On the Information Coding with Retinal Neuromorphic Spikes Toward the Spike-Based Microstimulation in Intra-Cortical Visual Prostheses

We have recently proposed an idea to employ the timing sequences of retinal neuromorphic spikes for the microstimulation pulses in intracortical visual prostheses. For a real-time emulation of the visual information encoding into the spikes, we have previously employed a combination of spatiotemporal linear filtering and static nonlinear scaling along with the Izhikevich spiking model, enabling us to simulate the biological post-spike refractoriness that improves spike timing precision and reduces the variance-to-mean ratio of spike counts in a certain time window. However, there has been little investigation into how the post-spike refractoriness affects the decodability of the spike sequences into visual information. In this study, we used a 3-dimensional convolutional neural network model, that was pre-trained to decode the retinal neuromorphic spikes into neural images, to compare the coding properties with the spikes computed with varying post-spike refractoriness in the Izhikevich model.