Amongst the biggest challenges of prosthetic vision is the puzzle of the neural code for perception. The complexity of the neural code suggests that prosthetic devices should rely on intact neural circuitry whenever possible in order to take advantage of any intact sensory processing available (Dagnelie and Schuchard, 2007). Thus, reducing the complexity of neural coding necessary could potentially be achieved by implanting the prosthetic device at the earliest point along the visual pathway that retains functional integrity. Following to this premise, the retina would represent the earliest site of potential neuronal interface.

Retinitis pigmentosa (RP) and age related macular degeneration (AMD) are two retinal disorders that contribute greatly to the incidence of inherited blindness and blindness in the elderly respectively (Bunker et al., 1984; Klein et al., 1997). Profound vision loss results largely due to the progressive degeneration of the light-capturing component of the outer-segment of the retina, that is, the photoreceptor cells. However, the remaining retinal elements within the inner retinal layers (e.g. the bipolar and ganglion cells that converge to form the optic nerve) appear to survive in large numbers. Furthermore, these elements remain responsive to electrical simulation even in highly advanced stages of the disease (Humayun et al., 1996). In essence, a retinal based visual prosthesis would replace the function of the degenerated photoreceptor cells by stimulating the surviving retinal neuronal machinery. A set of pivotal human experiments demonstrated that electrical stimulation of the retina of RP patients (Humayun et al., 1996; Rizzo et al., 2003b) as well as one patient with AMD (Humayun et al., 1999) led to the generation of phosphenes despite the fact that patients were profoundly blind for many years. Experiments lasted minutes to hours while patients remained awake in order to describe their visual experiences. Following electrical stimulation, patients reported visual patterned perceptions that were initially relatively crude. However, the gross geometric structure of the phosphene patterns could be altered in a controlled fashion by varying the position and number of the stimulating electrodes and the strength or duration of the delivered current (Humayun et al., 1996; Rizzo et al., 2003a, b). This demonstration of proof-of-principle has led many groups worldwide to pursue development of a variety of retinal-based designs and approaches. Currently, the retinal-based approach is arguably receiving the most attention as evidenced by size and number of on-going human clinical trials.

Two retinal-based approaches are being pursued that are largely differentiated by their location of implantation with respect to the retina. In the sub-retinal approach, the implant is placed in the region of degenerated photoreceptors by creating a pocket between the sensory retina and retinal pigment epithelium (RPE) layer. In the epi-retinal approach, the implant device is attached to the inner surface of the retina, close to the ganglion cell side (figure 1A).

The sub-retinal visual neuroprosthesis design is currently being pursued by the Boston Retinal Implant Project (a large joint collaborative effort that includes the Massachusetts Eye and Ear Infirmary and Harvard Medical School, the Massachusetts Institute of Technology, the Boston Veterans Affairs Healthcare System, and other partnering institutions) (see figure 2) (Shire et al., 2009). By virtue of being placed in juxtaposition to the nearest layer of surviving neurons (i.e. bipolar cells) the sub-retinal approach affords greater inherent mechanical stability. This is due to the fact that the ultra-thin electrode array is effectively “sandwiched” between the inner-segment of retina and the RPE layer. Furthermore, this approach has the theoretical advantage of not only being closer to surviving neuronal elements (thus potentially requiring lower amounts of electrical stimulating current) but also exploiting retinal signal pre-processing inherent to the bipolar cell layer. The placement of a sub-retinal device does require elaborate and complex surgical methods. For the Boston Retinal Implant device, this includes inserting an ultra-thin flexible microelectrode array through an incision made on the outside scleral wall of the ocular globe. This surgical approach is used so that the device resides within the sub-retinal space created (referred to as the “ab externo approach” as opposed to “ab interno”; where ones passes through the vitreous humor of the eye and inserts the device through an incision made directly in the retina, see (Javaheri et al., 2006)). Another feature of this configuration is that it leaves the bulk of the electronic hardware outside of the eye thus avoiding complications related to heat generation and corrosion and facilitates the exchange of electronic components as needed. For its operation, a miniature camera mounted on a pair of eyeglasses is used for image capture. These images are then analyzed by an externally worn portable microprocessor used to convert the image data into an electronic signal. The appropriate signal pulses (delivering data and power) are transferred to the implant wirelessly via radio frequency (RF) coils. The resulting signal is transmitted to the subretinal microelectrode array driving the surviving retinal neural elements (i.e. bipolar and ganglion cells) with appropriate patterned electrical stimulation. It is here that the signal processing begins and is further integrated as it passes down the optic nerve on to the visual cortex for final perception of the visual image. All electronic parts are hermetically sealed in a titanium case connected to an external flex circuit and the microelectronic array (Kelly et al., 2009). To date, the group has succeeded in developing a wireless retinal prosthesis prototype as the first step towards a human subretinal prosthesis implant. Initial studies in animal models have been successful in implanting active versions of the device and refining surgical techniques and mechanical design (Kelly et al., 2009). Human clinical trials are now being planned.

