Decoding network-mediated retinal response to electrical stimulation: implications for fidelity of prosthetic vision

摘要: Objective: Patients with the photovoltaic subretinal implant PRIMA demonstrated letter acuity by ~0.1 logMAR worse than sampling limit for 100μm pixels (1.3 logMAR) and performed slower healthy subjects, which exceeded at equivalently pixelated images ~0.2 logMAR. To explore underlying differences between natural prosthetic vision, we compare fidelity of retinal response to visual electrical stimulation through single-cell modeling ensemble decoding. Approach: Responses ganglion cells (RGC) optical or (1mm diameter arrays, 75μm pixels) white noise in degenerate rat retinas were recorded via MEA. Each RGC was fit linear-non-linear (LN) convolutional neural network (CNN) models. characterize level, compared statistics spike-triggered average (STA) RGCs responding retinas. At population constructed a linear decoder determine certainty can support N-way discrimination tasks. Main results: Although LN CNN models match responses pretty well (correlation ~0.6), they significantly spike timings elicited retina ~0.15). In retina, is equally bad. The signal-to-noise ratio STAs matched that when 78±6.5% spikes replaced random timing. However, contributed minimally errors determining factor accuracy decoding number cells. compensate fewer under larger presentations same stimulus are required deliver sufficient information image Significance: Slower pattern identification patients may be explained lower electrically activated compensated presentations.