Elucidation of simple optic stimuli based on detected nerve activity

Abstract

In this study, a minimally invasive interface to an intact peripheral nerve was developed. The interface was based on a cylindrical cuff installed around the nerve, and fitted with circumferentially dispersed surface electrodes. Differential measurements of bulk potentials were filtered, amplified, and subsequently processed with the objective of identifying unique patterns of electrical activity within the ensheathed nerve bundle. The system was implemented using an in vivo preparation of lateral eye and optic nerve from Limulus polyphemus. Recorded signals were filtered using the discrete wavelet transform and the Daubechies 4 wavelet basis to isolate signal components regarded as significant to the optical pattern. The entropies of the filtered signals were computed and used to build training and testing patterns for a cascade correlation neural network. The values on the eight outputs of the neural network represented the probability of one or more of the eight locations on the lateral eye having been illuminated by the stimulus. From these outputs, an image of the optical stimulus pattern was rendered. In tests, the system displayed an accuracy of 100% in distinguishing stimulus from nonstimulus conditions, and up to 91% in identifying the location(s) illuminated by a given pattern of optical stimulation.