gms | German Medical Science

Objective: Electronic epiretinal implants usually stimulate retinal ganglion cells non-selectively and indiscriminately, limiting their efficacy. One approach to improve selectivity is to use the implanted electrodes to record spikes evoked by a collection of electrical stimuli, and then only deliver those stimuli that reliably target specific neurons. However, obtaining these measurements can be slow and inaccurate due to complex, stochastic evoked activity and stimulus artifacts in the electrical recordings. Here we present a method to substantially improve this calibration.

Materials and Methods: 512-electrode stimulation and recording was performed using ex vivo preparations of the macaque retina. The spike responses of parasol retinal ganglion cells were identified using Kilosort and manual methods. An adaptive closed-loop stimulation and spike detection algorithm was developed to accurately and rapidly determine electrical activation thresholds while not requiring manual spike sorting and estimation of the stimulus artifact. The detection algorithm was based on two observations about electrically-evoked spikes: (1) spikes can be detected from partial waveforms to avoid partially overlapping artifacts, and electrodes located near the soma record the same spike waveform regardless of the spike activation site.

Results: Compared to prior work, the approach produced a tenfold reduction in the number of stimuli required to accurately determine ganglion cell activation thresholds. In some cases, calibration was achieved with the minimum possible number of stimuli, given the inherent variability in the membrane potential and spiking response.

Discussion: This work suggests that, in certain conditions, calibration of cellular resolution epiretinal implants can be performed in under a minute, potentially enabling rapid recalibration and improved operation in implanted devices.

Acknowledgments: Supported by Stanford Bio-X and NIH Biotechnology Training Grant, Stanford Wu Tsai Neurosciences Institute, and NIH NEI R01-EY021271.