Anna Kochnev Goldstein

A bstract: Retinal resistivity is a key parameter for modeling both the electric field in tissue and the resulting neural response - a crucial step in designing retinal prostheses, especially since both depend on resistivity nonlinearly. Previous studies using penetrating microelectrodes reported vastly different values, varying by as much as two orders of magnitude, likely due to parasitic capacitance in microelectrodes, compression of tissue, and its rupture by the inserted probe, allowing the conductive medium to seep in and alter the measurement. To measure retinal resistivity without compression or rupture, we applied electrical resistance tomography using a planar multielectrode array with a 30 µm pitch. Voltage waveforms generated in response to current injected from one electrode were measured on other electrodes in the array, at distances ranging from 30 to 600 µm. Data from wild-type and degenerated rat retinas were analyzed using two methods: a commercial inversion software, Res2DInv, and a custom machine learning-based decoder, with healthy retina approximated as a 4-layer structure and degenerated retina - as a 3-layer one. In healthy retina, average resistivity was 2.7 kΩ·cm for the combined ganglion cell-nerve fiber layer, 1.6 kΩ·cm for the inner plexiform layer, 2.3 kΩ·cm for the outer and inner nuclear layers, and 1.0 kΩ·cm for the photoreceptor inner and outer segments; a single-layer approximation yielded 1.7 kΩ·cm. In degenerated retina, average resistivity was 3.4 kΩ·cm for the ganglion cell-nerve fiber layer, 1.1 kΩ·cm for the inner plexiform layer, and 4 kΩ·cm for the inner nuclear layer, with a single-layer approximation of 2 kΩ·cm. S ignificance: Designing retinal prostheses to restore vision after loss of photoreceptors requires accurate modeling of electric field in tissue, but published resistivity values vary by two orders of magnitude due to inconsistencies with penetrating microelectrode-based measurements. We adapted electrical resistance tomography, a non-destructive technique from geophysics based on inverse problem solving, and developed a machine learning-based decoder to map resistivity in rat retinas using a planar microelectrode array. Our analysis yields quantitative, layer-specific resistivities for healthy and degenerated retina. Beyond retinal prostheses, this approach offers a general, deformation-free strategy for measuring electrical properties of biological tissues.

Objective: To preserve the encoding of visual information in prosthetic vision as close to natural as possible, subretinal photovoltaic implants, which replace the lost photoreceptors, strive to stimulate the second-order retinal neurons, the bipolar cells, while avoiding direct activation of the downstream retinal ganglion cells. To assess the range of such selective subretinal activation, we implanted the devices in rodent models of retinal degeneration and measured the stimulation thresholds based on the visually evoked potentials. After assessment of the bipolar cell-mediated thresholds, direct activation of retinal ganglion cells was measured following intraocular injection of synaptic blockers. Since these chemicals are toxic to the retina, this procedure can only be done once in each animal. Approach: We developed a machine-learning model that identifies the stimulation pathway directly from the recorded visually evoked potentials, eliminating the need for synaptic blockers. The model was trained on recordings from rats implanted with PRIMA subretinal arrays and evaluated on two additional implant architectures, a second rat species, and a different anesthesia protocol. Main Results: The classifier achieved a balanced accuracy of 92% in cross-validation on the training data. Generalization to all unseen experimental conditions yielded an average balanced accuracy of 91%. Integrated Gradients analysis showed that combined bipolar and ganglion cell responses were driven by the early P1 component, while bipolar cell responses relied on later waveform components, consistent with thalamocortical processing dynamics. Significance: The described computational alternative to pharmacological blockers should improve the experimental throughput, allow multiple recordings over the lifetime of the same animal, and might be applicable to optimization of the stimulation settings in patients.

Photovoltaic subretinal prosthesis can restore central vision in patients blinded by age-related macular degeneration with letter acuity matching its 100 µm pixel size. Improving resolution requires smaller pixels, but to still reach the target neurons, electric field should be less confined. However, wide-spreading field may engage adjacent photoreceptors and alter the visual perception. We studied the effects of residual photoreceptors on retinal responses to electrical stimulation using monopolar and bipolar photovoltaic arrays implanted subretinally in Long Evans rats with local photoreceptor loss, and compared that to RCS rats lacking all photoreceptors. Patterned retinal activation (880 nm, 0.5 to 10 ms) was assessed using visually evoked potentials under scotopic and photopic conditions, with and without the intravitreal injection of neurotransmitter blockers. Results were analyzed using a computational model of photoreceptor activation by various electric field configurations. We observed two mechanisms of photoreceptors engagement in electrical activation of the degenerate retina: 1) Dark-adapted photoreceptors near the implant can be simulated directly by a negative electric potential of the common return electrode along the edge of the array. 2) Light-adapted photoreceptors can reduce the stimulation threshold of bipolar cells within about 100 mm from the implant’s edge. Both effects may lead to reduced perceptual uniformity. Bipolar pixels with local return electrodes generate better confined electric fields than monopolar arrays and thus are less affected by the nearby photoreceptors. However, even such implants should be placed a few hundred micrometers from the edge of scotoma to minimize the unintended percepts.

