1. Advances in Engineering of Prostheses

Advances in Engineering of Prostheses

Material Science

Retinal prostheses employ two categories of electrode materials: inorganics (metals, silicones, carbon-based materials) and organics (polyimide, polydimethylsiloxane/PDMS)^[1]. Smaller-sized electrodes have been shown in the literature to optimize electrical signaling by matching the dimensionality of neural targets. However, further investigation is necessary to address the inherent physical limitations of current electrodes^[2].

Among conducting metals, iridium, gold, titanium, tin, and platinum are the most commonly used in metal microwires for retinal prosthetics. These metals exhibit chemical inertness, desirable electrical properties, and proven biocompatibility in neural interfaces^[1]. Planar microelectrodes are frequently employed due to their relatively high spatial resolution capacity. However, metal microwires have inherent limitations, including high impedance, lower charge storage capacities, rigidity complicating fixation, and increased risk of adverse surgical and long-term physiological complications associated with transcutaneous wire connections^[1].

To address these limitations, microelectromechanical system (MEMS)-based silicon materials offer potential solutions. These techniques utilize microfabrication methods to shape rigid planar structures into context-specific layouts, enabling higher density recording sites^[3]. However, a major drawback of MEMS-based silicon materials is their limited conformity with native tissue due to rigidity. Even after reducing electrode thickness for improved bending rigidity, the smallest MEMS-based electrodes remain larger than optimal for conformability^[1]. Hence, alternative materials such as conductive polymers, carbon-based materials, and nanomaterials show promise.

Conductive and semi-conductive polymers are organic materials with intrinsic electrical and optical characteristics that have demonstrated efficacy in electrode-tissue interfaces. They outperform purely metal-based materials due to their chemical stability, enhanced biocompatibility, and high conductivity. Examples include poly(3,4-ethylene dioxythiophene) (PEDOT), polyaniline, polythiophene, and polypyrrole (PPy), with PPy being widely used in brain-machine interfaces. PEDOT treated with polystyrene sulfonate has also shown superior performance in the mentioned metrics. However, weak interactions between the conductive polymer coating and the electrode pose a risk of delamination, which can disrupt signal registration^[4][5][6].

To address these challenges, Ouyang and colleagues (2017) proposed surface functionalization as a solution. They modified the surface properties of the electrode by functionalizing PEDOT with (2,3-dihydrothieno [3,4-b] [1,4] dioxin-2-yl) methenamine (EDOT-NH2), a methylamine-functionalized EDOT derivative, using electrografting techniques. Electrografting involves forming strong covalent bonds between organic materials and solid substrates. Their findings demonstrated improved adhesion with the P(EDOT-NH2) anchoring layer and enhanced electrode durability^[7].

PEDOT has been shown to exhibit significantly higher efficiency in charge transfers compared to traditional metal electrodes, as demonstrated by Green et al. (2010)^[8]. In their in vitro study, they coated platinum microelectrode arrays with PEDOT and conducted biphasic stimulation protocols. The findings revealed a 15-fold increase in the charge injection limit compared to platinum, along with a substantial reduction in potential excursions at the platinum electrode. When implanted into the suprachoroidal space of a cat retina, PEDOT also exhibited reduced potential excursions^[8].

Carbon-based materials have gained attention in the literature due to their favorable properties, including low modulus and electrical impedance^[1]. Carbon nanotubes (CNTs) offer large effective surface areas in electrode-tissue interfaces, resulting in enhanced charge transfer capacity and low interfacial impedance^[9]. Eleftheriou et al. (2017) investigated the structural and functional integration of CNTs in retinal prosthetics and observed a decrease in stimulation thresholds and increased cellular recruitment over three days. These findings indicate improved CNT-electrode-retinal ganglion cell coupling while mitigating negative glial responses^[10]. Such discoveries suggest the potential therapeutic benefits of employing CNTs in retinal prosthetics.

