gms | German Medical Science

Objective: Bi-directional retinal implants could enhance vision restoration by recording electrical activity to identify targeted cell types and to reproduce the neural code. However, the high power and data demands of spike recording require implants to compress the recorded data, potentially losing spike waveform information needed for cell type identification. Here we evaluate the recently-developed wired-OR compressive readout architecture, focusing on cell type identification.

Materials and Methods: A custom 1,024-channel stimulation and recording ASIC (36 μm pitch; 10 μm diameter; SIROF) with compressive recording capability was used to record spontaneous spiking activity of retinal ganglion cells from macaque retina with varying compression schemes. Spikes from individual cells were identified, and the signal fidelity as well as spike features useful for cell-typing were evaluated.

Results: The wired-OR readout effectively captured spike features critical for cell type identification while significantly reducing the recorded data rate by eliminating baseline signals not associated with spikes. Compressed data from each cell produced average spatiotemporal spike waveforms comparable to those obtained with full-bandwidth recordings, though minor discrepancies appeared at lower amplitudes.

Discussion: The results support wired-OR readout architecture in neural implants, enabling recovery of essential cell-type identification features with compressive recordings.

Acknowledgments: Supported by the Wu Tsai Neurosciences Institute, NIH NEI R01-EY021271 and R01-EY032900.