Objective. PRIMA subretinal implants provide pros thetic vision to patients blinded by age-related macular degeneration, with acuity closely matching the sampling limit of the pixel pitch: a single 100µm pixel per line of a letter corresponds to 20/420 acuity. Decreasing the pixel size in the same flat geometry is difficult due to the constrained electric field, especially consid ering a 40µm thick debris layer separating the implant from the target neurons. Here we optimize the electrode design to help overcome such limitations. Approach. An end-to-end modeling pipeline combines the retinal photovoltaic implant simulator (RPSim) based on the Xyce circuit simulator with an interface to COMSOL Multiphysics for electric field modelling. It was used to generate and characterize implants in an open-loop sampling based optimization. Implant performance was evaluated with respect to voltage drop across bipolar cells (representing the stimulation strength), pattern contrast, and neural selectivity. Main Results. The highest selectivity in stimulation of bipolar cells was achieved with arrays having active electrodes on pillars and return electrodes connected in a mesh surrounding the photovoltaic pixels in the array. Such a design, even with pixels down to 20µm, provides stimulation strength exceeding, and contrast similar to that of flat 100µm PRIMA pixels. Significance. Using a novel 3-D electrode design, the pitch of the photovoltaic array can be decreased to 20µm, while providing performance that exceeds the flat 100µm PRIMA pixels. In humans, 20µm resolution on the retina corresponds to a visual acuity of 20/80 a five times improvement compared to the current clinical device.

Objective: Patients implanted with the PRIMA photovoltaic subretinal prosthesis in geographic atrophy report form vision with the average acuity matching the 100 μm pixel size. Although this remarkable outcome enables them to read and write, they report difficulty with perceiving faces. Despite the pixelated stimulation, patients report seeing smooth patterns rather than dots. This paper provides a novel, non-pixelated algorithm for simulating prosthetic vision the way it is experienced by PRIMA patients, compares the algorithm’s predictions to clinical perceptual outcomes, and offers computer vision and machine learning (ML) methods to improve face representation. Approach: Our simulation algorithm (ProViSim) integrates a grayscale filter, spatial resolution filter, and contrast filter. This accounts for the limited sampling density of the retinal implant (pixel pitch), as well as the reduced contrast sensitivity of prosthetic vision. Patterns of Landolt C and faces created using this simulator are compared to reports from actual PRIMA users. To recover the facial features lost in prosthetic vision due to limited resolution or contrast, we apply an ML facial landmarking model, as well as contrast-adjusting tone curves to the image prior to its projection onto the photovoltaic retinal implant. Main results: Prosthetic vision simulated using the above algorithm matches the maximum letter acuity observed in clinical studies, as well as the patients’ subjective descriptions of perceived facial features. Applying the inversed contrast filter to the image prior to its projection onto the implant and accentuating the facial features using an ML facial landmarking model helps preserve the contrast in prosthetic vision, improves emotion recognition and reduces the response time. Significance: Spatial and contrast constraints of prosthetic vision limit resolvable features and degrade natural images. ML based methods and contrast adjustments prior to image projection onto the implant mitigate some limitations and improve face representation. Even though higher spatial resolution can be expected with implants having smaller pixels, contrast enhancement still remains essential for face recognition.

Previous studies exploring human sweat ducts as biological antennas in the sub-THz range have shown that the electromagnetic (EM) response of the skin is modulated by the person’s mental and physical stress. These findings naturally raised hopes of a new remote avenue for detecting human stress. However, as those studies unmasked stress using correlations with well-established markers such as the Galvanic Skin Response (GSR), the question of whether the EM response could serve as an independent marker of stress remained unanswered. Here, we provide a positive answer to this question by showing that machine learning models trained on EM reflections from 21 participants, subjected to physical and mental stress, were able to estimate the presence of stress in a signal from a new participant, in a matter of seconds, with above 90% accuracy.