Variations of the sub-retinal implant design have also been pursued by several large consortia efforts. The Artificial Silicone Retina (ASR) developed by Optobionics Corporation contains approximately 5,000 micro-photodiodes, each containing its own stimulating electrode (Chow et al., 2004). When implanted under the retina, photocurrents generated by absorbed light stimulate adjacent retinal neurons in a multi-site fashion. In a phase 1 trial of safety and efficacy carried out in six patients with profound vision loss from RP (followed from 6 to 18 months after implantation), patients reported an improvement in visual function after implantation. These reports were evidenced by an increase in visual field size and the ability to name more letters using a standardized visual acuity chart (Chow et al., 2004). While the relatively simple design of this device was intuitively appealing (note that no camera and subsequent image processing is required with this device), the apparent improvement in vision was not attributed to true prosthetic vision per se, but rather to a potential neurotrophic (or “cell rescue”) effect related to micro-electric currents generated by the device (Pardue et al., 2005a; Pardue et al., 2005b). With this limitation in mind, a multilayered subretinal chip device incorporating signal amplification is now being pursued by a German consortium (Retina Implant AG). This device has recently been implanted in profoundly blind RP patients and recent results have been encouraging. Early human clinical trial data suggests that stable visual percepts can be obtained and implanted patients profoundly blind with RP have been able to identify objects and letters (Besch et al., 2008; Zrenner, 2002).

As a contrasting design approach, the epi-retinal strategy entails placing an electrode array along the inner surface of the retina to stimulate the underlying ganglion cells. This procedure employs more typical vitreo-retinal surgery techniques so as to affix the microelectrode array on to the retinal surface (e.g. using a retinal tacks). The Artificial Retina Project has been actively pursued by a collaborative effort between the Doheny Eye Institute (University of Southern California) and Second Sight Medical Products. Like the Boston Retinal Implant design, this device incorporates a digital camera mounted on a pair of eyeglass capturing an image that in turn is converted into an electrical signal that is delivered to the retina (Humayun et al., 2003). Initial testing with a 16 electrode device (Argus I) in human volunteers with advanced RP has been successful. A large-scale multi-centered phase II FDA-sponsored clinical trial is currently underway to evaluate a second generation implant (Argus II; 60 electrodes) in the largest cohort of visual prosthesis recipients to date. Results suggest that patients chronically implanted with this device can detect phosphenes at individual electrodes, discriminate crude shapes upon multiple electrode stimulation, and recognize simple stimuli presented via a head-mounted camera (Humayun et al., 2009; Weiland et al., 2004). Very recently, the group reported that implanted subjects showed a significant improvement in accuracy in a spatial visual-motor target localization task comparing performance in patients implanted with their second generation device. Subjects were instructed to locate and touch a high contrast square target presented on a monitor. Nearly all subjects (26/27) showed a significant improvement in accuracy (Ahuja et al., 2010). This is consistent with the observation that implanted subjects were able to develop appropriate head-scanning techniques and good “camera-hand” coordination in using their visual prosthetic device (Ahuja et al., 2010).