However, concerns about the biotoxicity of carbon-based materials, including CNTs and graphenes, have been raised^[1]. Nevertheless, similar to conductive polymers, these materials can be functionalized to enhance biocompatibility and optimize electrode-tissue interfaces^[11][12]. Graphene-based materials, in particular, show promise as alternative electrodes for brain-machine interfaces due to their conformability, ease of functionalization, mechanical stability, and excellent electrical conductivity^[12][13]. Nguyen et al. (2021) conducted an in vivo biocompatibility study using a novel graphene electrode on P23H rat model retinas. The results indicated reduced inflammation, as evidenced by decreased microglial labeling, compared to biocompatible polymer-based electrodes^[14]. However, further investigations are necessary to evaluate the stimulation capacities and overall efficacy of graphene-based electrodes in retinal prosthetics.

Nanowires have also been investigated for their ability to stimulate neurons at the axonal and dendritic levels^[1]. Studies exploring two- and three-dimensional nanowire electrodes have shown improved sensitivity and signal-to-noise ratio due to increased surface-to-volume ratio and higher density of neuron-nanowire interfaces^[15]. Yang et al. (2019) discussed neuron-like electronics (NeuE) as a novel biomimetic electrode technology that closely resembles the structure and mechanics of neurons. Maintaining similarity between the neuron and electrode is crucial to avoid disruptions in signal recording and damage to surrounding tissue. Preliminary studies of NeuE in cerebral mouse models demonstrated minimal immune provocation and showed potential pro-regenerative properties by directing cells towards damaged tissue sites requiring repair^[16].

Visual Field Size

Recent advancements have focused on improving the visual field size of retinal prostheses. Felauto et al. (Switzerland) developed a foldable and photovoltaic wide-field epiretinal prosthetic called POLYRETINA, which addresses the limited visual field in current retinal prosthetic technology^[17][18]. Optimal visual perception requires a wide visual angle, which plays a crucial role in various processes, including attention, spatial recognition, and interacting with the environment^[19]. However, existing prosthetics like the Argus II offer a visual angle size (VAS) of only up to 20 degrees, insufficient for mobility tasks and object identification^[20][21]. Patients using current prosthetics must continually scan their environment to compensate for the limited VAS, which can be physically and cognitively burdensome^[21]. Preliminary studies suggest that the minimal visual field requirement for daily tasks is around 30 degrees, indicating the need for a higher VAS to adapt appropriately^[21]. The small size, limited retinal coverage, and challenges in surgical fixation make current retinal prosthetics inadequate^[21].

POLYRETINA not only improves visual angle size (VAS) but also enhances spatial resolution through its wireless photovoltaic technology, allowing for higher electrode number, density, and coverage^[20]. Similar to the retina's function, POLYRETINA projects artificial light onto the retina, where it is absorbed by a semiconductor layer embedded in the stimulating electrodes^[17][19]. Each pixel in the device converts incident light into information, resulting in 10,498 densely packaged photovoltaic pixels covering approximately 43 degrees^[17][19]. The increased electrode density reduces the need for frequent environmental scanning and improves spatial resolution^[20]. By utilizing network-mediated stimulation instead of direct activation, POLYRETINA avoids visual distortions caused by altering the retinotopic map^[21]. The wireless capabilities of POLYRETINA eliminate the need for space-consuming hardware, such as trans-scleral connectors and pulse generators, allowing for a high electrode number and density^[21]. Studies using ex vivo mouse models with retinitis pigmentosa demonstrate the device's ability to achieve high spatial resolution^[17].

POLYRETINA's conformable nature offers the advantage of easier surgical fixation^[20][21]. Unlike larger and more invasive implants, POLYRETINA can be folded and injected through a 6 mm scleral incision, making it a safer option. Wired implants pose risks of lead-wire damage, implant dysfunction, and tissue scarring due to mechanical wiring. POLYRETINA's wireless technology eliminates these issues by removing the need for unnecessary hardware.

Artificial Intelligence

Artificial intelligence (AI)-based imaging protocols show promise in addressing low sampling resolution. Recent AI research demonstrates the potential for computer vision algorithms to enhance image quality through preprocessing. For example, Ge et al. (2017) developed NeuCube, an obstacle-avoidance AI system based on a spiking neural network^[22]. NeuCube provides real-time obstacle analysis and guidance to prosthetic wearers, offering stability and accuracy without relying on sensors or additional hardware.