Patients with age-related macular degeneration (AMD) implanted with the PRIMA photovoltaic subretinal prosthesis demonstrated letter acuity closely matching the device's 100 µm pixel size. Improving visual acuity requires smaller pixels, which, in turn, require relaxation of electric field confinement to maintain effective stimulation of bipolar cells. Eliminating local return electrodes broadens the electric field but may inadvertently engage residual photoreceptors adjacent to the implant, thereby altering electrically evoked visual percepts. Here we quantify the contribution of residual photoreceptors to electrically evoked retinal responses across various implant configurations. Monopolar photovoltaic arrays with 20 µm pixels and bipolar arrays with 100 µm pixels were implanted subretinally in Long Evans rats, resulting in local degeneration of photoreceptors directly above the device. Responses were compared with those obtained in RCS rats, which lack functional photoreceptors. Implants were activated by patterns of 880 nm laser at pulse durations varying from 0.5 to 10 ms. Visually evoked potentials were measured in scotopic and photopic conditions, with and without the intravitreal application of mGluR6 agonist L-AP4 to block photoreceptor-driven ON pathways. Experimental thresholds were interpreted using a computational model of retinal network activation in distinct electric field geometries. In locally degenerate retina stimulated with monopolar arrays, blocking photoreceptor input yielded a rheobase (0.06 mW/mm ² ) and chronaxie (~3 ms) of the stimulation threshold, matching that measured in fully degenerate RCS retina, indicating direct activation of bipolar cells. In contrast, when photoreceptor input was intact, stimulation thresholds decreased significantly, and dark adaptation further modulated the threshold in a pulse-duration dependent manner. Bipolar arrays provided identical thresholds in locally degenerate and fully degenerate retina (0.2 mW/mm ² ) when stimulation was confined to the implant center; however, shifting stimulation to the implant's edge lowered the thresholds, revealing a contribution from adjacent photoreceptors. These findings demonstrate that residual photoreceptors can substantially influence responses evoked by subretinal prostheses in a degenerate retina, with direct consequences for perceptual uniformity in patients, such as edge brightening. These results provide guidance for design of the next-generation higher-resolution implants, supporting the use of local return electrodes to maximize resolution while minimizing the effect of photoreceptors in clinical applications.

Modeling of Multi-Electrode Arrays used in neural stimulation can be computationally challenging since it may involve incredibly dense circuits with millions of interconnected resistors, representing current pathways in an electrolyte (resistance matrix), coupled to nonlinear circuits of the stimulating pixels themselves. Here, we present a method for accelerating the modeling of such circuits with minimal error by using a sparse plus low-rank approximation of the resistance matrix. We prove that thresholding of the resistance matrix elements enables its sparsification with minimized error. This is accomplished with a sorting algorithm, implying efficient O (N log (N)) complexity. The eigenvalue-based low-rank compensation then helps achieve greater accuracy without significantly increasing the problem size. Utilizing these matrix techniques, we reduced the computation time of the simulation of multi-electrode arrays by about 10-fold, while maintaining an average error of less than 0.3% in the current injected from each electrode. We also show a case where acceleration reaches at least 133 times with additional error in the range of 4%, demonstrating the ability of this algorithm to perform under extreme conditions. Critical improvements in the efficiency of simulations of the electric field generated by multi-electrode arrays presented here enable the computational modeling of high-fidelity neural implants with thousands of pixels, previously impossible. Computational acceleration techniques described in this manuscript were developed for simulation of high-resolution photovoltaic retinal prostheses, but they are also immediately applicable to any circuits involving dense connections between nodes, and, with modifications, more generally to any systems involving non-sparse matrices.

Objective PRIMA subretinal implants provide prosthetic vision in patients blinded by age-related macular degeneration, with acuity closely matching the sampling limit of the pixel pitch: a single 100 µm pixel per line of a letter corresponds to 20/420 acuity. Decreasing the pixel size in the same flat geometry is difficult due to the constrained electric field, especially considering a 40 µm thick debris layer separating the implant from the target neurons. Here we optimize the electrode design to help overcome such limitations. Approach An end-to-end modeling pipeline combines the subretinal photovoltaic implant simulator (RPSim) based on the Xyce circuit simulator with an interface to COMSOL Multiphysics for electric field modelling. It was used to generate and characterize implants in an open-loop sampling-based optimization. Implant performance was evaluated with respect to voltage drop across bipolar cells (representing the stimulation strength), pattern contrast, and neural selectivity. Main Results The highest selectivity in stimulation of bipolar cells was achieved with arrays having active electrodes on pillars and return electrodes connected in a mesh surrounding the photovoltaic pixels in the array. Such a design with pixel size down to 20 µm provides stimulation strength exceeding, and contrast similar to that of flat 100 µm PRIMA pixels. Significance Using a novel 3-D electrode design, the pitch of the photovoltaic array can be decreased to 20 µm, while providing performance that exceeds the flat 100 µm PRIMA pixels. In humans, 20 µm resolution on the retina corresponds to a visual acuity of 20/80 – a five times improvement compared to the current clinical device.

In patients with atrophic age-related macular degeneration, subretinal photovoltaic implant (PRIMA) provided visual acuity up to 20/440, matching its 100 µm pixels size. Next-generation implants with smaller pixels should significantly improve the acuity. This study in rats evaluates removal of a subretinal implant, replacement with a newer device, and the resulting grating acuity in-vivo. Six weeks after the initial implantation with planar and 3-dimensional devices, the retina was re-detached, and the devices were successfully removed. Histology demonstrated a preserved inner nuclear layer. Re-implantation of new devices into the same location demonstrated retinal re-attachment to a new implant. New devices with 22 µm pixels increased the grating acuity from the 100 µm capability of PRIMA implants to 28 µm, reaching the limit of natural resolution in rats. Reimplanted devices exhibited the same stimulation threshold as for the first implantation of the same implants in a control group. This study demonstrates the feasibility of safely upgrading the subretinal photovoltaic implants to improve prosthetic visual acuity.