balance is shifted towards excitation, and synaptic connectivity is fluid. Synaptic rewiring includes physical pruning and homeostatic regrowth of synapses (controlled by the mediators tPA, TNFα, protein synthesis). Over time, responses to sensory input cause the promotion of inhibitory parvalbumin cell maturation and their increased GABA function through molecular triggers, such as orthodenticle homeobox 2 (Otx2), and brain-derived neurotrophic factor (BDNF, Huang et al. 1999). This shifts the excitation/inhibition balance towards a mature state of greater inhibition. The closure of the critical period is actively enforced by structural consolidation, including formation of a perineuronal extracellular matrix, and epigenetic brakes on plasticity (e.g., histone deacetylation) that silence the gene programs necessary for synaptic rewiring (Hensch 2005b, Takesian and Hensch 2013, Werker and Hensch 2015). This has led to a significant amount of research with the goal of ‘reawakening’ cortical plasticity via biochemical intervention that targets various aspects of this cascade.

In both animals and humans there is evidence that serotonin uptake inhibitors shift the excitatory-inhibitory balance towards a state of greater excitation, and may enhance plasticity. In adult animal models fluoxetine results in reduced GABAergic inhibition and increased BDNF expression, which results in reopening of the critical period (Maya-Vetencourt et al. 2008, Bachatene et al. 2013), but a study in humans did not find that fluoxetine enhanced perceptual learning (Lagas et al. 2016). In humans the FDA-approved cholinesterase inhibitor donepezil seems to enhance perceptual learning, perhaps by shifting the excitation/inhibition balance towards the ‘immature’ state of greater excitation (Rokem et al. 2010, Rokem and Silver 2013, Chamoun et al. 2017). While shifting the neurochemical balance towards greater excitation may be necessary to ‘reopen’ the critical period, there is growing evidence that it is not sufficient (Hensch 2005a, Wong 2012, Takesian and Hensch 2013). Epigenetic brakes and structural limitations must also be overcome.

As far as epigenetic limitations are concerned, in humans, valproate has been successfully used to enhance auditory cortex plasticity for training in pitch perception in adult humans (Gervain et al. 2013), although these findings have yet to be replicated and might not generalize to the visual system. The use of valproate to target visual cortical plasticity is complicated by its adverse side effects (Sandberg et al. 2011, Sisk 2012, Bhalla et al. 2013) and a literature leaving it unclear whether its effect on retinal function in patients with retinal dystrophies is positive (Clemson et al. 2011, Iraha et al. 2016) or negative (Sisk 2012, Bhalla et al. 2013).

As far as structural limitations are concerned, homeobox protein Otx2 has the potential to serve a critical role in reawakening the critical period. Disruption of Otx2 reduces parvalbumin (thereby influencing excitatory/inhibitory balance) and reduces perineuronal net expression (thereby reducing structural limitations on plasticity). In animal models disruption of Ox2 has proved sufficient to reopen plasticity in adult mice (Beurdeley et al. 2012, Spatazza et al. 2013). However, pharmacological enhancement of OTX2 raises serious safety concerns given that over-expression of OXT2 is associated with a variety of cancers (Di et al. 2005, Nagel et al. 2015).

In conclusion, as we develop a fuller understanding of the triggers, modulators and brakes that control the sensitive period, it appears increasingly likely that ‘reawakening plasticity’ in cortex is likely to require simultaneously targeting multiple parts of the cascade: it is not adequate to reduce structural brakes on synaptic rewiring if the excitatory-inhibitory balance is unsuitable for new connection patterns to be established, and vice versa. However, as described above, there are numerous FDA-approved drugs that seem likely to modestly enhance plasticity. Any reader over the age of forty is well aware that adult plasticity continues to decline with age. Given that most prosthetic implant recipients are likely to be elderly, even modest improvements in the capacity for rehabilitative plasticity might greatly enhance their ability to make functional use of their devices.

10. Conclusions

Visual prostheses and optogenetic methods for restoring retinal function offer a unique platform for studying cortical plasticity in adults that is likely to increase in translational importance as the field progresses. The literature described in this review strongly suggests that there exists a “tolerance envelope” of adequate sensory information which any successful prosthetic device must match. If distortions or the amount of information loss fall within that envelope, it seems plausible that cortical plasticity can successfully compensate. However, it must be recognized that cortical plasticity is finite and cannot compensate for distortions or information loss that falls outside that envelope.

Thus, it is critical to understand which visual distortions can be compensated for via cortical plasticity, which cannot, and what role training or pharmacological intervention can play. Engineers need to focus their energies on developing technologies that fall within the sensory ‘tolerance envelope’. This will require a shift from thinking of the brain as having ‘pluripotent plasticity’ towards a more nuanced understanding of which aspects of visual perception are adaptable, and under which conditions. It is not clear whether spatial distortions can be remapped within early visual areas, but higher level areas may be surprisingly competent at decoding a distorted world. However, learning to decode a distorted world must be done in the context of generalizable learning; this will require training across a wide range of stimulus sets or navigational environments, while maintaining high performance levels. Understanding these principles of effective and generalizable training will be critical for developing rehabilitation strategies that can efficiently help patients make best use of their implants. In the context of prosthetic vision, this may require increased use of virtual reality environments.

Finally low-level visual plasticity seems to be limited compared to the plasticity observed in other cognitive and sensory domains. This provides a significant technical challenge for engineers, and any way of enhancing low-level plasticity is likely to be extremely beneficial for rehabilitative training. In the short-term, gamification may provide significant benefits. In the longer term pharmacological intervention provides an exciting potential opportunity.

What is most exciting about this relationship between neural prosthesis development and basic neuroscience is that the flow of information will be reciprocal. The next decade is likely to yield fundamental insights within an important, but relatively poorly understood field of research – human adult plasticity. Perhaps the capacity that most differentiates humans from other animals is our capacity to continuously learn new perceptual and cognitive skills throughout our lives. Visual prosthesis patients are likely to provide a unique opportunity to gain insights into the fundamental principles that underlie these abilities.

Acknowledgments

References