Abstract
Purpose :
Electrical stimulation through retinal prosthesis elicits retinal ganglion cell (RGC) spikes directly and indirectly through synapses. Because directly-evoked spike is usually 1:1 matched with electric stimulus, it is very important to isolate spike obfuscated by electric stimulus artifact. Previously, we showed that topographic prominence (TP) discriminator based algorithm is valid and useful for artifact subtraction. In this study, we compared the performance of forward backward (FB) filter only vs. TP-adapted FB filter.
Methods :
C3H/HeJ strains (rd1 mice) at postnatal week 10 and higher were used for the retinal degeneration model (number of mouse = 3). Extracted retina was placed RGC layer down on the surface of the 8 × 8 multi-electrode array (MEA) in which one electrode was used for stimulation while all the others for recording. The recorded signals were classified into four groups by distances between the stimulation and recording electrodes on MEA (200-400, 400-600, 600-800, 800-1000 μm). Fifty cathodic phase-1st biphasic current pulses (duration 500 ms, intensity 5, 10, 20, 30, 40, 50, 60 μA) were applied at every 1 sec. Two subtraction filters were used: forward backward (FB) zero phase filter with cut-off frequency of 500 Hz high pass and TP-adapted FB filter. The performances of two algorithms were statistically compared in terms of false positive error and false negative error, considering variability of stimulus artifacts related with spatial distance and current intensity.
Results :
In terms of false positive error, TP-adapted FB filter shows better performance than simple FB filter (p < 0.001, number of channels vary from 32 to 83) except in the condition of 50 μA intensity, 200-400 μm distance. In terms of false negative error, simple FB filter shows better performance than TP-adapted FB filter (p < 0.01).
Conclusions :
Since false negative error of TP-adapted FB filter (0.08 ± 0.02 spike/pulse) was much less comparing with false positive error of FB filter (2.64 ± 0.07 spikes/pulse), it suggests that advantage of using TP-adapted FB filter would be higher than negligible disadvantage of slight increase of false negative error. Therefore, we propose that TP-adapted FB filter could be used as a computational algorithm to subtract the electric stimulus artifact.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.