Abstract
Purpose :
Recent clinical trials demonstrated the feasibility of retinal prostheses for somewhat useful artificial vision. Although the healthy retina is known for transmitting remarkable amount of neural information, the amount of artificial visual information created by prosthetic devices is still unknown. The quality of prosthetic vision is likely to be at least partially dependent on how much information would be transmitted. Here, we investigated neural information as a function of mean firing rate and spiking duration of retinal ganglion cells (RGCs).
Methods :
We used neural computational methods to generate correlated spike trains and calculate the amount of information of those simulated RGC spiking activities. First, we created five groups of 1-sec-long spike trains with different mean firing rates using a modified version of ‘Brian 2’, an open-source simulator for spiking neural networks. Second, we created five groups of spike trains that have 60 Hz average firing rate but different spiking duration. Then, we calculated the amount of neural information for each group to examine the effects of varying mean firing rate and spiking duration.
Results :
The mean firing rates of the first five groups were 20, 40, 60, 80, and 100 Hz. The amounts of neural information transmitted by 15 cells of each group were 2.95±0.11,4.32±0.15,5.19±0.13, 5.96±0.16, and 6.38±0.13 bits (mean±std), respectively. The spiking duration of the second five groups were 0.2, 0.4, 0.6, 0.8, and 1.0 sec. In each group, 15 cells transmitted neural information 1.54±0.07, 2.82±0.09, 3.76±0.09, 4.37±0.10, and 5.19±0.13 bits, respectively. In both cases, the higher firing rate or the longer spiking duration, the more neural information was transmitted as expected. However, the effects of increments in both parameters were gradually diminished. For example, when the spiking duration was 0.2 sec, the neural information increased by 0.77 bits for the mean firing rate change from 20 Hz to 60 Hz. When the duration was 1.0 sec, the neural information increased by just 2.24 bits for the same change in mean firing rate. Although the duration was 5 times bigger, the information was only ~2.9 times more.
Conclusions :
Our results indicate that we can transfer more information as the mean firing rate and spiking duration increase but energy-information efficiency should be considered.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.