Deep learning models, such as DeepGaze II by Kümmerer et al. (2016), play a crucial role in improving saliency mapping^[23]. DeepGaze II utilizes VGG-19, a trained convolutional neural network (CNN), to recognize objects and enhance saliency prediction. Integration of depth estimation protocols is an area of improvement, and Godard et al. (2019) addressed this by developing monodepth2, a self-supervised model for per-pixel monocular depth approximation^[24]. It significantly reduces visual artifacts and improves motion assumptions and reprojection loss.

To validate computational models of the retina, Han et al. (2021) used deep-learning-based scene simplification algorithms on psychophysically validated retina models, accurately predicting stimulation patterns for refined algorithms^[25].

While CNNs like DeepGaze II excel in two-dimensional image processing, they are ineffective in consolidating three-dimensional videos^[26]. An emerging alternative, convolution recurrent neural network (CRNN), optimizes spatiotemporal feature extraction. However, its energy demands make it unsuitable for retinal prosthetics. Wang et al. (2022) developed an energy-efficient version called SpikeSEE, which utilizes a dynamic scene-processing framework, improving prediction accuracy and reducing power consumption^[26].

Preserving Residual Visual Field

Retinal prostheses face concerns regarding their impact on residual functional retinal cells in patients with remaining peripheral vision, limiting their eligibility. A recent human study aimed to improve central vision in atrophic age-related macular degeneration without affecting peripheral vision^[27]. The study developed a wireless photovoltaic retinal implant (PRIMA), converting projected images from video glasses into electric current to stimulate inner retinal neurons, successfully implanted under the macula of five patients^[27]. The implants demonstrated visual acuity up to 20/460 without decreasing natural acuity^[28]. Additional studies confirmed PRIMA's robustness, with at least 10 years of in vitro reliability and resistance to corrosion and water ingress^[29]. Another clinical trial examined changes in macular structures and thickness associated with subretinal implantation in geographic atrophy, with stable implant-to-target cell distance and minor retinal thickness reduction observed^[30]. Overall, the surgical delivery of photovoltaic subretinal implants showed long-term stability and no adverse structural or functional effects^[30].

2. Outlook on Retinal Prostheses

Outlook on Retinal Prostheses

Retinal prostheses have gained significant interest, but improvements are still needed to optimize patient outcomes^[31]. Factors such as electrode-retina alignment, size, material, spatial selectivity, and bidirectional systems influence the effectiveness and visual acuity of retinal prostheses^[32]. Further research is necessary to address these limitations and enhance visual processing and psychophysical performance.

The clinical application of retinal prostheses faces several challenges:

1. Limited effectiveness: Current prosthetic devices cannot fully replicate the complexity and functionality of the natural retina, resulting in limited visual perception.
2. Surgical complexity: Implanting retinal prostheses requires delicate and technically challenging procedures with inherent risks.
3. Patient eligibility: Specific visual and anatomical characteristics must be considered to ensure candidates can benefit from the device, leading to a limited pool of eligible candidates.
4. Long-term durability: The longevity of retinal prostheses poses challenges due to potential mechanical failure, degradation, tissue response, and unknown possibilities for repair or replacement.
5. Cost and accessibility: Advanced technology and complex surgeries make retinal prostheses expensive, limiting access for many patients and hindering broader application.

Addressing these challenges requires advancements in technology, surgical techniques, and understanding of retinal physiology to improve retinal prostheses.

The future of retinal prostheses relies on improved clinical trial results to support device approval and increase user adoption. Several bioelectronic implants, including Argus II, Alpha IMS, BVT Bionic Eye System, and IMIE 256, have been used in clinical trials^[33]. While Argus II is the only FDA-approved retinal implant that restores some vision in advanced retinitis pigmentosa, it has limitations due to high stimulation and large electrodes, resulting in wide spacing or low density^[34]. IMIE 256, an upgraded version with smaller size and more electrodes, addresses these limitations and has shown favorable outcomes in safety and clinical efficacy^[33]. The continuous improvement of such devices will determine their wide adoption or potential replacement by other therapies.

Furthermore, the future of retinal prostheses depends on advancements in artificial intelligence (AI) and deep learning algorithms. The combination of understanding the retina's working principle and state-of-the-art computer vision models has led to significant progress in processing algorithms for retinal prostheses^[35]. AI-based image processing methods offer improved extraction capabilities, enhancing visual perception for patients with retinal diseases^[36].