Other groups are also pursuing the epi-retinal approach including a variety of German based consortiums. While still in earlier stages of development, early results have also been encouraging (Eckmiller et al., 2005; Gerding, 2007; Gerding et al., 2007)

Other notable downstream approaches have been developed. A Belgian consortium has developed a prosthesis designed to stimulate the optic nerve using a four-electrode cuff electrode design and driven by stimuli captured by an external camera (figure 1B). Two patients have been chronically implanted to date. Reports from one blind volunteer demonstrated that electrical stimulation evoked the perception of localized, and often colored, phosphenes throughout the visual field (Veraart et al., 2003). After four months of psychophysical testing, the patient could recognize and distinguish orientations of lines, some shapes and even certain letters (Brelen et al., 2005; Veraart et al., 2003).

For example, electrode geometry poses inherent limitations that must be carefully considered. This is particularly true with regards to how closely electrodes can be placed next to each other thus impacting the theoretical resolution the visual prosthesis can provide. Furthermore, electrode geometry is intimately related to the amount of current that can be delivered safely to the target neuronal tissue (i.e. charge/density limits). As these neuroprosthetic devices are meant to be implanted and used for very long-periods of time, the effect of prolonged and focal electrical stimulation delivered to delicate (and even potentially further degenerating) neuronal tissue remains unknown. In this direction, new electrode designs, materials and coatings are being actively pursued in order to improve and expand safety profiles. One intriguing possibility is the development of pillar electrode arrays. Implanting this electrode array design has shown that penetrating pillars are able to attract neuronal elements (e.g. ganglion cells of the retina). The closer electrode-neuron interface may allow for lower currents to be used and thus safer injection of current for prolonged electrical stimulation (Butterwick et al., 2009).

There also exists the issue of how a captured image is co-registered with the natural movement of the eye. Inappropriate compensatory eye movements may lead to perceptual mismatch, causing the patient wearing the implant to mis-localize objects in the external world. This potential confound is particularly true of implant designs that incorporate the use of an external mounted camera. To solve this issue, sophisticated eye-tracking mechanisms have been proposed and designed to generate appropriate shifts in the image (e.g. (Palanker et al., 2005)). However, these solutions await further development. Interestingly, recent work with visual simulations suggests that following training, implanted patients may learn to carry out appropriate compensatory head and camera movements to generate more stable percepts (Chen et al., 2009; Srivastava et al., 2009).

A review of human testing reveals that implanted recipients experience difficulties in fully understanding the visual information provided by these visual prosthetic devices. In the initial studies, the reported patterns of visual percepts often did not correspond to what was predicted based on the patterns of electrical stimulation delivered (e.g. (Rizzo et al., 2003a, b)). This key observation suggests that our intuitive sense as to how to generate patterned visual percepts (i.e. the “scoreboard approach”) may not prove to be an adequate strategy (Fernandez et al., 2005). This might be related to the fact that stimulation is carried out on neuronal tissue that is severely degenerated and, therefore, physiologically compromised. Certainly, the quality of visual percepts is likely to improve as remaining technical challenges continue to be solved. However, there may arrive a point when engineering and surgical issues will no longer represent the greatest impediment to future progress. Rather, the greatest barrier will likely lie in our ignorance of how to introduce visual information that is meaningful to the visually deprived brain. It is a misconception that simple perceptions generated from patterned light are sufficient to generate meaningful vision. Furthermore, increasing the resolution of images (for example, by increasing the number of stimulating electrodes) with the goal of generating more complex perceptions would initially be perceptually meaningless rather than helpful.

Several studies have highlighted that following the loss of vision, the brain undergoes profound neuroplastic transformation and that the occipital visual cortex is a major site of these changes (Bavelier and Neville, 2002; Fine et al., 2003; Merabet et al., 2005; Pascual-Leone et al., 1999). The extent and magnitude of these neuroplastic changes is likely to be influenced by such factors as the cause, onset and duration of blindness. The plasticity of the visual system may allow for a considerable degree of adaptation. However, understanding the precise constraints of these neuroplastic processes will be crucial and have clear implications for rehabilitative training and progress in device development (Fernandez et al., 2005).