A Patient-Specific Computational Framework to Optimize Retinal Stimulation

Abstract

Electronic visual prostheses stimulate neurons in the visual pathway to create light perception for blind patients. Electrically induced spots of light are called phosphenes, and visual prostheses aim to create interpretable scenes composed of phosphenes. Retinal prostheses stimulate inner retinal neurons for patients with degenerated photoreceptors. Clinical testing has demonstrated that retinal prostheses users can detect high-contrast objects and perform basic navigation tasks. However, the best restored visual acuity (20/438) is still below the level of legal blindness. Furthermore, phosphenes created by single-electrode stimuli vary in shape, size, and brightness. The future success of retinal prostheses depends on their ability to activate target neurons with improved spatiotemporal resolution. Computational models are a useful tool for investigating the mechanisms of neuromodulation therapies in a controlled environment. A common approach for modeling retinal stimulation is to: (1) create a three-dimensional bioelectric field model that solves for the spatial distribution of electric potential throughout the tissue, and (2) calculate the effects on retinal neurons using multi-compartment cable models. Prior studies have used simplified bioelectric field models that do not incorporate nonhomogeneous retinal thickness, fibrous tissue growth surrounding the implanted array, or variations in device placement. These canonical models cannot explain phosphene variability. In this thesis, we developed a novel methodology for building patient-specific field-cable models of retinal stimulation from optical coherence tomography images. We used these models to identify optimal stimulus parameters for individual electrodes. In Aim 1, we implemented a multi-compartment cable model of a retinal ganglion cell (RGC) and applied extracellular electric stimuli. We conducted sensitivity analyses to examine how dendrite representation, axon trajectory, and axon diameter influenced membrane dynamics and we optimized the spatial and temporal discretization of our model. We also implemented several simplified threshold prediction strategies based on activating function, but these did not match the prediction accuracy achieved by the cable equations. Through this work, we provide practical guidance for modeling the extracellular stimulation of RGCs to produce reliable predictions. In Aim 2, we created statistical and field-cable models to investigate inter-electrode variability across seven epiretinal prosthesis users. Not all electrodes induce visual perception at the same current amplitude, requiring the establishment of a perceptual threshold for each electrode. Our statistical analysis revealed that retinal thickness beneath the electrode correlated with measured perceptual thresholds with a fixed effect across participants. Electrode-retina distance and electrode impedance correlated with perceptual threshold as well, but these effects varied by individual. We developed a novel pipeline to build field-cable models from patient imaging data. These patient-specific models incorporate retinal anatomy and electrode position to predict inter-electrode differences. Model predictions significantly correlated with perceptual thresholds for 80% of participants. We also demonstrated that patient-specific field-cable models could make reasonable predictions of retinal activity and phosphene size. In Aim 3, we applied the Taguchi method for design of experiments to our patient-specific field-cable models to identify multi-electrode stimuli that focalize retinal activation, thus improving spatial resolution. We improved focality in all eight tested regions with a unique return electrode configuration, supporting the need for individualized current focusing strategies. This work lays the foundation for personalized, automated device programming, which will be essential for future retinal prostheses with thousands of electrodes.