Abstract
Purpose: :
Since the response of retinal ganglion cell (RGC) is strongly dependent on the parameters of electrical stimulation such as pulse rate, intensity and duration, it is necessary to develop a proper pulse generation strategy which can faithfully deliver visual information to the brain. As a method for the development and evaluation of the pulse generation strategies, we investigated how much information on visual/electrical input can be reconstructed from the RGC responses by spike train decoding.
Methods: :
A piece of retina was harvested from New Zealand white rabbit and was attached to 64 channel microelectrode array (MEA). Various full-field light stimuli were presented to the retina on MEA through beam projector and proper optics. The intensity variations of the stimuli were reconstructed by optimal linear filter trained from multiple single unit RGC spike trains. We also performed electrical stimulation of the retina by amplitude-modulated pulses and reconstructed the amplitude variation out of the RGC responses. The amplitudes of symmetric, biphasic pulses were modulated by sinusoid and the stimulus pulse train was presented to RGCs from one channel of MEA. (pulse duration: 500 µs per phase, current intensity range: 20 µA to 75 µA). We used half of the data to train the decoding filter, i.e., to calculate the filter coefficients, and the other half was used for the reconstruction. Correlation coefficient between actual and reconstructed stimulus was calculated to quantify the accuracy of the spike train decoding.
Results: :
In the case of repetitive ON-OFF stimulus, spike train decoding with less than 5 RGCs resulted in high correlation coefficient between the actual and reconstructed stimulus (as high as 0.9). Gaussian and binary random stimuli were also reconstructed so that the correlation coefficients were within 0.5 ~ 0.6 range when their bandwidth were limited to 1-2 Hz. It was also possible to decode the amplitudes of pulse trains successfully (correlation coefficient > 0.7) from 1 - 3 RGCs. The higher pulse rate (1 pulse/second) yielded better decoding performance than the lower pulse rate (0.5 pulse/second)
Conclusions: :
The present results suggest that spike train decoding is useful for developing and evaluating pulse generation strategies of visual prosthesis. Experiments to observe the responses from RGCs of retina-degenerated mice should be performed to set up more effective methods to develop a pulse generation strategy.
Keywords: retina • ganglion cells • electrophysiology: non-clinical